Grasp, Slide, Roll: Comparative Analysis of Contact Modes for Tactile-Based Shape Reconstruction
Chung Hee Kim, Shivani Kiran Kamtikar, Tye Brady, Taskin Padir, Joshua Migdal
AI summary
Problem
Acquiring useful tactile data for object reconstruction is often slow and inefficient, as naive grasp-and-release strategies yield limited information per interaction.
Approach
The authors compare three contact modes (grasp-releasing, finger-grazing, and palm-rolling) using a diffusion-based shape completion model and an information-theoretic framework to guide sampling toward high-uncertainty regions.
Key results
- 34% fewer physical interactions required for convergence
- 55% improvement in reconstruction accuracy over grasp-releasing
- Average of 8.4 interactions per object for accurate reconstruction
- Successful zero-shot sim-to-real transfer using a diffusion-based shape completion model
Why it matters
Enables robots to efficiently reconstruct unknown objects through touch, which is critical for manipulation in cluttered or visually occluded environments.
Abstract
Tactile sensing allows robots to gather detailed ge- ometric information about objects through physical interaction, complementing vision-based approaches. However, efficiently acquiring useful tactile data remains challenging due to the time-consuming nature of physical contact and the need to strategically choose contact locations that maximize information gain while minimizing physical interactions. This paper studies how different contact modes affect object shape reconstruction using a tactile-enabled dexterous gripper. We compare three contact interaction modes: grasp-releasing, sliding induced by finger-grazing, and palm-rolling. These contact modes are com- bined with an information-theoretic exploration framework that guides subsequent sampling locations using a shape completion model. Our results show that the improved tactile sensing efficiency of finger-grazing and palm-rolling translates into faster convergence in shape reconstruction, requiring 34% fewer physical interactions while improving reconstruction accuracy by 55%. We validate our approach using a UR5e robot arm equipped with an Inspire-Robots Dexterous Hand, showing robust performance across primitive object geometries.