Teaching a Quadruped Robot to Walk, Map, and Explore

MuJoCo • Reinforcement Learning • Walking Policy

Project Overview

I'm developing a quadruped robot to explore environments that may be unsafe for people. I trained a walking policy in MuJoCo using reinforcement learning, and the robot now walks forward in simulation. Next, I plan to add mapping and autonomous exploration.

Key Features

MuJoCo quadruped simulation
PPO walking policy trained with JAX
Gait evaluation using speed, stability, and foot-contact data

What I Learned

I started learning MuJoCo and reinforcement learning by configuring the simulation and writing reward and evaluation functions in Python. When a policy tracked speed but favored one diagonal pair, I learned to check foot contact and balance before treating the gait as reliable.