Physical AI Lab Three static Blender poses separate target reach, rigid tool-body clearance and projected working footprint. Moving the grinder axis changes clearance without proving a cleaner process.
Four static Blender states distinguish an absent screw, a reference screw, a longer screw and a tilted seat pose. Each result answers a different geometric question.
The outer screwdriver housing can stay unchanged while the loaded screw moves from a contained transfer state to exposed working reach. We checked both static states in Blender.
The generic interface kept its pilot clearance and mounting-hole pattern at 0 and 180 degrees, but the half-turn exchanged the measured aperture pairings.
Two generic 18 mm slides produced 15.588 mm and 9.000 mm of normal approach, showing why compliance direction matters before force tuning.
The same grasp and final pose produced different intermediate clearances: turning low obstructed the bin, while lifting first cleared every sampled pose.
Two Blender slider rows show why a half-speed reference can be 60 mm behind an unscaled schedule without that difference being tracking error.
Four measured Blender slider cases separate sampled path error from endpoint goal error and show why a later endpoint cannot rescue an earlier path-tolerance abort.
An Isaac Lab demonstration can load and step while state agreement or task success remains unresolved. Use this evidence ladder to diagnose replay results.
MoveIt parallel planning needs a declared selection criterion. See why early stopping, shortest joint-space paths and finite obstacle clearance answer different questions.