Gentle Object Retraction in Dense Clutter Using Multimodal Force Sensing and Imitation Learning
🤖 AI Summary
Safely inserting and grasping target objects amidst densely packed, movable objects remains challenging for robotic manipulation.
Method: Inspired by human non-grasping haptic perception, this paper proposes a gentle manipulation framework integrating multimodal force sensing and imitation learning. It systematically fuses non-grasping triaxial tactile sensing, contact wrench estimated from joint torque, eye-in-hand vision, proprioception, and suction cup status monitoring to establish a high-fidelity contact feedback loop. A policy network trained on this rich sensory input enables fine-grained contact force regulation.
Results: All force-aware variants significantly reduce excessive-force failures, improve task success rates, and shorten completion time versus the force-agnostic baseline. The fusion of tactile and wrench information achieves optimal performance, boosting retraction success rate by 80% over the baseline.
Application Category
📝 Abstract
Dense collections of movable objects are common in everyday spaces -- from cabinets in a home to shelves in a warehouse. Safely retracting objects from such collections is difficult for robots, yet people do it easily, using non-prehensile tactile sensing on the sides and backs of their hands and arms. We investigate the role of such sensing for training robots to gently reach into constrained clutter and extract objects. The available sensing modalities are (1) "eye-in-hand" vision, (2) proprioception, (3) non-prehensile triaxial tactile sensing, (4) contact wrenches estimated from joint torques, and (5) a measure of successful object acquisition obtained by monitoring the vacuum line of a suction cup. We use imitation learning to train policies from a set of demonstrations on randomly generated scenes, then conduct an ablation study of wrench and tactile information. We evaluate each policy's performance across 40 unseen environment configurations. Policies employing any force sensing show fewer excessive force failures, an increased overall success rate, and faster completion times. The best performance is achieved using both tactile and wrench information, producing an 80% improvement above the baseline without force information.
Problem
Research questions and friction points this paper is trying to address.
Retracting objects from dense clutter using multimodal sensing
Training robots with imitation learning for gentle object extraction
Evaluating force sensing impact on success rates and performance
Innovation
Methods, ideas, or system contributions that make the work stand out.
Multimodal force sensing for object retraction
Imitation learning from human demonstrations
Combining tactile and wrench information
🔎 Similar Papers
No similar papers found.
