Full Text

Turn on search term navigation

1. Introduction

Although encouraging accomplishments have been achieved in automatic control, scholars still strive to develop control methods toward human-level intelligence. With the widespread application of cyber-physical systems (CPSs), it is challenging to satisfy the robust and adaptive requirements of control systems owing to the unmodeled dynamics and uncertain environments. The adaptive capability of control methods based on analytical models is limited by some assumptions; the general intelligent control methods, such as neural network control and fuzzy control, are, in essence, function approximation methods, and the model structure of a system needs to be specified first before applying them [1,2,3,4]. Sometimes, a reliable model or the complete state space of a system cannot be established or defined. To make systems work effectively in such situations, it is imperative to investigate alternative control methods.

Humans can learn to deal with uncertainties and complete complex tasks, an ability which is called cognition in cognitive science [5,6]. As the germination of Wiener’s Cybernetics, it was suggested in [7] that humans and machines shared no fundamental differences, and the methods for studying them were similar, which laid the foundation for combining theories of control and cognitive science. Simply put, cognitive control is the application of cognitive science theories to the design of control systems, enabling them with cognitive abilities to learn and optimize themselves like humans.

Cognitive control has been studied from various perspectives. In [8], the cognitive control system was considered a control system that knew what it was undertaking and could actively adjust its behaviors rather than passively respond to input. In [9,10], the concept of cognitive technical system (CTS) was explained, that is, CPS with cognitive control capability. In [11], the idea of cognitive automation (CA) was introduced, and a cognitive system architecture (COSA) for CA was established based on the state, operator, and result (SOAR) cognitive model. In [12], cognitive control based on the cognitive dynamic system (CDS) was proposed, which corresponds to Fuster’s cognition paradigm [13], embodying four aspects of human cognition: the perception-action cycle (PAC), memory, attention, and intelligence.

Actually, since the introduction and implementation of Industry 4.0, revolutionary changes have taken place in social life and production. Based on CPS and characterized as being networking-data driven, Industry 4.0 promotes the deep integration of the digital world and the physical world, making industrial systems more intelligent [14,15,16]. In recent years, Industry 5.0 has received people’s attention as an emerging concept, which is a further development of Industry 4.0, highlighting the central position of humans in industrial processes as it is the integration of humans, information systems, and physical systems [17,18,19,20]. Industry 5.0 research involves many aspects, one of which is the deep cooperation between humans and machines. Industrial systems are autonomous to a certain extent, and humans can provide professional knowledge that systems cannot obtain and can respond to problems that systems do not know or are not aware of. Humans and machines cooperate to achieve key performance indicators [21,22,23]. The changes in the organization and operation of traditional industrial systems caused by Industry 4.0 and Industry 5.0 have made demands of some key capabilities for systems, such as self-perception, self-learning, self-diagnosis and self-optimization. Cognitive control organizes control systems according to the human cognition paradigm. Compared with other intelligent control methods, the intelligence of a cognitive control system is not limited to its controller; instead, the intelligence is embodied throughout the whole system. The working process of the cognitive control system is based on the perception–action cycle where each component of the cognitive control system embodies certain functions of cognition. Because the cognitive control system generally conforms to the human cognition paradigm, it has a certain degree of cognitive ability similar to humans. Cognitive control relies less on prior knowledge and human guidance and exhibits the stronger autonomous ability to collect information independently during its working process and carry out self-learning, self-diagnosis, and self-optimization. Hence, cognitive control systems have stronger abilities for self-adjustment and human-machine interaction, which is of theoretical and practical significance for the full realization of Industry 4.0 and the upcoming Industry 5.0.

However, a unified definition and method for cognitive control is still lacking for the following reasons. Firstly, cognitive control is an emerging research subject involving many disciplines, such as cognitive science, automation and control, communication, and computers. These subjects are progressing rapidly, as is cognitive control. Secondly, the problems encountered in different application scenarios are distinct, resulting in differences in cognitive control methods. Finally, the stability of cognitive control systems is difficult to prove currently and can only be verified through experiments and simulations. Nonetheless, cognitive control is meaningful in the pursuit of a control method that is as intelligent as humans.

Figure 1 presents the symbolized results of cognitive control. As can be seen from the figure, cognitive control has been investigated primarily in manufacturing, robotics, and CDS. The former two are application-oriented, while cognitive control based on CDS initially focuses on theories and methods, and the outcomes are then applied to engineering systems.

Cognitive control in manufacturing has been developed rapidly under the advocacy of some countries. For example, the cognitive technical system project funded by the German government and the cognitive system project sponsored by the British government both promote the development of cognitive control in manufacturing [9,10]. Cognitive control has also been widely studied in robotics, and many results have been obtained. Cognitive control based on CDS is primarily studied by Professor Haykin and his team, and the research content involves the theory, method, and application of cognitive control.

The development of cognitive control based on CDS was reviewed by [24,25,26,27] at different stages; cognitive control in manufacturing was reviewed by [28]; and the research of cognitive robots was reviewed by [29]. These papers are a review of one of the above three research fields. However, there is commonness among cognitive control research of these three fields. Therefore, the research results in these three fields should be reviewed and analyzed comprehensively; based on the review of these advancements, the general concept of cognitive control is summarized, the fundamental framework of cognitive control systems is presented and the differences and connections between cognitive control and similar control methods, such as intelligent control, robust control and adaptive control, are analyzed. Hence, based on the symbolized results of cognitive control shown in Figure 1, the representative papers that impact on the development of cognitive control are reviewed while ignoring those with less effect. These reviewed papers mainly involve the content of the following three aspects: (1) the discussion of cognitive control ideas and methods; (2) the establishment of architectures of cognitive control systems; and (3) the research results that come from the researchers or research groups who are engaged in cognitive control for a long period. These representative papers could embody the general situation of cognitive control.

In fact, there are other engineering applications concerned with cognition, such as cognitive radio [30,31], which aims to improve the utilization of the radio spectrum, and cognitive radar [32,33,34,35], which aims to enhance radar tracking accuracy. Their working processes can also be referred to as cognitive control in their respective areas. However, they do not belong to the research scope of automation. In this paper, cognitive control is discussed only from an automation and control perspective. Hence, although similar in thought, the cognitive control discussed in this paper differs from cognitive radio and cognitive radar.

The major contributions of this paper are as follows:

Firstly, a review of representative cognitive control papers in the three areas mentioned above is presented, and the commonness of the ideas and methods presented in these papers is analyzed and summarized.

Secondly, the relationship between automatic control and cognitive science is discussed, along with the definition and framework of cognitive control.

Thirdly, the advantages and disadvantages of cognitive control, compared with other control methods, such as intelligent control, robust control, and adaptive control, are analyzed.

Finally, the main issues, research directions, and development prospects of cognitive control are outlined.

The remainder of this paper is organized as follows. Section 2 presents a comprehensive overview of the advancements and related works on cognitive control. Section 3 introduces the concept and models of cognition and concludes with several meaningful remarks on cognitive control. Section 4 provides the definition of cognitive control. Section 5 elaborates on the fundamental framework of cognitive control systems. Section 6 compares cognitive control with similar concepts and control methods. Section 7 discusses the main issues, research directions, and development prospects. Finally, conclusions are presented in Section 8.

2. Cognitive Control Research Overview

Cognitive control can be applied to various manufacturing scenarios, such as production planning, machine monitoring, and fault diagnosis. In robotics, cognitive control is primarily related to robot behavioral control. The intention of cognitive control based on CDS is to reduce the risk faced by the system by bridging the information gap between relevant information and sufficient information.

In general, cognitive control has received less attention than other control methods, such as robust control, fault-tolerant control, event-triggered control, and intelligent control. We have pointed out in the introduction that cognitive control research is mainly conducted in manufacturing, robotics, and CDS because there are few papers regarding cognitive control in other fields, and their introduction to cognitive control concepts and methods needs to be more concrete. Therefore, we focused on representative papers in these three areas that could sufficiently reflect cognitive control development.

To reveal the commonality of cognitive control in different fields, we focused on papers that discussed the holistic concept and method of cognitive control rather than a particular cognitive function; the papers investigated are counted by year and application field, as shown in Figure 2. The figure shows that cognitive control research in manufacturing and robotics began relatively early, while cognitive control based on CDS has gained increased attention in recent years.

The uppermost feature of cognitive control methods is that most are developed based on cognitive models. A cognitive model is a reflection of the human cognition process and mechanism and includes two fundamental aspects: structure and algorithm. The structure is an organized framework composed of modules such as perception, reasoning, planning, and memory, and every module represents a specific cognition function. These modules are connected to fulfill the entire cognitive process. The algorithm illustrates the working principle of each module [36,37,38,39,40]. Various cognitive models have been formed due to the different perspectives of cognitive scientists, some focusing on calculation and some on reasoning. Although these models differ, their main modules are the same. The latest achievements in cognitive science confirm that these cognitive models are becoming unified [41,42]. The cognitive model is a comprehensive expression of cognitive theories, and research on cognitive control in engineering begins with it. It is expected that the functions of cognitive models can also be implemented in control systems to enable them to perform cognition as humans. Since most cognitive model algorithms cannot be directly applied to control systems, and the algorithm design is rarely discussed in cognitive control papers, we primarily focus on the concepts and methods of cognitive control. The cognition and cognitive models are introduced briefly in Section 3; therefore, the cognitive models mentioned in this section are not introduced in detail.

2.1. Cognitive Control in Manufacturing

Manufacturing systems are dynamic social and technical systems in which all levels, such as social production relationships, factories, production lines, machines, and production processes, change continually. By using cognitive control, control systems can overcome the limitations of preset conditions, avoid building an exhaustive knowledge base, and better adapt to the ever-changing production environment [43].

In [9,10], the architecture of CTS was established by functionally decomposing cognition into modules. The perception module collects information and estimates the system state and the learning and reasoning module is related to the self-adjustment and self-optimization of cognitive control systems; it learns and reasons (predicts) based on the cognitive control action and the sensory measurements and makes adjustments for systems themselves with the goal of achieving better control performance in the subsequent perception–action cycles. For example, it could identify parameters and adjust them adaptively in an adaptive control scenario. In artificial intelligence, there is also a large variety of information processing methods for learning, such as decision tree learners, support vector machines, and artificial neural networks, and for reasoning, such as Bayesian inferring. The knowledge and model module stores the models and algorithms used in the system and the planning and cognitive control module generates the cognitive control action, which is then executed by the execution module. CTS can also interact with humans.

However, in Figure 3 different cognitive functions, such as reasoning and learning, were integrated into the same modules. These considerations may be valid at a very high level but unreasonable in realistic application scenarios since these cognitive functions are well distinct, activate different processes, and are implemented with different techniques. Although interacting with each other, they are different in essence.

In [44,45], COSA was applied to human–machine cooperation and aircraft power management, respectively. The SOAR planning process is based on rules, which are also called knowledge, and stored in the knowledge base according to cognitive science theories. The planning part of SOAR will check all the rules to find the best one to generate the cognitive control action for the current PAC. Hence, the more complex the application scenario and the more time the system operates, the more rules will be generated and stored in the knowledge base, which leads to more matching items to be evaluated during planning, thereby decreasing the real-time capability of control systems. Hence, in the cognitive control unit designed in [46], a group of predetermined operations is directly performed without planning and coordination when the perceived information needs to be responded to immediately, as shown in Figure 4. In this way, the system’s response speed to safety-critical situations increases, and the security of human–machine cooperation is enhanced. The coordination module ensures that the planning results are properly executed at the hardware level.

In [47], a cognitive agent was constructed based on the beliefs-desires-intentions (BDI) cognitive model, as shown in Figure 5. Every machine on the shop floor is controlled by a cognitive agent, which can reduce the machine downtime or avoid production termination caused by internal disturbances (such as tool wear or machine breakdown) by making decisions independently or negotiating with other cognitive agents. Beliefs are information about the shop floor, such as the status of machines and production processes; desires are all possible states of tasks the production system would like to accomplish; and intentions are the states of tasks the system has decided to achieve.

In [48], an artificial cognitive control system was established based on the shared circuit model (SCM), which was verified to maintain cutting force stability during the drilling process by cognitively controlling the feed speed. SCM is a cognitive model that shows how subpersonal resources for control, mirroring, and simulation can enable the distinctive human sociocognitive skills of imitation, deliberation, and mind reading. In [49,50], a learning algorithm based on Q-learning was designed for the architecture established in [48].

During the development of cognitive control, some scholars suggest that the cognitive control system should not wholly copy a cognitive model as its portability will be reduced and the cost of secondary design will increase. When designing a cognitive control system, more attention should be paid to the realization of cognitive functions rather than formalizations [10,51]. Similar viewpoints can be found in [52,53]. In [52], a cognitive safe controller that contained a strategic controller and an operational controller was designed, as shown in Figure 6. The mission of the strategic controller is to optimize the standard controller and generate control actions to keep the plant away from safety-critical states; the operational controller becomes active when a warning level (before reaching the critical state) is encountered, and the operational controller ensures that a critical state is never entered. The author has explained the working principles of the strategic controller and operational controller in detail. However, the method for realizing safety sensors and safety actuators cannot be found in the paper. In [53], a cognitive control system with memory, reasoning, and planning functions was realized based on the combination of the graph model, radio frequency identification technology, and the Dijkstra algorithm. The experimental results showed that the machine downtime during a 3 h test run fell by 41.4%, and the waiting time of the production cell decreased by 17.7%.

The integration of cognitive control systems with manufacturing systems has attracted the attention of scholars. Since most modern manufacturing systems are CPSs, various open architectures have been formed for CPS to achieve unified access to systems software and hardware resources. When integrated with a CPS, the cognitive control system should follow the specifications of at least one of these open architectures. In [54], a cognitive control system was realized based on the production logistics and sustainability cockpit open architecture and the cognitive control architecture proposed in [48]. In [55], 5C architecture was employed to illustrate the relationship between cognitive control and CPS. The 5C structure is an open structure with five layers for the implementation of a CPS, as shown in Figure 7. The connection layer comprises sensors that collect data, the conversion layer extracts relevant information from the data, the cyber layer estimates the system state according to the relevant information, and the cognition layer presents the system’s state to users and helps them make control decisions, which are finally executed at the configuration layer [56]. In [57], a cognitive control system that conformed to the ISA-95 standard [58] was built using the cognitive model of simulation of mental apparatus (SiMA), which cooperated with the manufacturing execution system to automatically improve the health status of the production system.

