Robotics that adapts to your line.

Most robots in SMB manufacturing are programmed once and run one product. They handle pose variation badly, choke on part-family changes, and lose hours on changeover. The hardware is fine — what's missing is a software layer that lets a workcell adapt.

Asperity builds that layer. We use existing robot arms (Universal Robots, FANUC, ABB), pair them with cameras, force/torque, and tactile sensing where it matters, and run perception, anomaly detection, task adaptation, and recovery on top. The result is a cell that handles variation instead of breaking on it.

Operators will tell you the truth about what's breaking.

Year-1 focus: East Texas fabrication shops within driving distance of Tyler, and specialty electronics contract manufacturers nationally. Two reasons. The work in these places is high-mix and existing automation handles it badly.

Before we deploy on customer floors, we train in simulation. The clip below shows four manipulation policies — peg insertion, threaded coupling on an M16 bolt, and gear assembly with two different end-effectors on the same arm — trained from scratch in NVIDIA Isaac Lab.

Same training pipeline. Different robots. Different end-effectors. The point isn't that any one task is novel — it's that the underlying approach is robot- and gripper-agnostic, and it adapts to whatever task you put in front of it. That is the software layer the Sprint delivers on a customer floor.

Most automation feasibility studies are slide decks. The Sprint is a working cell.

We bring or configure a leased robot at your facility (or at our lab with your parts), instrument it for data collection, build a baseline automation, then add the AI layer that handles the variation your operators see every day.

Measured before/after numbers and a production-pilot proposal grounded in what we observed — not vendor optimism.