Compliant In-Hand Rolling Manipulation Using Tactile Feedback

• Source: https://arxiv.org/abs/2603.04301
• Type: paper
• ArXiv ID: 2603.04301
• Date Ingested: 2026-04-05T20:00:00Z
• Tags: tactile-sensing, dexterous-manipulation, in-hand, allegro-hand

Key Contribution

Compliant in-hand rolling manipulation using tactile feedback on Allegro hand with Visiflex fingertips. Combines forward/inverse mechanics with a vision-tactile feedback controller for robust object rolling.

Summary

This paper presents a framework for compliant in-hand rolling manipulation — the ability to roll objects between fingers while maintaining stable grasp. The system operates on the Allegro Hand equipped with Visiflex tactile fingertips that provide rich contact information.

Technical Approach

• Forward mechanics model: Predicts object motion from finger commands, accounting for contact mechanics and rolling constraints
• Inverse mechanics model: Computes required finger motions to achieve desired object trajectories
• Vision-tactile feedback controller: Fuses visual object tracking with tactile contact sensing to close the loop in real-time
• Compliant control: Maintains appropriate contact forces during rolling to prevent slip or excessive deformation

Hardware Platform

• Allegro Hand: 4-finger, 16-DOF robotic hand widely used in dexterous manipulation research
• Visiflex fingertips: Soft, vision-based tactile sensors that deform on contact, providing high-resolution contact geometry and force distribution
• Sensor fusion: Combines proprioceptive (joint angles/torques), tactile (contact patches), and visual (object pose) feedback

Key Results

• Demonstrates stable rolling of various objects between fingertips
• Tactile feedback significantly improves robustness compared to position-only control
• Compliant behavior allows handling of objects with different shapes and materials
• Real-time performance suitable for interactive manipulation tasks

Significance

Advances the state of tactile-driven dexterous manipulation by showing that rolling — one of the most challenging in-hand manipulation primitives — can be achieved reliably with appropriate tactile feedback and compliant control. This is a building block toward general-purpose robotic hands that can manipulate objects as dexterously as human hands.