In [59], a hierarchical cognitive control methodology was proposed. A hierarchical cognitive control system contains several cognitive controllers that solve problems at different levels. These controllers cascade together to accomplish production tasks. In [60], the cognitive control system was divided into four layers: two cognitive entity layers, a cognitive data-driven layer, and a cognitive communication and security layer.

Cognitive digital twin (CDT) has recently become a promising research subject associated with cognitive control in which the cognitive control process can be optimized by the virtual reality capability of digital twin technology. The concept of CDT was first proposed at an industry symposium in 2016; subsequently, in 2017, during the cognitive computing and artificial intelligence workshop at IBM, its functions were further envisioned [61]. In [62], CDT was engaged to manage the risk and safety of symbiotic autonomous systems. In [63], a CDT system architecture containing access, analytics, and cognition layers was designed. In [64], an innovative classification of manufacturing systems adaptability was introduced, differentiating static, dynamic, cognitive, and extreme adaptability, and cognitive adaptability was realized using CDT. In [65], a multidimensional cognitive framework for cognitive manufacturing was established based on the object-attribute-relation (OAR) cognitive model, which consists of physical and cognitive space. Physical space refers to production system entities. Cognitive space is a virtual production system constructed based on digital twin technology and OAR, including virtual physical objects, their interrelationships and attribute states, and various algorithms for recognition, learning, and decision-making.

In summary, most cognitive control methods in manufacturing have been developed based on cognitive models and applied to various manufacturing scenarios. The architectures of cognitive control systems have been elaborated in the references mentioned above; however, algorithms were rarely discussed in detail. The applications of these algorithms were described in detail in [49,50,52]. In [49,50], the Q-learning algorithm was applied to the learning part of the cognitive control system. In [52], the working principles of the strategic and operational controller were explained clearly. Other references also mention algorithms. For example, the ant colony algorithm was employed to facilitate the negotiation of cognitive agents in [47], and the fuzzy neural network was used in [48]; however, their applications were not explicitly introduced. Table 1 summarizes the references in manufacturing according to their research content.

2.2. Cognitive Control in Robotics

The purpose of applying cognitive control to robots is to enhance their behavioral control ability. Robots equipped with cognitive control are also called cognitive robots [29,66,67]. Research on cognitive robots is also based on cognitive models. For example, in [68], BDI, SOAR, and adaptive control of thought-rational (ACT-R) cognitive models were referred to when designing cognitive robots. In [69], a cognitive robot architecture was created based on the hybrid cognitive model; in [70], a cognitive robot architecture named multipurpose enhanced cognitive architecture (MECA) was proposed, which exhibited the advantages of cognitive models of SOAR, connectionist learning with adaptive rule induction online (CLARION), and learning intelligent distribution agent (LIDA) [71].

A type of three-layer cognitive robot architecture was continuously investigated by scholars, which was initially designed based on the human-operator model [72]. The human-operator model is a cognitive model for representing and selecting human behavior that divides behavior selection into three levels, as shown in Figure 8 [73]. The rules are the mapping from inputs to behaviors, which are equivalent to the control laws of the control system and are predetermined by designers. When no suitable rules are found, the stored knowledge is inferred by the system to establish new rules. The basic principle of establishing new rules is to make the system state as optimal or suboptimal as possible at the next moment. The three-layer architecture was successively improved in [74,75,76,77], and a software structure called ArmarX (original version) was designed in [78], which facilitated the development of cognitive robots with a three-layer architecture. At present, three-layer cognitive robot architecture has become an open research topic, and scholars are still exploring it.

In addition to three-layer architecture, many achievements have been made in cognitive robotics. In particular, researchers from Vanderbilt University’s Intelligent Robot Laboratory, University of Bremen’s Institute of Artificial Intelligence, and University of Naples Federico II have continuously studied this subject.

An intelligent machine architecture (IMA) was created by researchers from Vanderbilt University’s Intelligent Robot Laboratory. Cognitive robot components were classified into five types of atomic agents by IMA according to their functions [79]. In [80], a humanoid cognitive robot called intelligent soft-arm control (ISAC) was developed based on IMA, as shown in Figure 9. Originally, ISAC was designed to assist the physically disabled; then, it became a general-purpose cognitive robot intended to work with humans as an assistant in a variety of settings.

ISAC’s cognitive processes are performed by a human agent and a self-agent. The human agent collects information about a person’s characteristics and states, identifies a person’s identity, and infers his/her intentions. The self-agent observes the robot itself, including hardware monitoring and fault evaluation, and controls its behaviors [80].

In [81], the behavior control mechanism of ISAC was introduced, which was first proposed in [82]. When the central executive agent (CEA) of the self-agent receives the task, it first selects a group of potential behaviors from the long-term memory (LTM) according to the degree of correlation between the behavior and task and sends them to the working memory (WM). Subsequently, the behavior with the highest reward value will be implemented, as shown in Figure 10 [81]. The degree of correlation between the behavior and task is indicated by the corresponding weight value, and the reward value of a behavior is obtained by CEA through learning. Cognitive science research shows that the basic principle of reinforcement learning is in accordance with the learning mechanism of the human brain; therefore, the reinforcement learning algorithm was adopted as the learning algorithm of CEA [81].

Researchers have expanded and improved the ISAC functions. A multistrategy approach to reflection was proposed in [83], which improved the self-diagnosis and self-correction ability of ISAC. In [84], an automated monitoring system for ISAC was built, which was expected to provide the real-time status of its sensors and actuators and an analysis of the possible task failures given a set of system faults. In [85], a task-based dynamic feature mapping method was adopted to alleviate the dependence on preprogrammed heuristic functions, as well as on static feature representations.

Figure 9

IMA-based ISAC cognitive robot architecture [84].

[Figure omitted. See PDF]

Figure 10

Behavior control process in ISAC [81].

[Figure omitted. See PDF]

In order to evaluate the interaction ability of the ISAC cognitive robot, a model of the levels of interaction engagement is developed in [84], as shown in Figure 11. These levels progress the humanoid from a state of no interaction to the ultimate goal of completing a task with a person. The robot can be proactive and exhibit behaviors such as acknowledging the person’s presence, maintaining eye contact, or gesturing with the head or arm, such as nodding, waving, and pointing.

The supervisory attentional system (SAS) for cognitive robots was designed by researchers at the University of Naples Federico II. It also adopts the behavior control mechanism of [82]. The SAS architecture is shown in Figure 12 [86], which performs two critical functions: contention scheduling and supervisory attention, which are crucial for natural human–robot communication and collaborative task execution. Supervisory attention is performed by the attentional executive system, which dynamically evaluates the contributions of each behavior, and the evaluation results are stored in the WM associated with an artificial neural network. Contention scheduling is conducted by the behavior-based system. When several behaviors compete for the actuators of a cognitive robot simultaneously, the activation value of each behavior, that is, the weighted sum of the contributions, is calculated by the behavior-based system; the behavior with the maximum activation value is sent to the actuators for implementation. Research on SAS can also be found in [87,88,89,90,91,92], in which SAS was explored in different situations.

Contentions of conflicting tasks are solved by exploiting activation values, as shown in Figure 13. The activation values are online regulated by top-down and bottom-up stimuli. The bottom-up stimuli emphasize the hierarchy that is more accessible or attractive for the robot. For example, nodes associated with the objects glass can be stimulated by the glass proximity. On the other hand, top-down regulations affect nodes at all levels of abstraction and are propagated through the WM hierarchy (e.g., take(coffee) arouses goto(coffee) and pickup(coffee)). Each edge ( $i$ , $j$ ) of the WM structure is associated with a weight $w_{i, j}$ that indicates the attentional influence from the node $i$ to the node $j$ , and the emphasis value $e_{j}$ is the sum of contributions $c_{i}$ plus its corresponding weight $w_{i, j}$ ; the behaviors with the highest $e_{j}$ are considered the most activated behaviors which will then be sent to the actuators for execution [86].

A cognitive robot architecture called cognitive architecture for robot manipulation (CRAM) was created at the University of Bremen’s Institute of Artificial Intelligence [93]. CRAM is a complex architecture, as shown in Figure 14, and it comprises five main components: the CRAM Plan Language (CPL) executive, the suite of knowledge bases and associated reasoning mechanisms referred to as KnowRob 2.0, the perception executive referred to as RoboSherlock, the action executive referred to as Giskard, and the metacognition system referred to as COGITO. The concrete functions and mechanisms of these components are complicated and they are introduced in detail in [94]. The CRAM architecture has been tested in various life and production scenes, such as kitchens and production workshops. Studies on CRAM have also been published in [95,96,97,98,99,100,101]. In [95], a remote knowledge representation and processing service was established to facilitate the use of artificial intelligence technology to equip CRAM with knowledge and reasoning capabilities. In [96], an open-source software framework was developed to implement unstructured data management. Data are considered unstructured since they lack explicit semantics and structures. In [97], an object and scene description language were introduced that serves as a layer of abstraction between the structure of the perception logs and their semantic interpretation. In [98], a knowledge-based virtual scene rendering approach was proposed to specialize the perceptual capabilities of CRAM toward new objects. In [99], the virtual reality ability of the Unreal Engine 4 game engine was employed to generate various realistic simulated working scenes to train the cognitive robot’s perception ability without the need for a large amount of labeled data. In [100], a visual question-answering system for CRAM based on scene graphs and deep learning was created, which improved the reliability of robot manipulation in noisy dynamic environments. In [101], an experiment was conducted to investigate and evaluate the scalability and robustness of mobile manipulation.

In general, cognitive robot research relies on cognitive models with an emphasis on robot behavior control. Cognitive robots are studied initially to provide an experimental platform for testing the validity of various cognitive science theories but the development of cognitive robots has led to a more significant role for them in our daily lives and at work. Consequently, this has become an increasingly important research field. Researchers have proposed various types of cognitive robot architecture. Cognitive robot concepts and architectures are introduced in the papers mentioned in this subsection, and some also discuss algorithms. For example, the sensory egosphere algorithm was utilized to build ISAC’s short-term memory (STM) [80,81,83,84,85], and an artificial neural network was employed to create SAS’s LTM [86,87,88,89,90,91,92]. However, the application of these algorithms has not been adequately explained. It is necessary to note that there are many papers on cognitive robots; only a few representatives are introduced and they are summarized in Table 2 in terms of their research content.

Table 2

Cognitive control in robotics.

Research Content	References
1.. Three-layer cognitive robot architecture	[72,74,75,76,77,78]
2.. ISAC cognitive robot	[80,81,83,84,85]
3.. SAS for cognitive robots	[86,87,88,89,90,91,92]
4.. CRAM cognitive robot architecture	[93,94,95,96,97,98,99,100,101]
5.. Other cognitive robot architectures	[68,69,70,71]

2.3. Cognitive Control Based on CDS

CDS is also an engineering system that is inspired by human cognition. Research on CDS originates from cognitive radar [102] and cognitive radio [103] and is then extended to the automatic control field [12]. Research on CDS was carried out by Haykin; its concept was first proposed in [104]. In [12], the basic principles of CDS were explained in detail, and cognitive radar and cognitive radio were reconsidered from the perspective of CDS; that is, cognitive radar is an application that mimics the human brain, whereas cognitive radio is an application motivated by human cognition. More importantly, cognitive control based on CDS was proposed as the first special function of CDS, which is a control method that incorporates the state–space model and entropic state model.

Cognitive control based on CDS is established on the information gap. Haykin believes that since information is critical for control systems, a satisfactory level of control performance could be achieved, and the risk involved is reduced by properly defining the task-specific information gap and decreasing it during the control process. Specifically, available information is the information extracted from noisy measurements; regarding the task at hand, available information can be partitioned into relevant information and redundant information; sufficient information is defined as the required information for performing the task at hand with minimal risk; finally, the information gap is defined as the difference between sufficient information and relevant information [105]. The definition of risk presented in [106] is referenced by Haykin, i.e., the sufficiency of a statistic for a particular task is specified by the risk associated with a control or decision policy. In statistical decision theory, risk is usually defined as the expected loss.

In this subsection, cognitive control is described within the context of CDS. To simplify the description, cognitive control based on CDS is referred to as cognitive control for short, under the condition that there is no ambiguity.

The architecture of the cognitive control system is shown in Figure 15 [105]. The perceptual part and the executive part are connected to form a closed loop called PAC. In each PAC, the sensory measurements from the physical system and environment are processed by an estimation algorithm (such as a Bayesian filter) in the perceptual part to extract the relevant information. Then, the entropic state is calculated by the entropic information processor and sent to the executive part through the feedback channel. The entropic state is the quantification of the information gap, and its calculation method is determined by the specific tasks. A group of potential cognitive actions is selected from the actions space by the planner according to the current entropic state, which is evaluated by the cognitive controller through the shunt cycle, and the predicted entropic state of each cognitive action is obtained. The cognitive action to be executed is selected by policy from this group of cognitive actions according to the maximum Bayesian posterior and applied to the physical system or environment to reduce the entropic state of the system in the next PAC.

In [107], a cognitive control algorithm called value-to-go was proposed and demonstrated to be a special case of Bellman’s dynamic programming, which improved the Q-learning algorithm by solving the dimension disaster problem. Furthermore, the idea of improving the learning efficiency of the executive part through planning was discussed.

In [108], cyberattacks were defined as abnormal uncertainties; their occurrence is un-controllable, but the corresponding influence on systems, called risk, is controllable. To bring risk under control, cognitive control is developed based on the predictive adaption mechanism of the human brain [109], and the second special function of CDS, cognitive risk control (CRC), is established. The architecture of the CRC system is illustrated in Figure 16.

