Sim-to-Real RL for Manipulation

Fast and Efficient Simulation Training with MuJoCo + JAX (30min)

Zero-shot Real World Execution

We are developing reinforcement learning (RL) policies for contact-rich robotic manipulation tasks in simulation and transferring them effectively to the real world. Contact-rich interactions—such as sliding, pivoting, or insertion are especially challenging due to their sensitivity to dynamics and high precision requirements. By leveraging physics-based simulators, we train robust control policies and explore strategies for bridging the sim-to-real gap.

Large Scale Simulation RL Training

Zero-shot Real World Execution

Our goal is to enable real-world manipulation systems to learn complex skills without requiring large-scale real-world data collection.