R&D Engineer, ElasticRun
Why
A warehouse wants robots that pick, scan and sort parcels, and a sorter that moves them. Both are expensive to get wrong on the real floor. The work was to make them right in simulation first, and to check the simulator itself before anyone trusted it.
How
I built a two-armed robot platform end to end in simulation, from arm control to the vision that spots damaged parcels and reads their labels, and closed it into a pick, scan and sort loop. When one arm started shaking uncontrollably I traced it to a damping setting fighting the physics engine, and the shaking stopped. Separately, while the company evaluated a vendor proposal for sorter automation, I audited its production sorter simulator against its specification and found the places where it disagreed with itself.
What came out
A robot that stopped shaking, a vision layer that catches damaged goods before dispatch, and a simulator the company could trust in a vendor evaluation because its defects were found and fixed first.
What broke
The arm oscillation. It looked like a control bug for days and was actually the damping term and the physics solver pulling against each other at every step. The fix was small; finding it was the work.
Mind the gap
Internship. The code belongs to the company, so this station is described and not shown.