A risk-sensitive subsystem is integrated into the CRC system. The physical system and environment are controlled by the executive part in the same manner as the cognitive control when there is no attack. Once the system is attacked, the entropic state falls below the threshold, and the task-switch module switches the opening and closing of two sets of switches to bring the risk-sensitive subsystem into the control process. A perturbed cognitive action is first generated by the executive part; then, it is optimized by the classifier using empirical data in executive memory to generate a risk-sensitive cognitive action.

Cognitive control and CRC have been applied in communication-based train control (CBTC) systems, smart grids, vehicular radar, and communication systems.

2.3.1. Cognitive Control for CBTC Systems

CBTC systems transmit information through train–ground wireless communication, and the wireless local area network (WLAN) is generally adopted as the transmission medium. However, the time delay and packet loss caused by transmission errors and media access control (MAC) layer handoffs are inevitable because most of the current IEEE 802.11-based WLAN standards are not designed for high-mobility environments. Consequently, the state information of the front train cannot be received by the current train in time, resulting in unnecessary traction, braking, and energy loss.

In [110], cognitive control was used to alleviate the influence of uncertain communication on CBTC systems. The wireless channel was modeled as a finite state Markov model, the entropic state was used to quantify the time delay and packet loss, and the multiplexing gain and handoff action in multiple-input multiple-output (MIMO) wireless communication was adjusted by the cognitive controller. In [111], the interference in a CBTC system coexisting with a passenger information system (PIS) was mitigated based on cognitive control. Owing to the interference of the PIS, the signal-to-noise ratio (SNR) in the wireless transmission process becomes the signal-to-interference-plus-noise ratio (SINR). Hence, in addition to adjusting the multiplexing gain and handoff action, the packet size and packet interval were also controlled by the cognitive controller.

In [112], a cognitive control approach for CBTC systems was proposed based on a semi-Markov decision process. Two communication periods were set: regular and short. In each PAC, the MAC layer retry limit and communication period for the next PAC were adjusted by the cognitive controller. Specifically, when packet loss occurs, the retry limit was increased appropriately to reduce the packet loss rate, and the short communication period was chosen for the next PAC to shorten the time interval of the adjacent sampling times.

In [113], a cognitive control method for cost-efficient CBTC systems with smart grids was established. The train uses regenerative braking technology to feed energy back to the power grid; therefore, the cognitive controller needs to control the motion of the train and reduce the total financial cost of consuming electrical energy. A joint objective function composed of the information gap and the total financial cost was applied to generate an optimal policy [113]; the MAC layer retry limit and traction strategy were adjusted by the cognitive controller to decrease the packet loss rate and optimize the traction and braking process.

2.3.2. Cognitive Control and CRC for Smart Grids

Smart grids combine advanced technology with traditional power grids, bringing together power suppliers, consumers, distributed energy resources, and energy storage units to constitute a microgrid. Microgrids facilitate the interaction between power suppliers and electricity users and motivate users to participate in energy management. Demand response management (DRM) is critical for balancing the electricity supply and demand and maintaining microgrid stability. However, the DRM performance depends on the wireless communication quality of the microgrid. In [114], a cognitive control model was formulated on a distributed frame embedded in both the user agents and the microgrid control center agent. Cognitive control was carried out based on the real-time electricity price. The Q-learning algorithm was selected as the cognitive control algorithm to adjust the wireless communication channel, and the estimation-deviation level was adaptively adjusted. The experimental results showed that the global benefit for all subscribers was maximized; moreover, the electricity demand and supply were balanced.

In [115], optimal state estimation of smart grids was performed based on cognitive control. The entropy state was used to identify and detect bad data from sensory measurements. To improve the detection accuracy, an entropic state calculation method combining least-squares estimation, a Kalman filter, and a cumulative sum was proposed. When a sensor’s working environment is noisy, its corresponding weight is reduced by the cognitive controller to mitigate its influence on the state estimation of the system. The simulation results showed that the weight readjustment could be completed within a few PACs.

Furthermore, the authors showed that when faced with cyberattacks, known as false data injection (FDI) attacks, it was also possible to use the entropic state as a metric to detect FDI attacks. The disadvantage of this method is that it can only detect FDI attacks but cannot reduce or even eliminate its impact. Additionally, this algorithm is not suitable for large-scale smart grids, which have too many sensors (thousands), because this will increase the dimension of the matrix of the system model, leading to an excessive amount of computing resources being consumed during the calculation process, which will adversely affect the system performance in real-time.

In [116], CRC was employed to address the risk associated with FDI attacks. When an FDI attack occurs, the risk-sensitive subsystem is activated by the task-switch control module, which adjusts the system matrix to defend against FDI attacks.

2.3.3. CRC for Vehicular Radar and Communication Systems

Connected and autonomous vehicles (CAVs) are an integral part of intelligent transportation systems, and they have profoundly changed people’s transportation patterns and lives. With the development of mobile communication, artificial intelligence, and high-performance computing technologies, the automation level of CAV is increasing.

Vehicular radar is indispensable for CAV and its advanced driver assistance system [117]. It can sense road conditions and track targets and plays a significant role in adaptive cruise control, lane-changing assistance, and automatic parking assistance [118]. In addition to vehicular radar, vehicle-to-vehicle (V2V) communication is also an essential function of CAV, which makes tasks such as positioning and navigating more collaborative [119]. However, the driving environment of CAVs is imperfect; radar sensing and V2V communication can be declined by process and measurement noise, complex road conditions, and cyberattacks.

In [120], CRC was used for the adaptive selection of waveform parameters of the vehicular radar. When a CAV was only affected by process and measurement noise, the vehicular radar waveform parameters were adjusted by the executive part of the CRC. Once other uncertainties occur, such as structural uncertainty, the disturbed cognitive action is modified by the classifier following the shortest distance principle. In [121], the state information of other vehicles was used to formulate an extended state–space equation, based on which the accuracy of the entropic state calculation and optimization of vehicular radar waveform parameters were further improved.

In [122], CRC was applied to anti-jamming V2V communications with hybrid attackers. The transmission power and channel were reselected dynamically in each PAC to enhance the transmission success rate of the V2V communication under jamming attacks. In [123], the attacker’s eavesdropping on the V2V communication and the attack mode prediction made by the transmitting vehicle were further considered, and both results were incorporated into a utility function in the form of probability matrices to enhance the CRC system efficiency.

In [124], a coordinated CRC approach was proposed. Both the vehicular radar and the V2V communication system were considered separate CRC models with their own risk-sensitive subsystems. By establishing a mutual assistance relationship, the coordinated CRC provided a new safety mechanism that allowed one system to learn from and react to the risks that the other system has encountered.

In summary, CDS is a structured engineering system initially inspired by cognitive radar and cognitive radio, and its architecture is constructed in accordance with Fuster’s cognition paradigm, embodying four aspects of cognition: PAC, memory, attention, and intelligence. Cognitive control and CRC are extensions of CDS and have been applied to CBTC systems, smart grids, and vehicular radar and communication systems. These applications are based on the architectures depicted in Figure 16 and Figure 17. The differences among these applications are the algorithms employed in the perceptual part, the executive part, and the risk-sensitive subsystem, and the cognitive actions are also different according to the corresponding scenarios. In general, the Bayesian filter algorithm is primarily used as an estimation algorithm, whereas various reinforcement learning algorithms are widely used in cognitive controllers. In fact, cognitive control and CRC have also been applied to other fields, such as the cognitive Internet of Things; however, the related research contents are not attributed to the automatic control domain, which is beyond the scope of this paper. Table 3 summarizes the references in this subsection according to their research content.

3. Cognition, Cognitive Models, and Cognitive Control

From the introduction in Section 2, it can be concluded that although cognitive control has been investigated in different areas, and various architectures have been proposed, most architectures are created based on cognitive models.

The cognitive model reflects the architecture and process of human cognition; it is the concentrated embodiment of cognitive science theories. As stated in Section 1, cognitive control is the combination of automatic control and cognitive science. It can be seen from the cognitive control research results that the cognitive model is an effective combination point of automatic control and cognitive science. Control systems can be organized in the form of cognitive models and thus equipped with cognitive functions; as a result, they are capable of self-learning and self-optimization like humans, achieving better performance when facing uncertainties.

Hence, it is necessary to first introduce the basic concepts of cognition and cognitive models. Then, the concept and method of cognitive control are discussed without losing generality. The main viewpoints and conclusions are presented in the remarks.

The idea of combining automatic control with cognitive science can be traced back to Wiener’s Cybernetics, in which the information was used as a link to unify the control processes of humans and machines. Wiener emphasized that both human and machine control processes are feedback-based processes of information processing; this viewpoint opens the door to intelligent control [125]. Qian also pointed out in Engineering Cybernetics that various control algorithms are, in essence, algorithms for processing information and that the difference lies in how they process information [126].

Remark 1.

According to Wiener’s Cybernetics and Qian’s Engineering Cybernetics, the control process is essentially a process concerned with acquiring, organizing, and processing information. Therefore, cognitive control refers to a control method that processes information in a cognitive manner.

However, when Cybernetics was published, cognitive science was still in its infancy. With the development of science and technology, the emergence of technologies such as functional magnetic resonance imaging (fMRI) has provided scientists with the ability to observe brain cells and neurons working mechanisms in cognitive processes at the microscopic level, enabling them to better understand cognition and promote cognitive science development [127,128,129,130].

Intuitively understood, cognition refers to knowing, that is, understanding the development of things and making predictions for the future. Cognition has been studied not only by cognitive scientists but also by psychologists and neurophysiologists. There is still no unified definition of cognition in the academic world; however, scholars have obtained consistency in the cognition process and its related research content. In [131], it was pointed out that cognition is essentially a goal-oriented process of information processing, which further confirms that, from an information point of view, control systems could imitate the mode of human information processing.

Remark 2.

Cognitive science shows that cognitive theories can be applied to guide the design of control systems, changing the way they process information to the human cognition paradigm, which cognitive scientists consider the embodiment of the highest level of human intelligence.

Cognition has its unique architecture and process. The cognitive process is a circular process from perception to action, called PAC in cognitive science, and cognitive architecture is a reflection of the ways in which the human brain and sensory organs organize and work in the cognitive process [13,132].

Cognition in the human brain is mainly performed in the prefrontal cortex. After a long-term study, scientists found that different functions are performed in different areas of the prefrontal cortex during cognitive processes. According to the function of each area, the prefrontal cortex can be further divided into several functional parts, such as perception, attention, learning, reasoning, planning, and decision-making, each of which has its working mode and characteristics, and they are the main research objects of cognitive science [133,134]. At present, they have been thoroughly investigated and even artificially simulated by cognitive scientists, which has promoted the development of artificial intelligence. Because the main purpose of this paper is to discuss concepts and methods of cognitive control from automation perspective, the connotation of cognition and its main functions are not introduced in detail, although the related contents can be found in cognitive science publications, such as [133,135,136].

The cognitive functional parts of the prefrontal cortex are connected in a certain order to compose integral cognition; the cognitive model is the imitation of these cognitive functional parts and their relationships. Due to different perspectives, various cognitive models are formulated during cognitive science development, where some pay attention to reasoning, while others calculate. At present, dozens of cognitive models have gained wide attention, and many studies are related to them [137,138,139,140,141]. For example, the development of cognitive models over the past 20 years has been summarized in [142].

Since cognitive models are the embodiment of cognition, the most convenient and effective method for simulating cognition in control systems is to organize control systems according to cognitive models. Hence, it can be seen from Section 2 that most cognitive control system architectures in both manufacturing and robotics are established based on cognitive models, as is cognitive control based on CDS.

Table 4 summarizes the references in Section 2, in which the cognitive model referenced is clearly stated.

Surveys have shown that various cognitive models are continually evolving and gradually unifying, but many problems remain to be solved. Thus, cognitive models cannot be unified in the short term [142], as is the cognitive control method.

Remark 3.

At present, the design of the cognitive control system is mainly based on the cognitive model. Because a unified cognitive model is not formulated by cognitive scientists, and there are bigger differences in different application scenes in engineering, it is impossible to form a unified analysis and design method for cognitive control like other control methods in the short term.

Remark 4.

Algorithms are also very critical for cognitive control. However, because cognitive control development is still at an early stage, the main work currently is to build up its general concept and basic method. Therefore, discussions related to algorithms are few. In fact, each cognitive model includes corresponding algorithms, but most are unsuitable for direct use in engineering. With the development of cognitive control, the algorithm will also become a research hotspot and even form a unified analysis and synthesis method.

4. Definition of Cognitive Control

In Section 3, cognition and cognitive models are briefly introduced, and the current research status of cognitive control is discussed and summarized. Based on the previous analysis, we propose the following definition of cognitive control from an automation perspective without losing generality.

Cognitive Control Definition.

From an automation perspective, cognitive control is a combination of automatic control and cognitive science; control systems are designed under the guidance of cognitive science theories to imitate human cognition’s architecture and working process, obtaining and processing information as humans do. Cognitive control systems can realize active self-learning and self-optimization in interacting with the environment, enhancing control performance when systems operate in unknown environments or face uncertainties.

In general, cognitive control has the following characteristics:

Instead of trying to identify a permanent, precise, complete, situation-independent model, cognitive control can perform control tasks with temporal, imprecise, incomplete, and situation-dependent models.
The temporal and situation-dependent nature of these models allows substantial simplification: no need is generated for creating very sophisticated models; instead, iterative correcting actions are required.
Asymptotic stability is generally only possible if the controller possesses an analytically exact model with approximate parameters.

5. Fundamental Framework of a Cognitive Control System

In this paper, various cognitive control system architectures are reviewed and analyzed; they all conform to the cognition paradigm, and most are constructed based on cognitive models but with different details. Based on the review of these architectures, a fundamental framework of the cognitive control system is presented from an automation and control perspective, as shown in Figure 17.

As shown in Figure 17, the cognitive control system is mainly composed of two parts, i.e., the perceptual part and the cognitive part; its working process is summarized as follows: the perceptor of the perceptual part processes the sensory measurements, and the perceived relevant information is transmitted to the controlling part. The cognitive controller in the controlling part performs cognitive actions to the physical controller, actuators, and physical system based on the feedback information from the perceptual part. The physical controller generates physical actions, which are the input of actuators. The physical system is operated under the influences of physical actions and cognitive actions. The whole working process of the cognitive control system conforms to the PAC.

