MuJoCo Playground

• DOI: 10.48550/arxiv.2502.08844
• Publication Date: 2025-02-12
• Authors: Kevin Zakka; Baruch Tabanpour; Qiayuan Liao; Mustafa Haiderbhai; Samuel Holt; Jing Yuan Luo; Arthur Allshire; Erik Frey; Koushil Sreenath; Lüder A. Kahrs; Carmelo Sferrazza; Yuval Tassa; Pieter Abbeel
• Keywords: FOS: Computer and information sciences; Computer Science - Robotics; Robotics (cs.RO)

We introduce MuJoCo Playground, a fully open-source framework for robot learning built with MJX, the express goal of streamlining simulation, training, and sim-to-real transfer onto robots. With simple "pip install playground", researchers can train policies in minutes on single GPU. Playground supports diverse robotic platforms, including quadrupeds, humanoids, dexterous hands, arms, enabling zero-shot from both state pixel inputs. This is achieved through an integrated stack comprising physics engine, batch renderer, training environments. Along video results, entire freely available at playground.mujoco.org