urlab-sim/unrealroboticslab

URLab (Unreal Robotics Lab) is an open-source plugin for Unreal Engine 5 that lets you run accurate robot simulations inside a visually realistic environment. It does this by embedding MuJoCo, a physics engine from Google DeepMind known for precise contact and movement simulation, directly into Unreal Engine. The result is that you get Unreal Engine's cinematic-quality visuals alongside physics accurate enough for training and testing real robot controllers. To use it, you drag a standard robot description file (in MuJoCo's XML format) into the Unreal Engine editor, and the plugin automatically sets up a complete simulated robot with joints, sensors, and actuators. It supports over 40 sensor types and 8 actuator types, meaning you can simulate essentially any robot configuration. For control, the plugin communicates over ZMQ networking, a messaging protocol, so you can send commands from Python or ROS 2 while the simulation renders in Unreal Engine. There is also a companion Python package called urlab_bridge for deploying reinforcement learning policies and bridging to ROS 2. Additional features include a record-and-replay system for capturing and reviewing robot trajectories, debug visualization tools for inspecting physics in real time, cinematic camera tools, and support for Unreal's Blueprint visual scripting system so no C++ code is required for many tasks. It runs on Windows and Linux and requires Unreal Engine 5.7 or newer.