PAC. The cognitive control system includes two essential parts: the perceptual part resides on the right-hand side of Figure 17, and the controlling part resides on the left-hand side and they are connected to form the PAC. Therefore, the cognitive control system’s working process accords with the basic form of humans’ cognitive process. In each PAC, two control processes are completed, i.e., the cognitive control process and the physical control process. In order to distinguish the two control processes, the traditional controller is referred to as the physical controller, which performs physical control, and the controller with cognitive control function is called the cognitive controller, while the controlled system is called the physical system.
Perceptual part. The perceptual part receives the sensory data from sensors and extracts relevant information needed by the cognitive and physical controllers to perform control tasks. The extraction of relevant information is completed by the perceptor and the algorithm for the perceptor is decided by the specific application scenes. For example, Bayesian filtering can be used to estimate the system state from noisy measurements, and image processing algorithms can also be applied to analyze road conditions according to pictures captured by sensors [143,144,145,146,147,148]. The perceptual results are fed back to the physical controller and the cognitive controller through the feedback channel; they are used to generate physical actions and cognitive actions to control the physical system. At the same time, the perceptual results are also stored in perceptual memory for extraction when necessary.
Controlling part. The controlling part includes the physical controller, the cognitive controller, and the controlling memory. The physical controller imposes physical actions on the physical system and the cognitive controller imposes cognitive actions on the system.

Cognitive actions for the physical controller: How cognitive actions influence the physical controller depends on the physical control algorithm applied in the physical controller. For instance, cognitive actions could adjust the weight values of PID controller parameters or change the length of the control horizon of the model predictive control algorithm.

Cognitive actions for actuators: Cognitive actions could be applied to actuators to affect their behaviors directly, such as driving motors, enhancing the ability to control actuators. However, overriding control actions by acting directly on the actuators may lead to instability or compromise the quality of control.

Cognitive actions for the physical system: Cognitive actions could change the parameters of the physical system depending on application scenarios. For example, cognitive actions could increase the transmission power and adjust the upper limit of the MAC layer’s retransmission times, reducing the influence of uncertain factors on the wireless communication process.

Cognitive actions are stored in the controlling memory to guide the control process. The cognitive controller is a critical part of the cognitive control system; however, as mentioned in Section 3, there is no unified method for cognitive controller designation. In manufacturing, the cognitive controller is realized through a combination of several modules, such as reasoning, planning, and decision-making, as shown in Figure 3 and Figure 4. In cognitive control systems based on CDS, reinforcement learning algorithms are generally chosen as the algorithm of the cognitive controller because it is shown in cognitive science that reinforcement learning algorithms conform to the working mechanism of the prefrontal cortex and embody human cognitive characteristics [149,150,151].

Memory. Memory includes perceptual memory, controlling memory, and working memory. Both perceptual memory and controlling memory belong to long-term memories, storing information that is necessary for cognitive control, such as groups of physical parameters, perceptual results, and performed cognitive actions for a long time, which are regarded as experiences in cognitive science [152,153] and play an important role when cognitive control systems face unexpected situations.

Working memory is short-term memory and its content could come from the preceptor and cognitive controller, as well as perceptual memory and controlling memory. Working memory is fundamentally used to help select cognitive actions, but in practical applications, it could be used or not [154].

How working memory receives and exchanges information with other memories and modules and helps the cognitive controller select cognitive actions depending on the specific application scenarios is conceptually described in Figure 17.

Attention and intelligence. Attention and intelligence are not shown in Figure 17. Attention has its own physiological structure in the human brain [155,156,157,158], which manifests the goal-oriented characteristics of the human cognitive process. However, control systems are designed for specific tasks, unlike humans, who need to control their own attention. The algorithm in cognitive control systems exhibits task-oriented characteristics; attention manifests itself in an algorithmic manner as perceptual attention in the perceptor and cognitive attention in the cognitive controller. Similar to attention, intelligence does not occupy a physical space within the cognitive control system; instead, its influence is distributed throughout the system, and thereby, it derives its information processing power by exploiting all feedback loops within the system. In short, we may say that the entire PAC is intelligent, and the cognitive control system is an intelligent control system that conforms to the human cognitive paradigm.

Figure 18 illustrates the application of cognitive control which is used to adjust the control horizon of the model predictive-control algorithm and the number of MAC layer retransmissions, mitigating the influences of packet loss and time delay on the wireless network control system (WNCS) [159].

In each PAC, the packet loss rate and time delay of the network are calculated by the perceptor, and then the network state value is calculated based on the packet loss rate and time delay. The value of the ideal network state is defined as 0, which means there is no packet loss and time delay in the network. The information gap is network state minus the ideal network state.

The Q-learning algorithm is used as the cognitive control algorithm; based on the information gap feedbacked by the perceptor, the cognitive controller selects the control horizon and the number of MAC layer retransmissions for the next PAC from the predefined parameters set for the purpose of reducing the information gap.

6. Comparison of Cognitive Control and Similar Control Methods

6.1. Cognitive Control, Cognitive Radio, and Cognitive Radar

Cognitive radio and cognitive radar are both engineering applications related to cognition, and they are conceptually similar to cognitive control, causing misunderstandings for researchers who are not familiar with them. There are similarities and differences among cognitive control, cognitive radio, and cognitive radar; they are all associated with cognition, but their applications, systematic structures, working methods, and task goals are different.

Cognitive radio aims to improve the utilization of the radio spectrum. Simply put, cognitive radio senses the spectrum to identify spectrum holes that represent sub-bands belonging to primary users that are not employed at a particular point in time and space. Spectrum holes are a spectral resource that is essential to cognitive radio, as they can be utilized by secondary users for as long as these holes remain available; hence, the utilization of the radio spectrum is improved.

Cognitive radar optimizes the parameters of the radar transmitted waveform according to prior information about the target and environment, thus improving the radar’s ability to detect and track targets and suppress clutter.

Generally, both cognitive radio and cognitive radar contain control processes, but their control processes differ from those discussed in this paper. The most important feature of cognitive control in the automation field is that the controlled objects are physical entities with the purpose of controlling their movement or dynamic process according to the desired goals. However, controlled objects of cognitive radio and cognitive radar are not physical entities. In addition, cognitive control systems in the automation field have two control processes, that is, the physical control process and the cognitive control process, as shown in Figure 17; however, cognitive radio and cognitive radar only have cognitive control processes. The control area of the system is extended using cognitive control, which can control actuators, physical controllers, and physical systems.

The main characteristics of cognitive control, cognitive radio, and cognitive radar are presented in Table 5.

6.2. Cognitive Control and Intelligent Control

Cognitive control and intelligent control are not separated from each other. Conceptually, cognitive control and intelligent control are both related to the concept of artificial intelligence; hence, cognitive control essentially belongs to the scope of intelligent control. However, there are some differences between cognitive control and other intelligent control methods. Hence, the purpose of comparing and analyzing cognitive control and intelligent control in this subsection is to illustrate further their concepts rather than put them in opposition. According to our investigation, the major differences between cognitive control and intelligent control lie in the following three aspects.

Firstly, the scope of intelligence is different between cognitive control systems and other intelligent control methods. For many intelligent control systems, when we say they are intelligent, in most situations, we mean that their controllers are intelligent. For example, adaptive controllers are intelligent since they can adjust their parameters; the model predictive controllers are intelligent to some extent since they are capable of predicting for several subsequent periods; the controllers established by utilizing the artificial neural network are also considered intelligent since the artificial neural network is considered an intelligent tool. Certainly, other parts of intelligent control systems could also be considered intelligent as intelligence does not always refer to the controller. For cognitive control systems, intelligence is distributed throughout the whole system. The cognitive control systems are established according to the human cognition paradigm; the intelligence of cognitive control systems generally means the systems could perform the whole cognitive process, just like the human, rather than a certain function of cognition, although the constitutional parts of cognitive control systems are usually considered intelligent since they are the functionally decomposing of cognition.

Secondly, cognitive control is still in its infancy. The cognition mechanism of the human has not been fully investigated currently by cognitive scientists, psychologists, and neurophysiologists. Since cognitive science is the significant theoretical guidance for cognitive control in automation and it is developed fast, the methods for establishing cognitive control systems and analyzing their performance are progressing rapidly.

Finally, the establishment of cognitive systems has made higher demands for hardware than other intelligent control methods. The software or programs used to realize the algorithms, such as reinforcement learning, usually need more computing resources due to the nature of algorithms. Their online training is exhaustive for computing resources.

Owing to the author’s limited knowledge and space, various intelligent control methods have not been analyzed in detail. To avoid ambiguity, we want to explain that all control methods, including cognitive control and intelligent control, are continuously developing and improving, and each control method has its own advantages, which are important for realizing control at the human intelligence level.

6.3. Cognitive Control, Robust Control and Adaptive Control

With the popularization of Industry 4.0, the CPS has been widely used in engineering systems, and the uncertainties become non-negligible influences. The common uncertainties include structured uncertainties, parameter uncertainties, network uncertainties such as packet drop and time delay, and other unexpected uncertainties such as malicious cyber-attacks.

In robust control, the uncertainties are described by a perturbation that is considered limited, and as such a set of objects is established. For every object in the set, the controller could ensure the feedback system is internally stable; this controller is considered to have robust stability.

In adaptive control, the control problem is formulated in a way that the controller has some adjustable parameters. The controller is also equipped with an adaptation mechanism for updating the parameters according to variations in dynamics of the system with which it interacts as well as the nature of disturbances.

Typically, these controllers function well under the prespecified conditions for which they are designed. However, they will not function properly if the system of interest confronts highly uncertain environments and so the presence of a human operator in the control loop is indispensable. In such environments, the controller may reach points of surprise for which it has not been programmed. This issue normally arises because the controller is unable to collect the sufficient information it requires to achieve its goals in a self-organized manner.

Cognitive control systems equipped with cognitive abilities are capable of self-learning, self-diagnosis, self-adjustment, and self-optimization based on less prior knowledge; they could perform better when operating in unknown environments and/or confronted with unexpected uncertainties. In [107], it is considered that the lack of memory is the main difference between cognitive control, robust control, and adaptive control; memories, including long-term memory and short-term memory, are critical for self-x abilities and they play important roles in behaviors selection and adjustment, which is significant for the continual progress of the control system in unknown environments and for realizing optimal or suboptimal control. However, equipping the control system with memories will increase the complexity of the system; therefore, the enhancement of the robustness and adaptiveness of the cognitive control system is at the cost of increasing the complexity of the system architecture, which in turn increases the security risks of the system.

A comparison between cognitive control, robust control, and adaptive control is presented in Table 6.

7. Main Issues, Research Directions, and Development Prospects of Cognitive Control

7.1. Main Issues of Cognitive Control

As mentioned in Section 2, cognitive control has mainly been investigated in manufacturing, robotics, and CDS. The former two are application oriented, while cognitive control based on CDS initially focuses on theories and methods, and the outcomes are then applied to engineering systems. As an emerging intelligent control method, cognitive control faces many problems and difficulties during its development; these problems can be summarized in the following four aspects:

Firstly, it is challenging to develop a general theory and method of cognitive control in the short term. Currently, cognitive control system design is primarily based on cognitive models. However, cognitive scientists have yet to obtain consistency in cognitive models. Although common elements exist in various cognitive models, their details are different. In addition, problems faced in different fields are not the same. Consequently, cognitive control methods differ not only in different fields but also within a single field.

Because the algorithms in cognitive models are unsuitable for direct application in cognitive control systems, they need to be redesigned when establishing a cognitive control system based on cognitive models. However, there is no unified standard for the designation of algorithms, and sometimes, no suitable algorithm can be found when realizing a specific cognitive function, which leads to the failure of fully embodying the autonomous learning and optimization ability of cognitive control.

Secondly, there are only a few applications related to cognitive control. Cognitive control is primarily applied in manufacturing and robotics; the research on cognitive control based on CDS is initially focused on theories and methods, and even though it has been applied to CBTC systems, microgrid systems, and vehicular radar and V2V communication systems, very few studies have been conducted. In addition, the portability of established cognitive control systems is low and it is challenging to apply them to other application cases without marked changes. It can be said that the application of cognitive control is more difficult than theoretical research, and instead, an abundance of experimental simulations have been carried out to verify the proposed research.

Thirdly, algorithms in cognitive control systems, such as reinforcement learning, demand a lot of computing power, and the algorithm’s training time is long in complex scenes. To meet the needs of the cognitive control system, more than the existing computer software and hardware is needed. Therefore, ensuring the real-time performance of cognitive control systems in complex scenes is a significant issue in practical applications.

Finally, there is no uniform standard for the performance evaluation of cognitive-control systems. Owing to the differences in research backgrounds and goals of cognitive control, it is difficult to build a reliable evaluation system for cognitive control systems or make transverse comparisons for them.

However, research on cognitive control has made great progress over the past ten years. We believe that, as research develops, cognitive control plays a more important role in the field of intelligent control.

7.2. Research Directions of Cognitive Control

As an emerging control method, much work remains to be undertaken to make cognitive control a relatively independent research field. Cognitive control must be improved in theory and in its applications. First, the general realization and evaluation methods of cognitive control should be explored further. Second, it is necessary to apply the latest research results on machine learning and artificial intelligence to cognitive control systems to improve their performance and efficiency. Third, the application scope of cognitive control needs to be extended and various complex dynamic systems and environments, such as supply chain management systems, healthcare systems, and prognostics and health management systems, should be brought into the application scope. The development of cognitive control can be promoted when it is applied to more scenarios.

