Pick a bay along the top. Each one is a different simulation, from tag and soccer to walking creatures and a predator/prey habitat.
Press Start training. A background thread plays thousands of practice rounds at full speed, while the arena shows a live match using the newest version of the brain.
Watch the reward curve go up. PPO improves one neural network with gradient updates. Neuroevolution breeds a population of networks and keeps the best players.
Tinker. In Rewards, change what the agents are paid for. In Arena, add walls and change team sizes or physics. In Brain, look inside an agent's network as it decides.
Step in. Choose Take control, click a robot, and play against what you trained using WASD.
Brains save in your browser automatically, and you can export them as files. Nothing is uploaded.