Creating a MuJoCo model has always meant writing XML by hand. You specify every body, joint, actuator, and collision geometry in MJCF format, tweaking inertias and damping coefficients until the simulation behaves correctly. It's slow, error-prone, and requires deep familiarity with physics simulation internals.
RebelAI changes this. Describe the robot you want—"a 6-DOF arm with a parallel gripper"—and get a complete MuJoCo model ready for simulation. The generated model includes proper mass distributions, joint limits, actuator gains, and collision geometries. You can start training policies immediately.
From text to MJCF
The generation pipeline interprets your natural language description and produces valid MJCF. It understands common robot morphologies: manipulator arms, mobile bases, legged systems, grippers. When you say "quadruped," it knows you mean four legs with appropriate joint configurations.
The system infers reasonable physics parameters based on the robot type and scale. A tabletop manipulator gets different inertias than an industrial arm. Joint damping is set to produce stable control. Actuator limits match what's physically plausible for the specified configuration.
Iterating through prompts
The real power is iteration speed. Traditional MuJoCo development means editing XML, reloading, testing, and repeating. With text-based generation, you modify your description and regenerate. Want to add a wrist camera? Update the prompt. Need softer joint limits? Describe that constraint.
This speed matters when you're exploring robot designs or task configurations. You can test dozens of variations in the time it would take to manually author one model.
Integration with World Labs
Generated robots need environments to operate in. RebelAI connects to World Labs for photorealistic scene generation. The same text-based workflow applies: describe the scene, generate the environment, and your MuJoCo robot spawns into a visually rich world ready for training perception and control policies together.
Getting started
Install the package and generate your first model:
pip install rebelaiA single function call takes your description and returns a simulation-ready MuJoCo model. The documentation covers the full prompt syntax, customization options, and examples for common robot types.