Physical Intelligence Lab, ZJU-UIUC | Advisor: Prof. Hua Chen
- Used UMI data to generate full-body motion sequences for G1 from wrist-mounted RGB images alone.
- Used Flow Matching to integrate motion prior information, enabling zero-shot motion generation.
- Used WBC and incorporated ResMimic as an additional module to improve hand precision.
Physical Intelligence Lab, ZJU-UIUC | Advisor: Prof. Hua Chen
- Built a two-layer loco-manipulation framework transferring manipulation skills to a bipedal platform.
- Trained a DiT Diffusion Policy on UMI data to predict 6D end-effector trajectories from RGB.
- Trained an RL whole-body controller in Isaac Lab to track SE(3) commands on the bipedal robot.
RoboTouch Lab, UIUC | Advisor: Prof. Wenzhen Yuan
- Realized stable pose selection via a tactile-feedback-driven learning-based planner.
- Enabled stability prediction of poses by training a model with data collected on the Taccel simulator.
AVIATE Center, UIUC | Advisor: Dr. Sheng Cheng, Prof. Naira Hovakimyan
- Developed minimum-snap trajectory with flatness for smooth transitions between taut and slack cable states.
- Improved trajectory fidelity to real maneuvers via bilevel optimization of release parameters.
- Validated control strategies through simulation testing on Gazebo.
Physical Intelligence Lab, ZJU-UIUC | Advisor: Prof. Hua Chen, Prof. Liangjing Yang
- Facilitated blind walking by implementing a teacher-student reinforcement learning framework.
- Enabled quadruped locomotion under unknown payloads with a proprioception-based payload estimator.
- Validated through MuJoCo simulations and sim-to-real deployment.
Academic Training SRTP, ZJU | Advisor: Prof. Jian Chen
- Simplified route planning for unmanned mining trucks, overcoming traditional method limitations.
- Developed a two-step strategy: front-end path planning and back-end trajectory optimization.
- Used Hard-Constraints and Minimum-Snap to ensure safe, dynamically feasible paths.
