We make it easy to deploy policies on different robots. Whether it is a quadruped robot, a wheeled quadruped, or even a humanoid robot. Unileg provides an easy interface to plug-and-play different robots.
UniLeg is a unified and modular ROS framework for accelerating legged locomotion, navigation, and manipulation research. We believe that a streamlined pipeline for sim-to-sim-to-real transfer is essential for ease of use and reproducibility in embodied AI research.
Low-level control is often safety-critical and requires a lot of engineering effort. When our policy does not work, we often spend a lot of time debugging without knowing where the issue stems from. UniLeg provides a streamlined pipeline for sim-to-sim-to-real transfer in legged locomotion research without the need of preparing different inference script. Our pipeline supports seamless sim-to-sim deployment in Mujoco and Gazebo, as well as sim-to-real deployment in popular commercial robots (Unitree and DeepRobotics). By doing so, we can easily focus on policy design and ensuring safety through sim-to-sim validation.
We make it easy to deploy policies on different robots. Whether it is a quadruped robot, a wheeled quadruped, or even a humanoid robot. Unileg provides an easy interface to plug-and-play different robots.
Our modular pipeline makes it easy to even combines two different robots. With UniLeg, we are able to combine a Go2 robot with a ViperX robot arm to perform dynamic loco-manipulation.