sim-environment-select
Use when choosing or setting up a simulator for a robot task — match the engine to the job (MuJoCo for contact-rich manipulation, Isaac Lab for large-scale RL, Gazebo for ROS navigation) and reuse the
Files1
View on GitHubSKILL.mdproject-trajectory/skills/sim-environment-select/SKILL.md
GitHubWhen to use. Standing up or picking a sim. Why: engines differ in contact fidelity, throughput, and ROS integration; the wrong pick wastes weeks and misleads sim-to-real.
Procedure.
- Classify the task → engine: manipulation/sim-to-real → MuJoCo/MJX; parallel RL locomotion → Isaac Lab; ROS nav/SLAM/multi-robot → Gazebo.
- Load the same URDF/description used on hardware (converge, don't fork).
- For high stakes, plan dual-sim validation (train Isaac Lab → re-score MuJoCo MJX).
- Done when: the robot loads and a scripted motion runs in-sim matching the real kinematics.
Knowledge: KNOWLEDGE-LIBRARY.md §B3. Example: the MJCF snippet there.