7.3. Development Prospects of Cognitive Control

By establishing control systems to conform to the cognition paradigm, cognitive control changes the way to build control systems from an automation perspective. In traditional control systems and general intelligent control systems, there is usually one kind of controller, which is referred to as the physical controller in this article; the physical controller is used to control actuators or the physical systems (when there exist no actuators in the system). In cognitive control systems, there exist two kinds of controllers, i.e., physical controllers and cognitive controllers. The existence of cognitive controllers makes control systems convenient for coping with uncertainties coming from the environment. For example, in [110,111,112,113], the influences of packet drop and time delay in wireless communication are mitigated by using the cognitive controller to change the number of retransmissions of the MAC layer cognitively; in [121,122,123], the cognitive controller is employed to select communication channels to avoid the negative influences of cyber-attacks. Hence, we could not make assumptions about uncertainties and design physical controllers that are tolerable to the assumed uncertainties. From this point of view, cognitive control is prospective in systems that operate in unknown or highly uncertain environments. With the popularization of Industry 4.0, the CPS has been widely used in engineering systems, and the uncertainties coming from the environment become non-negligible influences; the performance of the following kinds of control systems could be improved by cognitive control.

Networked control systems (NCSs). In NCSs, the data and information are transmitted through the wired or wireless network, reducing the geographical restrictions on control systems; however, due to the uncertainties, such as packet drop and time delay, and the unexpected influences, such as cyber-attacks, the NCSs face risks. In traditional control systems, the way to address the influences induced by the network is to make the physical controller tolerable to them. As introduced in [110,111,112,113,121,122,123], cognitive control provides a new way to address these uncertainties; in cognitive control systems, the cognitive controller could be used to adjust the parameters of communication to mitigate the effects of the network on control systems.
Distributed control systems (DCSs). The basic thought of distributed control is decentralized control and centralized management. Generally, a DCS contains many subsystems which are coupled with others. When a certain subsystem fails, the topological structure of the DCS needs to be adjusted; in cognitive control systems, the physical controllers are used to control the corresponding subsystems, and it is convenient to use the cognitive controller to adjust the topological structure, which is a difficult task for the physical controller.
Fault-tolerant control systems. Fault-tolerant control is to make the system have the ability to tolerate faults. In the case of a system failure, the basic function of the system can still be realized. When the actuator, sensor, or component failure occurs in the system, the control system can still ensure its own safe operation through fault-tolerant control adjustment and meets certain performance indicators and requirements. One of the research focuses on fault-tolerant control is to control fault systems. Since the kinds of faults are diverse, they are hard to control by using control methods such as reconfiguring controllers and integrity controller design. At the same time, when a fault occurs, the new controller implemented after the system is reorganized is generally difficult to match the dynamic characteristics of the fault system, which will lead to control failure. In addition, due to the wide application of CPS in engineering systems, some faults are unpredictable and fuzzy, and general fault-tolerant controllers cannot be used to control such faults. In cognitive control systems, cognitive abilities enable the control systems to acquire the fault information and reason on this information during systems operation, hence improving control decisions cycle by cycle until the optimal fault control decision is achieved. The cognitive control systems could operate with small prior knowledge of controlled objects and the cognitive controller could help reconfigure physical controllers or perform cognitive control actions as supplementary to physical actions to make faults better controlled.

8. Conclusions

Cognitive control is a control method that is inspired by the human brain and human cognition. In this paper, the primary research results of cognitive control are reviewed and summarized; then, the general definition and framework are presented with an example to illustrate realization. Some relevant topics are also discussed, such as the main issues, the research directions, and the development prospects of cognitive control. In the final section of this paper, we list eleven major conclusions drawn from our study.

Cognitive control is a control method that establishes control systems based on the human cognition paradigm.
Although cognitive control has been studied for several decades, the unified definition and method are still lacking, and cognitive control is far less mature than other general intelligent control methods.
Cognitive control has been applied in many fields, such as manufacturing, humanoid robots, and transportation systems; although the existing cognitive architectures are different, they share common characteristics, i.e., based on various kinds of cognitive models.
In this paper, cognitive control is defined from an automation perspective: cognitive control is a combination of automatic control and cognitive science; control systems are designed under the guidance of cognitive science theories to imitate human cognition’s architecture and working process, obtaining and processing information as humans do.
The main characteristic of cognitive control systems is that they contain two kinds of controller, i.e., a physical controller to perform physical actions and a cognitive controller to perform cognitive actions. The existence of cognitive controllers makes the control systems flexible to cope with uncertainties that the physical controllers are difficult to address.
Cognitive control is different from the concepts of cognitive radio and cognitive radar.
Cognitive control systems manifest intelligence in a way different from other intelligent control methods.
Compared with robust control and adaptive control, cognitive control is more for engineering systems working in unknown and highly uncertain environments.
The implementation of cognitive control makes higher demands for computing resources.
With the popularization of Industry 4.0, the CPS has been widely used in engineering systems, and the uncertainties coming from the environment become non-negligible influences; cognitive control can be applied to situations such as NCSs, DCSs, and fault-tolerant control systems.
Although cognitive control enhances the capability of traditional systems to cope with uncertainties, the performance and stability are hard to prove theoretically and so m more experiments should be conducted to verify its availability.

Author Contributions

Conceptualization, X.Y.; methodology, X.Y.; formal analysis, Z.L.; investigation, Z.L.; resources, X.Y. and Z.L.; data curation, Z.L.; writing—original draft preparation, Z.L.; writing—review and editing, X.Y.; supervision, X.Y. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

No new data were created or analyzed in this study.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

ACT-R	Adaptive Control of Thought-Rational
BDI	Beliefs-Desires-Intentions
CA	Cognitive Automation
CAV	Connected and Autonomous Vehicle
CBTC	Communication-Based Train Control
CDS	Cognitive Dynamic System
CDT	Cognitive Digital Twin
CEA	Central Executive Agent
CLARION	Connectionist Learning with Adaptive Rule Induction Online
COSA	Cognitive System Architecture
CPS	Cyber-Physical System
CRAM	Cognitive Architecture for Robot Manipulation
CRC	Cognitive Risk Control
CTS	Cognitive Technical System
DCS	Distributed Control System
DRM	Demand Response Management
FDI	False Data Injection
fMRI	functional Magnetic Resonance Imaging
IMA	Intelligent Machine Architecture
ISAC	Intelligent Soft-Arm Control
LIDA	Learning Intelligent Distribution Agent
LTM	Long-Term Memory
MAC	Media Access Control
MECA	Multipurpose Enhanced Cognitive Architecture
MIMO	Multiple-Input Multiple-Output
NCS	Networked Control System
OAR	Object-Attribute-Relation
PAC	Perception-Action Cycle
PIS	Passenger Information System
SAS	Supervisory Attentional System
SCM	Shared Circuit Model
SiMA	Simulation of Mental Apparatus
SINR	Signal-to-Interference-Plus-Noise Ratio
SNR	Signal-to-Noise Ratio
SOAR	State, Operator, and Result
STM	Short-Term Memory
V2V	Vehicle-to-Vehicle
WLAN	Wireless Local Area Network
WM	Working Memory

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. The timeline of the symbolized results of cognitive control, as surveyed in this paper.

Figure 2. The statistics of cognitive control papers investigated in this paper.

Figure 3. The architecture of CTS [9,10].

Figure 4. The cognitive control unit based on SOAR [46].

Figure 5. The cognitive agent based on BDI [47].

Figure 6. Architecture of the cognitive safety controller [52].

Figure 7. 5C architecture for the implementation of CPS [56].

Figure 8. Human operator model [73].

Figure 11. Levels of interaction engagement [84].

Figure 12. SAS for cognitive robots [86].

View Image - Figure 13. Conflicting tasks in WM (left) and the associated network representation (right). Concrete behaviors (dark-gray ovals) compete for shared resources (robot actuators). Their emphasis values ([Forumla omitted. See PDF.] and [Forumla omitted. See PDF.]) are affected by object proximity (bottom-up) and (top-down) aroused by task continuation drives (teleology). In the associated network (right), [Forumla omitted. See PDF.] nodes are for behaviors enabled by provided inputs along with regulations tel, gla, bot, and emphasis values [Forumla omitted. See PDF.] are the outputs [86].

Figure 13. Conflicting tasks in WM (left) and the associated network representation (right). Concrete behaviors (dark-gray ovals) compete for shared resources (robot actuators). Their emphasis values ([Forumla omitted. See PDF.] and [Forumla omitted. See PDF.]) are affected by object proximity (bottom-up) and (top-down) aroused by task continuation drives (teleology). In the associated network (right), [Forumla omitted. See PDF.] nodes are for behaviors enabled by provided inputs along with regulations tel, gla, bot, and emphasis values [Forumla omitted. See PDF.] are the outputs [86].

Figure 14. A schematic representation of the CRAM cognitive architecture [94].

Figure 15. The cognitive control system architecture based on CDS [105].

Figure 16. The CRC system architecture [108].

Figure 17. Fundamental framework of the cognitive control system.

Figure 18. The application of cognitive control in WNCS [159].

Table 1

Cognitive control in manufacturing.

Research Content	References
The concept, characteristics, and architecture of cognitive control	[9,10,11,12,28,44,45,46,47,48,52,53,59,60]
2. The integration of cognitive control system with CPS	[54,55,56,57]
3. Cognitive control based on CDT	[61,62,63,64,65]
4. Algorithm design for the cognitive control system	[49,50,52]

Table 3

Research on cognitive control based on the CDS.

Research Content	References
The concepts and methods of CDS, cognitive control, and CRC	[12,102,103,104,105,107]
2. Cognitive control for CBTC systems	[110,111,112,113]
3. Cognitive control and CRC for smart grids	[114,115,116]
4. CRC for vehicular radar and communication systems	[120,121,122,123,124]

Table 4

The cognitive models utilized in the references.

References	Research Areas	Cognitive Models
[11,36,37]	Manufacturing	SOAR
[48,50,54]	Manufacturing	SCM
[47]	Manufacturing	BDI
[57]	Manufacturing	SiMA
[55]	Manufacturing	SOAR, ACT-R
[54]	Manufacturing	OAR
[68]	Cognitive robot	SOAR, BDI, ACT-R
[70]	Cognitive robot	SOAR, CLARION, LIDA
[72,74,75,76,77]	Cognitive robot	Human operator model
[80,81,83,84,85,86,87,88,89,90,91,92]	Cognitive robot	Baddeley behavior control model
[105,107,108,110,111,112,113,114,115,116,117,118,119,120,121]	CDS	Fuster’s cognitive paradigm

Table 5

The main characteristics of cognitive control, cognitive radio, and cognitive radar.

Technology	Discipline	Purpose	ControlledObjects	PerceptualContent	Controller
Cognitive control	Automation and Control	Ensuring the physicalsystem perform as expected	Physical controller, actuator, and physical system	Physical systemand environment	Physical controllerand cognitive controller
Cognitive radio	Communication	Improving the utilization of radio spectrum	Radio spectrum	Radio spectrum	Dynamic spectrum manager and transmit power controller
Cognitive radar	Electronics, communication, and signal processing	Improving the radar’sabilities for target tracking and clutter suppression	Radar transmitted waveform parameters	Radar returnsand environment	Transmitter

Table 6

Cognitive control, robust control, and adaptive control.

Control Methods	Cognitive Control	Robust Control	Adaptive Control
Concepts	The control system is established based on the human cognition paradigm, with the ability of self-learning, self-diagnosis, self-adjustment, and self-optimization relying on less prior knowledge.	Robust control can ensure the feedback system is internally stable if the uncertainties are under the prespecified limitation.	Adaptive control can adjust system parameters automatically to achieve optimal control.
Characteristics	The system is equipped with cognitive abilities.	The controller has robust ability under limited perturbation.	Identification of system structure, parameters, and performance indicators.
Objects	Unknown and highly uncertain environments.	Uncertainties under the prespecified limitation.	The uncertainties of system and environment
Targets	The system performs well in highly uncertain environments.	The system is stable under the influences of limited perturbation.	When the environment changes, adaptive control can be realized by adjusting the parameters.

References

1. Harapanahalli, A.; Jafarpour, S.; Coogan, S. Forward invariance in neural network controlled systems. IEEE Contr. Syst. Lett.; 2023; 7, pp. 3962-3967. [DOI: https://dx.doi.org/10.1109/LCSYS.2023.3341980]

2. Liu, K.; Hashemkhani, S.; Rubin, J.; Kubendran, R. BioNN: Bio-mimetic neural networks on hardware using nonlinear multi-timescale mixed-feedback control for neuromodulatory bursting rhythms. IEEE J. Emerg. Sel. Topics Power Electron.; 2023; 23, pp. 914-926. [DOI: https://dx.doi.org/10.1109/JETCAS.2023.3330084]

3. Wei, L.; Wang, X.; Li, L.; Fan, Z.; Dou, R.; Lin, J. T-S fuzzy model predictive control for vehicle yaw stability in nonlinear region. IEEE Trans. Veh. Technol.; 2021; 70, pp. 7536-7546. [DOI: https://dx.doi.org/10.1109/TVT.2021.3091809]

4. Zhong, Z.; Wang, X.; Lam, H. Finite-time fuzzy sliding mode control for nonlinear descriptor systems. IEEE-CAA J. Autom. Sin.; 2021; 8, pp. 1141-1152. [DOI: https://dx.doi.org/10.1109/JAS.2021.1004024]

5. Huang, X.; Huss, J.; North, L.; Williams, K.; Boyd-Devine, A. Cognitive and motivational benefits of a theory-based immersive virtual reality design in science learning. Comput. Educ. Open; 2023; 4, 100124. [DOI: https://dx.doi.org/10.1016/j.caeo.2023.100124]

6. Lewis, N.A. What would make cognitive science more useful?. Trends Cogn. Sci.; 2022; 26, pp. 1038-1039. [DOI: https://dx.doi.org/10.1016/j.tics.2022.07.005]

7. Rosenblueth, A.; Wiener, N.; Bigelow, J. Behavior, purpose and teleology. Philos. Sci.; 1943; 10, pp. 18-24. [DOI: https://dx.doi.org/10.1086/286788]

8. Brachman, R.J. Systems that know what they’re doing. IEEE Intell. Syst.; 2002; 17, pp. 67-71. [DOI: https://dx.doi.org/10.1109/MIS.2002.1134363]

9. Beetz, M.; Buss, M.; Wollherr, D. Cognitive technical systems—What is the role of artificial intelligence?. Proceedings of the 30th Annual German Conference on AI; Osnabrück, Germany, 10–13 September 2007.

