Grasp, Slide, Roll: Comparative Analysis of Contact Modes for Tactile-Based Shape Reconstruction

TL;DR

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%.

Abstract

Tactile sensing allows robots to gather detailed geometric 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 combined 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.

Figures (11)

• Figure 1: Tactile information is promising to unlock contact-rich manipulation capabilities for robots. We investigate tactile data acquisition efficiency of various contact modes (sliding, rolling, and discrete contacts) in the context of 3D object reconstruction using tactile data only (no vision).
• Figure 2: Comparative examples of original SPVD results and our improved approach. Partial input points and shape completed output points are colored blue and red, respectively. (a) Shape completion with (right) and without (left) normal vectors (shown as black arrows) illustrates how normal guidance disambiguate shape generation direction. (b) Novel point generation filters noisy measurements (right), while the original approach preserves noisy inputs (left). (c) Rotation-invariant shape completion achieved through random rotation augmentation, demonstrated with a chair point cloud.
• Figure 3: Data augmentation pipeline on a simulated tactile point cloud including contact point truncation, Gaussian noise injection, and random rotation, illustrated using contact data from a sphere. Points are colored to show correspondence before and after applying the augmentation.
• Figure 4: Contact motion sequence of the tactile contact simulator. The target object is a sphere shaded in purple.
• Figure 5: Process diagram of the information-theoretic tactile exploration. Measured tactile points $P_{meas}$ are colored blue (Step 1), while the shape completed point cloud $P_{pred}$ is colored red (Step 2). The coordinate frames in Step 3 represent candidate contact poses, while gray coordinate frames in Step 4 represents filtered candidates. The contact pose with the highest score is selected (Step 5) and executed (Step 6).