ROBOTICS & PHYSICAL AI
MuJoCo batched CPU rollouts (MjBatch)
MuJoCo batched CPU rollouts (MjBatch) is an OpenEnv-compatible reinforcement learning environment in our Robotics & Physical AI catalog. An agent acts through the tools below, and a deterministic verifier scores every episode, so the reward can be used for GRPO or other RLVR post-training.
Task and verification
Run native MuJoCo 3.11.0 / MjBatch 0.1.0 CPU batches with per-replica friction and actuator controls. Evaluate Go1 velocity tracking with a released PPO actor, G1 backflip MPC, arm friction identification from encoder recordings, and throwing-arm co-design.
Tools and runtime
Physical state, rotation, flight, stable landing, contact and work gates verify every replica. G1 uses a bounded reference task at nominal friction 0.6. Simulation qualification; model training speed and hardware performance are unmeasured.
NATIVE MUJOCO / MJBATCH
Related Robotics & Physical AI gyms
- Autonomous driving (CARLA)
- ManiSkill articulation & contact
- LeRobot imitation & policy learning
- PX4 drone aerial navigation
- Dynamic balance & locomotion (Unitree G1 / Go1)
- Sim-to-real robotic calibration
- Real robot arm manipulation (Viam)
- Mobile robot indoor navigation (Gazebo)
- Multi-robot swarm coordination (Robotarium)
- Bimanual stationery manipulation
See how these environments are run privately on the custom RL environments page, and how their rewards are made hard to game in our verifier design guide.
Plan a training run
Check the environment’s availability, compatible models and compute in Studio. A catalog entry does not guarantee that hosted training is enabled for your workspace.