In progress · Phase 1 closed, Phase 2 bench-proven in Isaac Sim 6.1 · updated 29 Sep 2026
A wafer cell, and its shadow
A two-robot semiconductor wafer-handling cell at ASU's Robotics Systems Instructional Lab, and the simulation that mirrors it. I work the simulation track: the Gazebo and Isaac Sim scenes, both joint mappings, and the read-only bridge. Phase 1 closed on 17 September 2026, when the mirror first ran live in Gazebo against both physical arms on the bench; Phase 2 was bench-proven on 28 September 2026, the same two arms live in NVIDIA Isaac Sim 6.1.
- One generated description for the whole cell - both arms and the conveyor axis, 11 joints under a single controller manager - running the full pick-transfer-place wafer cycle autonomously in simulation, on ROS 2 Jazzy and Gazebo Harmonic.
- A read-only real-to-sim digital shadow: the M1 Pro's 1440-byte feedback packet decoded at 123 Hz beside the session already commanding it, and republished as a simulator-agnostic JointState interface at 50 Hz. The laptop writes nothing back to either robot.
- Both joint mappings solved; the M1 Pro's verified to 0.1 degrees against every taught waypoint. On the 17 September bench run it tracked within 1.9 degrees / 7.6 mm and the Pro 600 within 4.4 degrees over a full cycle - figures noted at the bench that day.
- One-way, real to sim: a digital shadow rather than a bidirectional twin. Phase 2, bench-proven on 28 September 2026: the same two arms mirrored live in NVIDIA Isaac Sim 6.1, tracking within 0.5 degrees (M1 Pro) and 2.2 degrees (Pro 600) over a 461 s recorded cycle. The physical cell build and the PLC program - including the taught poses the arms run here - are teammates' work.
Live real-to-sim in NVIDIA Isaac Sim 6.1, the current milestone
Live real-to-sim in Gazebo, Phase 1
The full simulated cycle, for context
