RBQ Simulator

The RBQ open-source package ships several simulators that fall into two groups with different purposes:

• Open-source physics simulators — MuJoCo and Gazebo run the full RBQ software stack (Motion / Vision / GUI) against a simulation, so you can develop and test robot software without hardware.
• Reinforcement-learning training environments — Isaac Gym and Isaac Lab let you train a locomotion policy and deploy it to the real robot.

Simulators

Simulator	Engine	Package path	Purpose
MuJoCo	MuJoCo	rbq_simulator/rbq_mujoco/	Run the full RBQ stack in simulation
Gazebo	Ignition Gazebo (Fortress)	rbq_simulator/rbq_gazebo/	Run the full RBQ stack in simulation, with a ROS 2 sensor bridge
Isaac Gym	IsaacGym (Preview 4)	rbq_simulator/rbq_gym/	RL training — fast parallel training
Isaac Lab	Isaac Sim / Isaac Lab	rbq_simulator/rbq_lab/	RL training — high-fidelity simulation

Both full-stack simulators support two robot variants — the RBQ quadruped (default) and the wheeled RBQ (--wheel, 12 leg joints + 4 driven wheels). Pass the same variant to the RBQ stack with scripts/sim.bash -r wheel so Motion drives the wheels too.

LiDAR is opt-in in both, with the same flags — --livox (Livox Mid-360 x2, front / rear) or --ouster (Ouster OS1 x1, rear deck). With no flag the robot carries no LiDAR.

Running the full RBQ stack (MuJoCo / Gazebo)

MuJoCo and Gazebo join the running RBQ stack as DDS peers on domain 0 / loopback. Start the stack, then the simulator. scripts/sim.bash runs the stack in Docker; the first run builds its image, which takes a few minutes:

Quadruped

bash scripts/sim.bash --vision --no-sim
cd rbq_simulator/rbq_mujoco && ./bin/rbq_mujoco   # or rbq_gazebo's gazebo.bash

Wheeled RBQ

bash scripts/sim.bash --vision --no-sim -r wheel
cd rbq_simulator/rbq_mujoco && ./bin/rbq_mujoco --wheel   # or rbq_gazebo's gazebo.bash --wheel

See MuJoCo and Gazebo for the full build and run instructions.

Training & deploying a policy (Isaac Gym / Isaac Lab)

The RL environments follow the same Train → Play → Deploy pipeline:

Train (Isaac Gym or Isaac Lab)
  ↓
Play  (verify policy in same simulator)
  ↓
Sim2Sim (MuJoCo, via rbq_low_level)
  ↓
Deploy (real robot)

After training, the policy.onnx and info.json artifacts are loaded by rbq_low_level for on-robot execution.

Repository structure

All simulator code lives under rbq_simulator/:

RBQ/
└── rbq_simulator/
    ├── rbq_mujoco/        # MuJoCo simulator (full-stack)
    ├── rbq_gazebo/        # Ignition Gazebo simulator (full-stack + ROS 2 bridge)
    ├── rbq_gym/           # Isaac Gym RL environment
    └── rbq_lab/           # Isaac Lab RL environment

Next steps

• MuJoCo → — run the full stack in MuJoCo
• Gazebo → — run the full stack in Gazebo
• Isaac Gym → — Isaac Gym training guide
• Isaac Lab → — Isaac Lab training guide
• Low Level (RL Policy) → — deploy a trained policy to the real robot