10. Buss, M.; Beetz, M. COTESYS—Cognition for technical systems. Künstl. Intell.; 2010; 24, pp. 323-327. [DOI: https://dx.doi.org/10.1007/s13218-010-0061-z]

11. Putzer, H.; Onken, R. COSA—A generic cognitive system architecture based on a cognitive model of human behavior. Cogn. Technol. Work; 2003; 5, pp. 140-151. [DOI: https://dx.doi.org/10.1007/s10111-003-0120-4]

12. Haykin, S. Cognitive dynamic systems: Radar, control, and radio. Proc. IEEE; 2012; 100, pp. 2095-2103. [DOI: https://dx.doi.org/10.1109/JPROC.2012.2193709]

13. Fuster, J.M. Cortex and Mind: Unifying Cognition; Oxford University Press: Oxford, UK, 2010.

14. Keliang, Z.; Taigang, L.; Lifeng, Z. Industry 4.0: Towards future industrial opportunities and challenges. Proceedings of the 12th International Conference on Fuzzy Systems and Knowledge Discovery; Zhangjiajie, China, 15–17 August 2015.

15. Jazdi, N. Cyber physical systems in the context of Industry 4.0. Proceedings of the International Conference on Automation, Quality and Testing, Robotics; Cluj-Napoca, Romania, 22–24 May 2014.

16. Mohammad, N.H.R.; Chinnasamy, A.N.M.; Sreenivasan, J.; Muhammad, M. Industry 4.0—Technological revolution and sustainable firm performance. Proceedings of the 2021 Emerging Trends in Industry 4.0; Raigarh, India, 19–21 May 2021.

17. Saeid, N. Industry 5.0—A human-centric solution. Sustainability; 2019; 11, 4371. [DOI: https://dx.doi.org/10.3390/su11164371]

18. Alexey, Y.Z.; Vasily, N.G. Industry 4.0: Underlying technologies. Industry 5.0: Human-computer interaction as a tech bridge from Industry 4.0 to Industry 5.0. Proceedings of the 9th International Conference on Web Research; Tehran, Iran, 3–4 May 2023.

19. Ndala, Y.M. Industry 5.0 a novel technological concept. Proceedings of the International Conference on Smart Applications, Communications and Networking; Harrisonburg, VA, USA, 28–30 May 2024.

20. Aeshita, M.; Ameesha, D.; Nikhil, S. Evolution from Industry 1.0 to Industry 5.0. Proceedings of the 4th International Conference on Advances in Computing, Communication Control and Networking; Greater Noida, India, 16–17 December 2022.

21. Thomas, G.; Mario, H.; Benjamin, K.; Reinhard, B.; Alf, J.I. The autonomous industrial plant-future of process engineering, operations and maintenance. IFAC-PapersOnLine; 2020; 52, pp. 454-460.

22. Ji, Z.; Yanhong, Z.; Baicun, W.; Jiyuan, Z. Human–cyber–physical systems (HCPSs) in the context of new-generation intelligent manufacturing. Engineering; 2019; 5, pp. 624-636.

23. Ziang, L.; Jianhua, S.; Ziren, L.; Minghao, C.; Jiafu, W. Intelligent manufacturing from the perspective of Industry 5.0: Application review and prospects. IEEE Access; 2024; 12, pp. 167436-167451.

24. Giuliano, A.; Hilal, W.; Alsadi, N.; Gadsden, S.A.; Yawney, J. A review of cognitive dynamic systems and cognitive IoT. Proceedings of the IOT, Electronics and Mechatronics Conference; Toronto, ON, Canada, 1–4 June 2022.

25. Hilal, W.; Giuliano, A.; Gadsden, S.A.; Yawney, J. A review of cognitive dynamic systems and its overarching functions. Proceedings of the IOT, Electronics and Mechatronics Conference; Toronto, ON, Canada, 1–4 June 2022.

26. Hilal, W.; Gadsden, S.A.; Yawney, J. Cognitive dynamic systems: A review of theory, applications, and recent advances. Proc. IEEE; 2023; 111, pp. 575-622. [DOI: https://dx.doi.org/10.1109/JPROC.2023.3272577]

27. Naghshvarianjahromi, M.; Kumar, S.; Deen, M.J. Natural intelligence as the brain of intelligent systems. Sensors; 2023; 23, 2859. [DOI: https://dx.doi.org/10.3390/s23052859]

28. Kalach, F.E.; Yousif, I.; Wuest, T.; Sheth, A.; Harik, R. Cognitive manufacturing: Definition and current trends. J. Intell. Manuf; 2024; pp. 1-21. [DOI: https://dx.doi.org/10.1007/s10845-024-02429-9]

29. Liu, S.; Wang, L.; Gao, R.X. Cognitive neuroscience and robotics: Advancements and future research directions. Robot. Comput. Integr. Manuf.; 2024; 85, 102610. [DOI: https://dx.doi.org/10.1016/j.rcim.2023.102610]

30. Wu, X.; Ma, J.; Xue, X. Joint beamforming for secure communication in RIS-assisted cognitive radio networks. J. Commun. Netw.; 2022; 24, pp. 518-529. [DOI: https://dx.doi.org/10.23919/JCN.2022.000044]

31. Kumar, A.; Kumar, K. A game theory based hybrid NOMA for efficient resource optimization in cognitive radio networks. IEEE Trans. Netw. Sci. Eng.; 2021; 8, pp. 3501-3514. [DOI: https://dx.doi.org/10.1109/TNSE.2021.3116669]

32. Zhang, C.; Pei, J.; Zhang, Y.; Zhang, Y.; Huo, W.; Huang, Y. Slow-time ambiguity function shaping with spectral coexistence for cognitive radar. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2023; 16, pp. 8418-8430. [DOI: https://dx.doi.org/10.1109/JSTARS.2023.3312516]

33. Shan, Z.; Liu, P.; Wang, L.; Liu, Y. A cognitive multi-carrier radar for communication interference avoidance via deep reinforcement learning. IEEE Trans. Cogn. Commun. Netw.; 2023; 9, pp. 1561-1578. [DOI: https://dx.doi.org/10.1109/TCCN.2023.3306854]

34. Zhu, P.; Liang, J.; Luo, Z.; Shen, X. Cognitive radar target tracking using intelligent waveforms based on reinforcement learning. IEEE Trans. Geosci. Remote Sens.; 2023; 61, 5107315. [DOI: https://dx.doi.org/10.1109/TGRS.2023.3298355]

35. Howard, W.W.; Buehrer, R.M. Hybrid cognition for target tracking in cognitive radar networks. IEEE Trans. Radar Syst.; 2023; 1, pp. 118-131. [DOI: https://dx.doi.org/10.1109/TRS.2023.3282846]

36. Rajangam, M.; Annamalai, C. Extractive document summarization using an adaptive, knowledge based cognitive model. Cogn. Syst. Res.; 2019; 56, pp. 56-71. [DOI: https://dx.doi.org/10.1016/j.cogsys.2018.11.005]

37. García-Mieres, H.; Lundin, N.B.; Minor, K.S.; Dimaggio, G.; Popolo, R.; Cheli, S.; Lysaker, P.H. A cognitive model of diminished expression in schizophrenia: The interface of metacognition, cognitive symptoms and language disturbances. J. Psychiatr. Res.; 2020; 131, pp. 169-176. [DOI: https://dx.doi.org/10.1016/j.jpsychires.2020.09.008]

38. Worzberger, M.; Borst, K.P.; Krems, J.F.; Rey, G.D. Memory-related cognitive load effects in an interrupted learning task: A model-based explanation. Trends Neurosci. Educ.; 2020; 20, 100139. [DOI: https://dx.doi.org/10.1016/j.tine.2020.100139]

39. Faghani, N.; Akbari, M.; Hasani, J.; Marino, C. An emotional and cognitive model of problematic Internet use among college students: The full mediating role of cognitive factors. Addict. Behav.; 2020; 105, 106252. [DOI: https://dx.doi.org/10.1016/j.addbeh.2019.106252]

40. Sebastian, A.; Forstmann, B.U.; Matzke, D. Towards a model-based cognitive neuroscience of stopping—A neuroimaging perspective. Neurosci. Biobehav. Rev.; 2018; 90, pp. 130-136. [DOI: https://dx.doi.org/10.1016/j.neubiorev.2018.04.011]

41. Zhang, X.; Fan, Z.; Shen, Y.; Dai, J. Evidence accumulation is not essential for generating intertemporal preference: A comparison of dynamic cognitive models of matching tasks. Cogn. Psychol.; 2023; 147, 101615. [DOI: https://dx.doi.org/10.1016/j.cogpsych.2023.101615] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37871413]

42. Nejati, V.; Famininejad, Z.; Rad, J.A. Neural symphony of risky decision making in children with ADHD: Insights from transcranial alternating current stimulation and cognitive modeling. Neurophysiol. Clin.; 2023; 53, 102898. [DOI: https://dx.doi.org/10.1016/j.neucli.2023.102898] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37659136]

43. Avishek, M.; Divya, A.B.; Sivvani, M.; Surjya, K.P. Cognitive intelligence in industrial robots and manufacturing. Comput. Ind. Eng.; 2024; 191, 110106.

44. Schulte, A.; Meitinger, C.; Onken, R. Human factors in the guidance of uninhabited vehicles: Oxymoron or tautology?. Cogn. Technol. Work; 2009; 11, pp. 71-86. [DOI: https://dx.doi.org/10.1007/s10111-008-0123-2]

45. Pecher, W.; Brüggenwirth, S.; Schulte, A. Using cognitive automation for aircraft general systems management. Proceedings of the 5th International Conference on System of Systems Engineering; Loughborough, UK, 22–24 June 2010.

46. Hauck, E.; Gramatke, A.; Henning, K. Cognitive technical systems in a production environment: Outline of a possible approach. Proceedings of the 5th International Conference on Informatics in Control, Automation and Robotics; Funchal, Portugal, 11–15 May 2008.

47. Park, H.; Tran, N. A cognitive agent based manufacturing system adapting to disturbances. Int. J. Control Autom. Syst.; 2012; 10, pp. 806-816. [DOI: https://dx.doi.org/10.1007/s12555-012-0417-2]

48. Boza, A.S.; Guerra, R.H.; Gajate, A. Artificial cognitive control system based on the shared circuits model of sociocognitive capacities. A first approach. Eng. Appl. Artif. Intell.; 2011; 24, pp. 209-219. [DOI: https://dx.doi.org/10.1016/j.engappai.2010.10.005]

49. Beruvides, G.; Juanes, C.; Castaño, F.; Haber, R.E. A self-learning strategy for artificial cognitive control systems. Proceedings of the IEEE 13th International Conference on Industrial Informatics; Cambridge, UK, 22–24 July 2015.

50. Haber, R.E.; Juanes, C.; del Toro, R.; Beruvides, G. Artificial cognitive control with self-x capabilities: A case study of a micro-manufacturing process. Comput. Ind.; 2015; 74, pp. 135-150. [DOI: https://dx.doi.org/10.1016/j.compind.2015.05.001]

51. Wendt, A.; Kollmann, S.; Bratukhin, A.; Estaji, A.; Sauter, T.; Jantsch, A. Cognitive architectures for process monitoring—An Analysis. Proceedings of the IEEE 18th International Conference on Industrial Informatics; Warwick, UK, 20–23 July 2020.

52. Kain, S.; Ding, H.; Schiller, F.; Stursberg, O. Controller architecture for safe cognitive technical systems. Proceedings of the 26th International Conference on Computer Safety, Reliability, and Security; Nuremberg, Germany, 18–21 September 2007.

53. Zaeh, M.F.; Reinhart, G.; Ostgathe, M.; Geiger, F.; Lau, C. A holistic approach for the cognitive control of production systems. Adv. Eng. Inform.; 2010; 24, pp. 300-307. [DOI: https://dx.doi.org/10.1016/j.aei.2010.05.014]

54. Iarovyi, S.; Lastra, J.L.M.; Haber, R.; Del Toro, R. From artificial cognitive systems and open architectures to cognitive manufacturing systems. Proceedings of the IEEE 13th International Conference on Industrial Informatics; Cambridge, UK, 22–24 July 2015.

55. Bunte, A.; Fischbach, A.; Strohschein, J.; Bartz-Beielstein, T.; Faeskorn-Woyke, H.; Niggemann, O. Evaluation of cognitive architectures for cyber-physical production systems. Proceedings of the IEEE International Conference on Emerging Technologies and Factory Automation; Zaragoza, Spain, 22 February 2019.

56. Lee, J.; Bagheri, B.; Kao, H. A cyber-physical systems architecture for industry 4.0 based manufacturing systems. Manuf. Lett.; 2015; 3, pp. 18-23. [DOI: https://dx.doi.org/10.1016/j.mfglet.2014.12.001]

57. Siafara, L.C.; Kholerdi, H.; Bratukhin, A.; Taherinejad, N.; Jantsch, A. SAMBA—An architecture for adaptive cognitive control of distributed cyber-physical production systems based on its self-awareness. Elektrotechnik Informationstechnik; 2018; 135, pp. 270-277. [DOI: https://dx.doi.org/10.1007/s00502-018-0614-7]

58. Unver, H.O. An ISA-95-based manufacturing intelligence system in support of lean initiatives. Int. J. Adv. Manuf. Technol.; 2013; 65, pp. 853-866. [DOI: https://dx.doi.org/10.1007/s00170-012-4223-z]

59. Carpanzano, E.; Knüttel, D. Advances in artificial intelligence methods applications in industrial control systems: Towards cognitive self-optimizing manufacturing systems. Appl. Sci.; 2022; 12, 10962. [DOI: https://dx.doi.org/10.3390/app122110962]

60. Hadi, M.A.; Kraus, D.; Kajmakovic, A.; Suschnigg, J.; Guiza, O.; Gashi, M.; Sopidis, G.; Vukovic, M.; Milenkovic, K.; Haslgruebler, M. et al. Towards flexible and cognitive production—Addressing the production challenges. Appl. Sci.; 2022; 12, 8696. [DOI: https://dx.doi.org/10.3390/app12178696]

61. Alimam, H.; Mazzuto, G.; Tozzi, N.; Ciarapica, F.E. The resurrection of digital triplet: A cognitive pillar of human-machine integration at the dawn of industry 5.0. J. King Saud Univ.—Comput. Inf. Sci.; 2023; 35, 101846. [DOI: https://dx.doi.org/10.1016/j.jksuci.2023.101846]

62. Fernández, F.; Sánchez, Á.; Vélez, J.F.; Moreno, A.B. Symbiotic autonomous systems with consciousness using digital twins. Proceedings of the 8th International Work-Conference on the Interplay Between Natural and Artificial Computation; Pueto de la Cruz, Spain, 3–7 June 2019.

