MuJoCo — Fairino FR5 & NACS Charging Simulation

Testing charging code on the physical robot for every change slows down iteration. I built this MuJoCo environment to inspect arm motion, camera views, and charge-port detection locally, then use those observations to prepare for hardware tests.

The recording shows the simulated Fairino FR5 charging approach alongside the wrist-camera detector input and depth visualization.

Assembled the Fairino FR5 from robot description and mesh assets, mounted it to the Kiwi charging platform, and integrated the custom flange, ZED camera, and NACS tool. Downloaded Tesla connector and inlet CAD provides a more representative charging interface than basic placeholder shapes.

The garage includes Tesla vehicle meshes and parking geometry so the charging approach can be inspected in the context of the car and surrounding space.

MuJoCo provides the articulated robot scene and camera rendering. Local robot-control interfaces route commands into the simulator, while ROS 2 camera data exposes synthetic RGB, depth, camera calibration, and transforms for perception experiments.

A trained NACS segmentation model was exercised on rendered camera frames. Viewing detection output next to the robot makes camera alignment, visibility, and approach behavior easier to diagnose.

Created a local testbed for iterating on robot motion and NACS perception without requiring the physical robot for every experiment. The September 7, 2026 recording captures the simulation and camera views together.

This is a development simulation: contact behavior, insertion tolerances, and perception across varied conditions still require validation. A successful simulated approach does not establish reliable physical insertion.