Physical AI Lab

Averaging wheel travel gives path length, not a straight world displacement. Follow the finite turn, preserve heading and check the small-angle limit.

Equal edge lengths can hide a reflected asset. Distinguish object scale from changed mesh coordinates, and check a proper rotation before transferring a robot pose.

Validate the same finite-mesh endpoint in the same frame. Compare separate joint perturbations with geometric derivatives before converting tip velocity to joint rates.

A central tensor, an origin tensor and a rotated tensor describe different references. Compare finite assigned-mass cuboids before transferring robot model values.

A mesh can remain in place while the render camera clips it away. Compare actual frames and finite depth bounds, including a slab with no original corners inside.

The same one-pixel correspondence change gives a larger, asymmetric depth interval farther away. Compare audited synthetic rays at two depths and baselines.

A rotating tool can keep its flange coordinates yet move its tip. Compare the same orientation with fixed flange position or a compensated fixed TCP point.

Constant sprocket rpm can preserve average chain travel while hiding a within-tooth speed cycle. Compare discrete pin geometry at matched mean transport.

A 2 mm thread pitch can mean 6 mm per revolution. Follow one connected start to distinguish pitch from lead and avoid a 3× ideal conversion-scale error.

Planet spin in the fixed frame differs from spin relative to a moving carrier. Check parent rotation and the missing second member speed in a planetary model.