63. Ali, M.I.; Patel, P.; Breslin, J.G.; Harik, R.; Sheth, A. Cognitive digital twins for smart manufacturing. IEEE Intell. Syst.; 2021; 36, pp. 96-100.

64. EIMaraghy, H.; EIMaraghy, W. Adaptive cognitive manufacturing system (ACMS)—A new paradigm. Int. J. Prod. Res.; 2022; 60, pp. 7436-7449. [DOI: https://dx.doi.org/10.1080/00207543.2022.2078248]

65. Jiang, T.; Zhou, J.; Zhao, J.; Wang, M.; Zhang, S. A multi-dimensional cognitive framework for cognitive manufacturing based on OAR model. J. Manuf. Syst.; 2022; 65, pp. 469-485. [DOI: https://dx.doi.org/10.1016/j.jmsy.2022.09.019]

66. Sun, F.; Huang, W.; Luo, Y.; Ji, T.; Liu, H.; Liu, H.; Zhang, J. Robot cognitive learning by considering physical properties. Engineering; 2024; [DOI: https://dx.doi.org/10.1016/j.eng.2024.10.013]

67. Tusseyeva, I.; Sandygulova, A.; Rubagotti, M. Perceived intelligence in human-robot interaction: A review. IEEE Access; 2024; 12, pp. 151348-151359. [DOI: https://dx.doi.org/10.1109/ACCESS.2024.3478751]

68. Wei, C.; Hindriks, K.V. An agent-based cognitive robot architecture. Proceedings of the International Workshop on Programming Multi-Agent Systems; Valencia, Spain, 5 June 2012.

69. Liu, Y.; Tian, Z.; Liu, Y.; Li, J.; Fu, F. Cognitive architecture based simulation of perception and behavior controls for robot. Proceedings of the International Conference on Neuroergonomics and Cognitive Engineering; Orlando, FL, USA, 21–25 July 2008.

70. Gudwin, R.; Rohmer, E.; Paraense, A.; Fróes, E.; Gibaut, W.; Oliveira, I.; Rocha, S.; Raizer, K.; Feljan, A.V. The TROCA project: An autonomous transportation robot controlled by a cognitive architecture. Cogn. Syst. Res.; 2020; 59, pp. 179-197. [DOI: https://dx.doi.org/10.1016/j.cogsys.2019.09.011]

71. Gudwin, R.; Paraense, A.; de Paula, S.M.; Fróes, E.; Gibaut, W.; Castro, E.; Figueiredo, V.; Raizer, K. The multipurpose enhanced cognitive architecture (MECA). Biol. Inspired Cogn. Archit.; 2017; 22, pp. 20-34. [DOI: https://dx.doi.org/10.1016/j.bica.2017.09.006]

72. Brooks, R.A. A robust layered control system for a mobile robot. IEEE J. Robot. Autom.; 1986; RA-2, pp. 14-23. [DOI: https://dx.doi.org/10.1109/JRA.1986.1087032]

73. Rasmussen, J. Skills, rules, and knowledge: Signals, signs, and symbols, and other distinctions in human performance models. IEEE Trans. Syst. Man. Cybern.; 1983; SMC-13, pp. 257-266. [DOI: https://dx.doi.org/10.1109/TSMC.1983.6313160]

74. Breazeal, C. Socially intelligent robots: Research, development, and applications. Proceedings of the IEEE International Conference on Systems, Man and Cybernetics. e-Systems and e-Man for Cybernetics in Cyberspace; Tucson, AZ, USA, 7–10 October 2001.

75. Arkin, R.C.; Fujita, M.; Takagi, T.; Hasegawa, R. An ethological and emotional basis for human–robot interaction. Robot. Auton. Syst.; 2003; 42, pp. 191-201. [DOI: https://dx.doi.org/10.1016/S0921-8890(02)00375-5]

76. Hackel, M.; Schwope, S.; Fritsch, J.; Wrede, B.; Sagerer, G. A humanoid robot platform suitable for studying embodied interaction. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems; Edmonton, AB, Canada, 2–6 August 2005.

77. Burghart, C.; Mikut, R.; Stiefelhagen, R.; Asfour, T.; Holzapfel, H.; Steinhaus, P.; Dillmann, R. A cognitive architecture for a humanoid robot: A first approach. Proceedings of the 5th IEEE-RAS International Conference on Humanoid Robots; Tsukuba, Japan, 5 December 2005.

78. Vahrenkamp, N.; Wächter, M.; Kröhnert, M.; Welke, K.; Asfour, T. The robot software framework ArmarX. Inf. Technol.; 2015; 57, pp. 99-111. [DOI: https://dx.doi.org/10.1515/itit-2014-1066]

79. Pack, R.T.; Wilkes, D.M.; Kawamura, K. A software architecture for integrated service robot development. Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics. Computational Cybernetics and Simulation; Orlando, FL, USA, 12–15 October 1997.

80. Kawamura, K.; Rogers, T.E.; Ao, X. Development of a cognitive model of humans in a multi-agent framework for human-robot interaction. Proceedings of the First International Joint Conference on Autonomous Agents and Multiagent Systems; Bologna, Italy, 15–19 July 2002.

81. Ratanaswasd, P.; Dodd, W.; Kawamura, K.; Noelle, D.C. Modular behavior control for a cognitive robot. Proceedings of the 12th International Conference on Advanced Robotics; Seattle, WA, USA, 18–20 July 2005.

82. Baddeley, A.D. Working Memory; Clarendon Press: Oxford, UK, 1986.

83. Kawamura, K.; Noelle, D.C.; Hambuchen, K.A.; Rogers, T.E.; Turkay, E. A multi-agent approach to self-reflection for cognitive robotics. Proceedings of the 11th International Conference on Advanced Robotic; Coimbra, Portugal, 30 June–3 July 2003.

84. Kawamura, K.; Peters, R.A., II; Bodenheimer, R.E.; Sarkar, N.; Park, J.; Clifton, C.A.; Spratley, A.W. A parallel distributed cognitive control system for a humanoid robot. Int. J. Humanoid Robot.; 2004; 1, pp. 65-93. [DOI: https://dx.doi.org/10.1142/S021984360400006X]

85. Gordon, S.M.; Kawamura, K.; Wilkes, D.M. Neuromorphically inspired appraisal-based decision making in a cognitive robot. IEEE Trans. Auton. Ment. Dev.; 2010; 2, pp. 17-39. [DOI: https://dx.doi.org/10.1109/TAMD.2010.2043530]

86. Caccavale, R.; Finzi, A. A robotic cognitive control framework for collaborative task execution and learning. Top. Cogn. Sci.; 2021; 14, pp. 327-343. [DOI: https://dx.doi.org/10.1111/tops.12587]

87. Broquère, X.; Finzi, A.; Mainprice, J.; Rossi, S.; Sidobre, D.; Staffa, M. An attentional approach to human-robot interactive manipulation. Int. J. Soc. Robot.; 2014; 6, pp. 533-553. [DOI: https://dx.doi.org/10.1007/s12369-014-0236-0]

88. Caccavale, R.; Cacace, J.; Fiore, M.; Alami, R.; Finzi, A. Attentional supervision of human-robot collaborative plans. Proceedings of the 25th IEEE International Symposium on Robot and Human Interactive Communication; New York, NY, USA, 26–31 August 2016.

89. Caccavale, R.; Saveriano, M.; Fontanelli, G.A.; Ficuciello, F.; Lee, D.; Finzi, A. Imitation learning and attentional supervision of dual-arm structured tasks. Proceedings of the Joint IEEE International Conference on Development and Learning and Epigenetic Robotics; Lisbon, Portugal, 18–21 September 2017.

90. Caccavale, R.; Finzi, A. Flexible task execution and attentional regulations in human–robot interaction. IEEE Trans. Cogn. Dev. Syst.; 2017; 9, pp. 68-79. [DOI: https://dx.doi.org/10.1109/TCDS.2016.2614690]

91. Caccavale, R.; Finzi, A. Learning attentional regulations for structured tasks execution in robotic cognitive control. Auton. Robot.; 2019; 43, pp. 2229-2243. [DOI: https://dx.doi.org/10.1007/s10514-019-09876-x]

92. Caccavale, R.; Saveriano, M.; Finzi, A.; Lee, D. Kinesthetic teaching and attentional supervision of structured tasks in human–robot interaction. Auton. Robot.; 2018; 43, pp. 1291-1307. [DOI: https://dx.doi.org/10.1007/s10514-018-9706-9]

93. Beetz, M.; Mösenlechner, L.; Tenorth, M. CRAM—A cognitive robot abstract machine for everyday manipulation in human environments. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems; Taipei, Taiwan, 18–22 October 2010.

94. Beetz, M.; Kazhoyan, G.; Vernon, D. The CRAM cognitive architecture for robot manipulation in everydday activities. arXiv; 2023; arXiv: 2304.14119

95. Beetz, M.; Tenorth, M.; Winkler, J. Open-EASE—A knowledge processing service for robots and robotics/AI researchers. Proceedings of the IEEE International Conference on Robotics and Automation; Seattle, WA, USA, 26–30 May 2015.

96. Beetz, M.; Bálint-Benczédi, F.; Blodow, N.; Nyga, D. ROBOSHERLOCK: Unstructured information processing for robot perception. Proceedings of the IEEE International Conference on Robotics and Automation; Seattle, WA, USA, 26–30 May 2015.

97. Bálint-Benczédi, F.; Márton, Z.; Durner, M.; Beetz, M. Storing and retrieving perceptual episodic memories for long-term manipulation tasks. Proceedings of the 18th International Conference on Advanced Robotics; Hong Kong, China, 10–12 July 2017.

98. Bálint-Benczédi, F.; Beetz, M. Variations on a theme: “It’s a poor sort of memory that only works backwards”. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems; Madrid, Spain, 1–5 October 2018.

99. Mania, P.; Beetz, M. A framework for self-training perceptual agents in simulated photorealistic environments. Proceedings of the International Conference on Robotics and Automation; Montreal, QC, Canada, 20–24 May 2019.

100. Kenfack, F.K.; Siddiky, F.A.; Bálint-Benczédi, F.; Beetz, M. RobotVQA—A scene graph- and deep-learning-based visual question answering system for robot manipulation. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems; Las Vegas, NV, USA, 24 October 2020–24 January 2021.

101. Kazhoyan, G.; Stelter, S.; Kenfack, F.K.; Koralewski, S.; Beetz, M. The robot household marathon experiment. Proceedings of the IEEE International Conference on Robotics and Automation; Xi’an, China, 30 May–5 June 2021.

102. Haykin, S. Cognitive radar: A way of the future. IEEE Signal Process. Mag.; 2006; 23, pp. 30-40. [DOI: https://dx.doi.org/10.1109/MSP.2006.1593335]

103. Haykin, S. Cognitive radio: Brain-empowered wireless communications. IEEE J. Sel. Areas Commun.; 2005; 23, pp. 201-220. [DOI: https://dx.doi.org/10.1109/JSAC.2004.839380]

104. Haykin, S. Cognitive dynamic systems. Proc. IEEE; 2006; 94, pp. 1910-1911. [DOI: https://dx.doi.org/10.1109/JPROC.2006.886014]

105. Haykin, S.; Fatemi, M. Cognitive control. Proc. IEEE; 2012; 100, pp. 3156-3169. [DOI: https://dx.doi.org/10.1109/JPROC.2012.2215773]

106. Berger, E.L. Statistical Decision Theory and Bayesian Analysis; 2nd ed. Springer: New York, NY, USA, 1985.

107. Haykin, S.; Fuster, J.M.; Findlay, D.; Feng, S. Cognitive risk control for physical systems. IEEE Access; 2017; 5, pp. 14664-14679. [DOI: https://dx.doi.org/10.1109/ACCESS.2017.2726439]

108. Fatemi, M.; Haykin, S. Cognitive control: Theory and application. IEEE Access; 2014; 2, pp. 698-710. [DOI: https://dx.doi.org/10.1109/ACCESS.2014.2332333]

109. Fuster, J.M. The prefrontal cortex makes the brain a preadaptive system. Proc. IEEE; 2014; 102, pp. 417-426. [DOI: https://dx.doi.org/10.1109/JPROC.2014.2306250]

110. Wang, H.; Yu, F.R.; Zhu, L.; Tang, T.; Ning, B. A cognitive control approach to communication-based train control systems. IEEE Trans. Intell. Transp. Syst.; 2015; 16, pp. 1676-1689. [DOI: https://dx.doi.org/10.1109/TITS.2014.2377115]

111. Wang, H.; Yu, F.R.; Wang, H. A cognitive control approach to interference mitigation in communications-based train control (CBTC) co-existing with passenger information systems (PISs). EURASIP J. Wirel. Commun. Netw.; 2017; 2017, 186. [DOI: https://dx.doi.org/10.1186/s13638-017-0959-3]

112. Sun, W.; Yu, F.R.; Tang, T. Performance improvement in communication-based train control (CBTC) systems using cognitive control. Proceedings of the 17th International IEEE Conference on Intelligent Transportation Systems; Qingdao, China, 8–11 October 2014.

