In-Hand Manipulation with Enforced Grasp Stability for Contact-Rich Tasks

Published in IEEE ICRA 2025 Workshop on Contact-Rich Manipulation (CRM), Atlanta, GA, 2025

In-Hand Manipulation with Enforced Grasp Stability for Contact-Rich Tasks

  • Grasp Stability Objective: Introduced a reinforcement learning framework that incorporates fingertip force-torque sensing into the observation space to encourage stable manipulation.
  • Force-Closure Reward: Designed a physics-based reward using grasp wrench space analysis and singular value decomposition to promote contact-rich, secure grasps.
  • Policy Evaluation: Demonstrated improved robustness and early learning efficiency in a cube reorientation task with a downward-facing multi-fingered hand.
  • Reinforcement Learning: Utilized Soft Actor-Critic (SAC) with rich multi-modal feedback and parallelized simulation environments.
  • Video & Results: View Paper on OpenReview

Recommended citation: Y. Chen*, S. Lu*, H. Zhang, and K. Lynch. (2025). "In-Hand Manipulation with Enforced Grasp Stability for Contact-Rich Tasks." IEEE ICRA 2025 Workshop on Contact-Rich Manipulation (CRM), Atlanta, GA. [Online]. Available: https://openreview.net/forum?id=YK2YeuL2rU
Download Paper