A Comparative Analysis of Machine Learning-Based and Conventional Techniques for Real-Time Path Planning in Robotics

Abstract

Robot performance and efficiency are greatly affected by motion planning, which is an essential component of robotic control. This paper compares path planning algorithms, including traditional and machine learning-based approaches, for real-time obstacle avoidance and target tracking. The motion planning network (MPNet), a learning-based neural planner, is evaluated alongside several established algorithms: the safe artificial potential field (SAPF), standard artificial potential field (APF), vortex APF (VAPF), and the dynamic window approach (DWA). Simulation results indicate that MPNet outperforms conventional techniques across critical metrics, including path efficiency and collision avoidance. According to simulation data, MPNet outperforms conventional methods like collision avoidance and path efficiency in crucial areas. These findings demonstrate the respective benefits and drawbacks of each algorithm and the effectiveness of learning-based strategies like MPNet in resolving the challenges associated with real-time path planning in dynamic circumstances.