lab

Why we included this project

DeepMind Lab is a familiar stop for deep reinforcement learning researchers who want to test agents on navigation and puzzle-solving in a first-person 3D world. It runs on the Quake III engine, ships with a set of built-in levels, and exposes a Lua scripting API, so you can either train on the stock tasks or build custom environments that isolate skills like spatial memory and reward shaping. A Python API and hooks into DeepMind's dm_env interface make it easy to connect your own agent and start collecting data. It does build from source on Linux with Bazel, so it suits people who want a serious research testbed more than a plug-and-play demo.