Content area
Robotic object placement constitutes a critical component of pick-and-place operation. While significant advancements have been made in grasp planning and object acquisition strategies, the subsequent phase of safely depositing objects - particularly those with thin profiles or complex material properties - remains an open research problem. This dissertation bridges this gap by introducing model-based manipulation frameworks tailored for the precise placement of thin-rigid and deformable linear objects (DLOs), addressing both theoretical and practical challenges in robotic manipulation.
First, we present a closed-loop system designed for handling thin-rigid objects, which are prone to damage during placement due to their mechanical brittleness. The system integrates vision-based tactile sensors with pixel-level resolution and soft-contact capabilities, enabling real-time perception of interaction dynamics. A motion controller processes this sensory feedback to execute optimized in-hand rotation and sliding maneuvers, ensuring stable and damagefree placement on target surfaces. Second, we develop an open-loop framework for deformable linear objects, where the primary challenge lies in controlling infinite-degree-of-freedom systems with limited actuation points. Our method analytically relates object deformation to boundary constraint forces, enabling global shape control via a single grasp point. This approach achieves curvature matching with target surfaces without requiring full-state feedback.
The proposed techniques are object-agnostic and dimensionally scalable. Real-world experimental validation confirms their efficacy and robustness across industrial and domestic applications, diverse object classes, and varying environmental conditions.