Humanoid Gait Imitation with Reinforcement Learning

A reinforcement-learning framework in MuJoCo that teaches a simulated humanoid to walk by imitating human motion-capture data.

MuJoCo
Reinforcement Learning
Imitation Learning
OpenSim

← All projects Get in touch

Overview

Instead of hand-designing a walking controller, the humanoid learns to walk by imitating reference motion from real people. The reference comes from motion-capture recordings, converted into joint trajectories with OpenSim inverse kinematics.

What I built

A MuJoCo-based RL training framework for humanoid locomotion.
A pipeline that turns motion-capture data into reference joint trajectories using OpenSim inverse kinematics.
Reward shaping and training pipelines that balance gait stability, motion-tracking accuracy, and physically plausible movement.

More work

Other projects

Safe RL for Quadrotor Recovery

PPO policy with a Control Barrier Function safety filter that recovers a quadrotor from fully inverted states, deployed on a real Agilicious drone.

Foundation-Model Driving on F1TENTH

ROS 2 autonomous-driving stack for an F1TENTH car where vision-language and large language models make high-level driving decisions on top of classical control.

UAV Cave Exploration

ROS 2 quadrotor that autonomously explores and maps an unknown cave with OctoMap, frontier-based exploration, sampling-based planning and a geometric SE(3) controller.