Event

Four-Engine Simulation Comparison: Consistent Success/Failure but Contact Forces Differ Tenfold

1 reports · 1 sources

Manda Robotics released a simulator comparison report in early October, running experiments with the same Franka seven-axis robotic arm and the same control program on four physics engines: PhysX, MuJoCo, MuJoCo Warp, and Genesis. The results showed that task success/failure judgments were consistent, but values involving force and contact differed by up to tenfold. Harry wrote an interpretation of the report and, combined with his own SO-101 simulation experience, offered selection recommendations. The report's scope is limited to engine-to-engine comparison; it did not use a real robot as a reference. Harry recommends that any conclusion related to force and contact be re-run on a different engine; if the task involves soft objects such as cloth, sponge, food, or cables, it needs to be verified separately in each engine used.