Robotic Disassembly and Grasping

Our research in robotic disassembly and grasping addresses the difficulty of manipulating objects, particularly in tasks involving non-rigid or non-regular components, which are common in manufacturing and recycling. A major focus is the use of multimodal sensing and Reinforcement Learning (RL) to handle the high positional and orientation uncertainties inherent in these tasks (missing reference).

We have successfully applied these principles to the challenging task of non-regular peg extraction from industrial trays. By leveraging compliant tactile sensing modules—like the BioIn-Tacto sensor—integrated with an RL framework, we enable the robot to adapt its grasp approach and extraction strategy in real-time, even when the initial object pose is unknown (missing reference). This work demonstrates a significant step toward developing robust, generalized policies for robotic manipulation in unstructured environments.

Furthermore, we investigated the use of kinesthetic and tactile information for data-driven analysis to enhance object classification and recognition during complex disassembly and grasping tasks (missing reference). This foundational work has laid the groundwork for using rich sensory data to improve the robotic understanding of object properties and facilitate more intelligent and adaptable manipulation strategies.