Active Visuo-Tactile Interactive Robotic Perception for Accurate Object Pose Estimation in Dense Clutter

This work presents a novel active visuo-tactile based framework for robotic systems to accurately estimate pose of objects in dense cluttered environments. The scene representation is derived using declutter graph (DG) which describes the relationship among decluttering by leveraging semantic segmentation and grasp affordances networks. formulation allows robots efficiently workspace autonomously selecting next best object remove optimal action (prehensile or non-prehensile) perform. Furthermore, we propose translation-invariant Quaternion filter (TIQF) vision tactile estimation. Both visual points are selected maximizing expected information gain. We evaluate our proposed on system with two coordinating randomized scenes perform ablation studies static estimation prior post as baselines. Our interactive perception shows upto 36% improvement accuracy compared baseline.