ml-agents

Why we included this project

Unity ML-Agents turns the Unity engine into a training environment for intelligent agents. You define agents and reward signals directly in a scene, train them with PyTorch-based algorithms like PPO or self-play, then drop the trained model back into the running game or simulation. That loop makes it practical for controlling NPC behavior, stress-testing builds, and checking design decisions before release. Researchers get a clean Python API for reinforcement learning, imitation learning, and custom training methods, while game developers get a Unity SDK that fits into existing projects. If you want to prototype embodied or multi-agent behavior in a rich visual world instead of a bare code-only simulator, this is a solid starting point.