Sim-to-Real Dexterous Hand

Simulation infrastructure for a multi-fingered robotic palm in Webots and Isaac Sim.

In this project, I established robust simulation environments to support future Reinforcement Learning (RL) research on a multi-fingered robotic palm.

Simulation Demos

Isaac Sim
Closed-loop simulation of the robotic palm in NVIDIA Isaac Sim.
Webots
Closed-loop simulation of the robotic palm in Webots.

Work Done

  • URDF Refinement: Optimized the URDF model’s collision geometries and kinematic constraints to ensure physical accuracy.
  • Environment Setup: Configured physics parameters (e.g., friction, joint limits) in both Isaac Sim and Webots.
  • Foundational Pipeline: Established a simulation-to-training pipeline for future RL grasping policies.