What happens when artificial intelligence takes on the challenge of evolution? In this experiment, we train AI models to walk through different stages of life — starting from insect-like crawlers, through jumpers, dinosaurs, quadrupeds, and finally, upright humans. This isn’t just a test of reinforcement learning — it’s AI vs NATURE. This video consisted of 4 stages with 4 different agents: Ant (Insect Stage) Hopper (Kangaroo Stage) Quadruped (Cheetah/Dog Stage) Humanoid (Human Stage) HOW I MADE THIS VIDEO: The AIs were trained completely by myself on my local machine. I used the process of Reinforcement Learning. These environments are from the PyBullet envs package for Python. I used a mix of special RL algorithms implemented by the team at Stable Baselines3 to train these agents. Here are the training steps took for each stage: Ant: 3M steps Hopper: less than 1M steps, around 500K Cheetah: 2M steps Humanoid: 8M steps, stopped because it didnt make any progress whatsoever from after 5M. In the end, I think it's safe to say that humans win the challenge, as AI could complete the other challenges, however it failed when it comes to the humanoid level. This video took a lot of work (and a lot of my free time) so the fact that you guys watched this video means a lot to me. Thank you for your contribution to getting these videos to more people :D See you in the next video. COURTESY: Reuters (for the real-life robot walking clip near the end of the video)