projects

2026

1. 

   In-Hand Cube Rotation via LeapHand

   Dexterous Manipulation 2026

   LeapHand Reproduction:

   Hardware: Completed the assembly and commissioning of the LeapHand dexterous hand hardware platform.

   Software: Developed a in-hand cuboid manipulation policy via Reinforcement Learning on IsaacLab.

2025

1. 

   Learning One-Foot Camel Spin on the TITA Wheel-Legged Biped Robot

   Model-Free RL 2025

   Intern at Direct Drive Tech

   Learned a robust one-foot spin maneuver for a wheel-legged biped (TITA Robot of Direct Drive Tech) via Reinforcement Learning.

   Built on Isaacgym for training and valuated through sim-to-sim transfer to Webots.

2. 

   Agile Drone Pursuit-Evasion via Self-Play Reinforcement Learning

   Self-Play RL 2025

   Spring at ZJU FAST LAB

   We propose a Reinforcement Learning (RL) framework that conducts large-scale self-play to obtain high-quality interaction data, while iteratively optimizing the competing policies under the Policy-Space Response Oracle (PSRO) algorithm.