Closing the Reality Gap: Zero-Shot Sim-to-Real Deployment for Dexterous Force-Based Grasping and Manipulation

TL;DR

The paper tackles the reality gap in dexterous manipulation by presenting a zero-shot sim-to-real reinforcement learning framework that relies on dense tactile feedback and a current-based torque proxy. It introduces a computationally efficient tactile simulator, a per-joint current-to-torque calibration to eliminate torque sensors, and broad actuator dynamics randomization to bridge the sim-to-real divide. Trained entirely in simulation with an asymmetric actor-critic PPO, the policies are deployed directly on a real $5$-finger, $12$-DoF hand to achieve force-controllable grasping and in-hand object rotation, two force-sensitive manipulation tasks. The results demonstrate robust transfer to real hardware, generalization to unseen objects, and provide a practical, reproducible recipe for training full-state tactile-torque policies in simulation for dexterous manipulation.

Abstract

Human-like dexterous hands with multiple fingers offer human-level manipulation capabilities, but training control policies that can directly deploy on real hardware remains difficult due to contact-rich physics and imperfect actuation. We close this gap with a practical sim-to-real reinforcement learning (RL) framework that utilizes dense tactile feedback combined with joint torque sensing to explicitly regulate physical interactions. To enable effective sim-to-real transfer, we introduce (i) a computationally fast tactile simulation that computes distances between dense virtual tactile units and the object via parallel forward kinematics, providing high-rate, high-resolution touch signals needed by RL; (ii) a current-to-torque calibration that eliminates the need for torque sensors on dexterous hands by mapping motor current to joint torque; and (iii) actuator dynamics modeling to bridge the actuation gaps with randomization of non-ideal effects such as backlash, torque-speed saturation. Using an asymmetric actor-critic PPO pipeline trained entirely in simulation, our policies deploy directly to a five-finger hand. The resulting policies demonstrated two essential skills: (1) command-based, controllable grasp force tracking, and (2) reorientation of objects in the hand, both of which were robustly executed without fine-tuning on the robot. By combining tactile and torque in the observation space with effective sensing/actuation modeling, our system provides a practical solution to achieve reliable dexterous manipulation. To our knowledge, this is the first demonstration of controllable grasping on a multi-finger dexterous hand trained entirely in simulation and transferred zero-shot on real hardware.

Figures (9)

• Figure 1: A framework for learning a full-state policy integrating tactile sensing and joint torques for dexterous grasping and in-hand manipulation.
• Figure 2: Calibration and alignment of current-force (real robot) versus torque-force (simulation) properties.
• Figure 3: Contact point modeling and material properties.
• Figure 4: Visualization of real-world and simulated contact data during an in-hand rotation task. The close alignment between the contact points (Top) and contact forces (Bottom) shows the high fidelity of our contact simulation.
• Figure 5: Joint torque and contact forces under controllable force commands with different reward settings.