AI Chicken learns to RUN

A developer showcases an AI locomotion project using Proximal Policy Optimization (PPO) instead of the NEAT evolutionary algorithm. The video documents training progressively complex bipedal structures, culminating in an AI-powered chicken model that learns to run at nearly 10 km/h. Key improvements include a graphical editor for building structures, tuning joint range of motion, and adding synchronization penalties to reward functions. The training process is shown iteratively, with the agent developing increasingly natural and fluid gaits over time.