Learned Locomotion

Key Insight

Train a legged locomotion control policy using reinforcement learning in the GPU-accelerated Isaac Lab simulation environment. Instead of hand-crafting gait sequences or footstep planners, the policy learns to coordinate the robot's joints directly from joint angles and inertial measurements to match a target velocity. By applying domain randomization to physics parameters like friction, mass, and latencies, the learned policy develops robustness for successful sim-to-real transfer.