113. Sun, W.; Yu, F.R.; Tang, T.; You, S. A cognitive control method for cost-efficient CBTC systems with smart grids. IEEE Trans. Intell. Transp. Syst.; 2017; 18, pp. 568-582. [DOI: https://dx.doi.org/10.1109/TITS.2016.2586938]

114. Fang, X.; Han, Y.; Wang, J.; Zhao, Q. A cognitive control approach for microgird performance optimization in unstable wireless communication. Neurocomputing; 2019; 355, pp. 168-182. [DOI: https://dx.doi.org/10.1016/j.neucom.2019.04.048]

115. Oozeer, M.I.; Haykin, S. Cognitive dynamic system for control and cyber-attack detection in smart grid. IEEE Access; 2019; 7, pp. 78320-78335. [DOI: https://dx.doi.org/10.1109/ACCESS.2019.2922410]

116. Oozeer, M.I.; Haykin, S. Cognitive risk control for mitigating cyber-attack in smart grid. IEEE Access; 2019; 7, pp. 125806-125826. [DOI: https://dx.doi.org/10.1109/ACCESS.2019.2939089]

117. Kukkala, V.K.; Tunnell, J.; Pasricha, S.; Bradley, T. Advanced driverassistance systems: A path toward autonomous vehicles. IEEE Consum. Electron. Mag.; 2018; 7, pp. 18-25. [DOI: https://dx.doi.org/10.1109/MCE.2018.2828440]

118. Hasch, J.; Topak, E.; Schnabel, R.; Zwick, T.; Weigel, R.; Waldschmidt, C. Millimeter-wave technology for automotive radar sensors in the 77 GHz frequency band. IEEE Trans. Microw. Theory Techn.; 2012; 60, pp. 845-860. [DOI: https://dx.doi.org/10.1109/TMTT.2011.2178427]

119. Schwarz, S.; Philosof, T.; Rupp, M. Signal processing challenges in cellular-assisted vehicular communications: Efforts and developments within 3GPP LTE and beyond. IEEE Signal Process. Mag.; 2017; 34, pp. 47-59. [DOI: https://dx.doi.org/10.1109/MSP.2016.2637938]

120. Feng, S.; Haykin, S. Cognitive risk control for transmit-waveform selection in vehicular radar systems. IEEE Trans. Veh. Technol.; 2018; 67, pp. 9542-9556. [DOI: https://dx.doi.org/10.1109/TVT.2018.2857718]

121. Feng, S.; Haykin, S. V2V communication-assisted transmit-waveform selection for cognitive vehicular radars. Proceedings of the IEEE Canadian Conference of Electrical and Computer Engineering; Edmonton, AB, Canada, 5–8 May 2019.

122. Feng, S.; Haykin, S. Anti-jamming V2V communication in an integrated UAV-CAV network with hybrid attackers. Proceedings of the IEEE International Conference on Communications; Shanghai, China, 20–24 May 2019.

123. Feng, S.; Haykin, S. Cognitive risk control for anti-jamming V2V communications in autonomous vehicle networks. IEEE Trans. Veh. Technol.; 2019; 68, pp. 9920-9934. [DOI: https://dx.doi.org/10.1109/TVT.2019.2935999]

124. Feng, S.; Haykin, S. Coordinated cognitive risk control for bridging vehicular radar and communication systems. IEEE Trans. Intell. Transp. Syst.; 2022; 23, pp. 4135-4150. [DOI: https://dx.doi.org/10.1109/TITS.2020.3041647]

125. Wiener, N. Cybernetics: Or the Control and Communication in the Animal and the Machine; 2nd ed. MIT Press: Cambridge, MA, USA, 1965.

126. Tsien, H.S. Engineering Cybernetics; McGraw-Hill: New York, NY, USA, 1954.

127. Steel, A.; Garcia, B.D.; Silson, E.H.; Robertson, C.E. Evaluating the efficacy of multi-echo ICA denoising on model-based fMRI. NeuroImage; 2022; 264, 119723. [DOI: https://dx.doi.org/10.1016/j.neuroimage.2022.119723] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36328274]

128. Heugel, N.; Beardsley, S.A.; Liebenthal, E. EEG and fMRI coupling and decoupling based on joint independent component analysis (jICA). J. Neurosci. Methods; 2022; 369, 109477. [DOI: https://dx.doi.org/10.1016/j.jneumeth.2022.109477] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34998799]

129. Gutierrez-Barragan, D.; Basson, M.A.; Panzeri, S. Infraslow state fluctuations govern spontaneous fMRI network dynamics. Curr. Biol.; 2019; 29, pp. 2295-2306. [DOI: https://dx.doi.org/10.1016/j.cub.2019.06.017]

130. Hojjati, S.H.; Ebrahimzadeh, A.; Khazaee, A.; Babajani-Feremi, A. Predicting conversion from MCI to AD by integrating rs-fMRI and structural MRI. Comput. Biol. Med.; 2018; 102, pp. 30-39. [DOI: https://dx.doi.org/10.1016/j.compbiomed.2018.09.004]

131. Amari, S. Information geometry of the EM and em algorithms for neural networks. Neural Netw.; 1995; 8, pp. 1379-1408. [DOI: https://dx.doi.org/10.1016/0893-6080(95)00003-8]

132. Anderson, J.R. Human symbol manipulation within an integrated cognitive architecture. Cogn. Sci.; 2005; 29, pp. 313-341. [DOI: https://dx.doi.org/10.1207/s15516709cog0000_22]

133. Gazzaniga, M.G.; Ivry, R.B.; Mangun, G.R. Cognitive Neuroscience: The Biology of the Mind; 5th ed. Norton: New York, NY, USA, 2019.

134. Miller, G.A. The cognitive revolution: A historical perspective. Trends Cogn. Sci.; 2003; 7, pp. 141-144. [DOI: https://dx.doi.org/10.1016/S1364-6613(03)00029-9]

135. Langley, P.; Laird, J.E.; Rogers, S. Cognitive architectures: Research issues and challenges. Cogn. Syst. Res.; 2009; 10, pp. 141-160. [DOI: https://dx.doi.org/10.1016/j.cogsys.2006.07.004]

136. Thorisson, K.R.; Helgasson, H.P. Cognitive architectures and autonomy: A comparative review. J. Artif. Gen. Intell.; 2012; 3, pp. 1-50. [DOI: https://dx.doi.org/10.2478/v10229-011-0015-3]

137. Cina, M.; Rad, A.B. Categorized review of drive simulators and driver behavior analysis focusing on ACT-R architecture in autonomous vehicles. Sustain. Energy Technol. Assess.; 2023; 56, 103044. [DOI: https://dx.doi.org/10.1016/j.seta.2023.103044]

138. Li, H.; Dou, R.; Keil, A.; Principe, J.C. A self-learning cognitive architecture exploiting causality from rewards. Neural Netw.; 2022; 150, pp. 274-292. [DOI: https://dx.doi.org/10.1016/j.neunet.2022.02.029] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35339009]

139. Itoh, H.; Nakano, H.; Tokushima, R.; Fukumoto, H.; Wakuya, H. A partially observable markov-decision-process-based blackboard architecture for cognitive agents in partially observable environments. IEEE Trans. Cogn. Dev. Syst.; 2022; 14, pp. 189-204. [DOI: https://dx.doi.org/10.1109/TCDS.2020.3034428]

140. Huang, K.; Ma, X.; Song, R.; Rong, X.; Tian, X.; Li, Y. A self-organizing developmental cognitive architecture with interactive reinforcement learning. Neurocomputing; 2020; 377, pp. 269-285. [DOI: https://dx.doi.org/10.1016/j.neucom.2019.07.109]

141. Zall, R.; Kangavari, M.R. Comparative analytical survey on cognitive agents with emotional intelligence. Cogn. Comput.; 2022; 14, pp. 1223-1246. [DOI: https://dx.doi.org/10.1007/s12559-022-10007-5]

142. Ye, P.; Wang, T.; Wang, F. A survey of cognitive architectures in the past 20 years. IEEE Trans. Cybern.; 2018; 48, pp. 3280-3290. [DOI: https://dx.doi.org/10.1109/TCYB.2018.2857704]

143. Julier, S.; Uhlmann, J.; Durrant-Whyte, H. A new method for the nonlinear transformation of means and covariances in filters and estimators. IEEE Trans. Autom. Control; 2000; 45, pp. 477-482. [DOI: https://dx.doi.org/10.1109/9.847726]

144. Arasaratnam, I.; Haykin, S. Cubature kalman filters. IEEE Trans. Autom. Control; 2009; 54, pp. 1254-1269. [DOI: https://dx.doi.org/10.1109/TAC.2009.2019800]

145. Ristic, B.; Arulampalam, S.; Gordon, N. Beyond the Kalman Filter: Particle Filters for Tracking Applications; Artech House: Reading, PA, USA, 2004.

146. Gordon, N.; Salmond, D.; Smith, A. Novel approach to nonlinear/non-Gaussian Bayesian state estimation. Inst. Electr. Eng. Proc. F-Radar Signal Process.; 1993; 140, pp. 107-113. [DOI: https://dx.doi.org/10.1049/ip-f-2.1993.0015]

147. Lin, H.; Si, J.; Abousleman, G.P. Orthogonal rotation-invariant moments for digital image processing. IEEE Trans. Image Process.; 2008; 17, pp. 272-282.

148. Alnashwan, R.O.; Chelloug, S.A.; Almalki, N.S.; Issaoui, I.; Motwakel, A.; Sayed, A. Lighting search algorithm with convolutional neural network-based image captioning system for natural language processing. IEEE Access; 2023; 11, pp. 142643-142651. [DOI: https://dx.doi.org/10.1109/ACCESS.2023.3342703]

149. Zhu, Q.; Wu, X.; Lin, Q.; Ma, L.; Li, J.; Ming, Z. A survey on evolutionary reinforcement learning algorithms. Neurocomputing; 2023; 556, 126628. [DOI: https://dx.doi.org/10.1016/j.neucom.2023.126628]

150. Schifani, C.; Sukhanov, I.; Dorofeikova, M.; Bespalov, A. Novel reinforcement learning paradigm based on response patterning under interval schedules of reinforcement. Behav. Brain Res.; 2017; 331, pp. 276-281. [DOI: https://dx.doi.org/10.1016/j.bbr.2017.04.043] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28457882]

151. Radulescu, A.; Niv, Y.; Ballard, I. Holistic reinforcement learning: The Role of structure and attention. Trends Cogn. Sci.; 2019; 23, pp. 278-292. [DOI: https://dx.doi.org/10.1016/j.tics.2019.01.010]

152. Wang, X.; Nie, G.; Li, B.; Zhao, Y.; Kang, M.; Liu, B. Hierarchical memory-guided long-term tracking with meta transformer inquiry network. Knowl. Based Syst.; 2023; 269, 110504. [DOI: https://dx.doi.org/10.1016/j.knosys.2023.110504]

153. Lee, T.; Shin, J.; Kim, J.; Singh, V.P. Stochastic simulation on reproducing long-term memory of hydroclimatological variables using deep learning model. J. Hydrol.; 2020; 582, 1245540. [DOI: https://dx.doi.org/10.1016/j.jhydrol.2019.124540]

154. Haykin, S.; Fuster, J.M. On cognitive dynamic systems: Cognitive neuroscience and engineering learning from each other. Proc. IEEE; 2014; 102, pp. 608-628. [DOI: https://dx.doi.org/10.1109/JPROC.2014.2311211]

155. Maidan, I.; Jacob, Y.; Giladi, N.; Hausdorff, J.M.; Mirelman, A. Altered organization of the dorsal attention network is associated with freezing of gait in Parkinson’s disease. Park. Relat. Disord.; 2019; 63, pp. 77-82. [DOI: https://dx.doi.org/10.1016/j.parkreldis.2019.02.036]

156. Roberts, B.Z.; Young, J.W. Translational cognitive systems: Focus on attention. Emerg. Top. Life Sci.; 2022; 6, pp. 529-539.

157. Carlson, K.S.; Gadziola, M.A.; Dauster, E.S.; Wesson, D.W. Selective attention controls olfactory decisions and the neural encoding of ddors. Curr. Biol.; 2018; 28, pp. 2195-2205. [DOI: https://dx.doi.org/10.1016/j.cub.2018.05.011] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30056854]

158. Song, R.; Tao, G.; Guo, F.; Ma, H.; Zhang, J.; Wang, Y. The change of attention network functions and physiological adaptation during high-altitude hypoxia and reoxygenation. Physiol. Behav.; 2023; 268, 114240. [DOI: https://dx.doi.org/10.1016/j.physbeh.2023.114240] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37201691]

159. Yin, X.; Liu, Z. Cognitive control from the automation’s perspective. Proceedings of the 42nd Chinese Control Conference; Tianjin, China, 24–26 July 2023.

Word count: 17678

Show less

© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.

Abstract

Translate

The operational environments of engineering systems are becoming increasingly complex and require automatic control systems to be more intelligent. Cognitive control extends the domain of intelligent control, whereby cognitive science theories are applied to guide the design of automatic control systems to make them conform to the human cognition paradigm and behave like a real person, hence improving physical systems performance. Cognitive control has been investigated in several fields, but a comprehensive review covering all these fields has yet to be provided in any paper. This paper first presents a review of cognitive control development and related works. Then, the relationship between cognitive control and cognitive science is analyzed, based on which the definition and framework of cognitive control are summarized from the perspective of automation and control. Cognitive control is then compared with similar concepts, such as cognitive radio and cognitive radar, and similar control methods, such as intelligent control, robust control, and adaptive control. Finally, the main issues, research directions, and development prospects are discussed. We expect that this paper will contribute to the development of cognitive control.

Details

Title

A Review of Cognitive Control: Advancement, Definition, Framework, and Prospect

Author

Liu, Zhenfei

; Yin, Xunhe

First page

Publication year

2025

Publication date

2025

Publisher

MDPI AG

ISSN

20760825

Source type

Scholarly Journal

Language of publication

English

DOI

https://doi.org/10.3390/act14010032

ProQuest document ID

3159142547