Robots that touch, grasp, and assemble.
Most robots can move; few can manipulate. Work in this area studies contact-rich control: the tactile sensing, force control, and dexterous, often bimanual coordination needed to handle real objects and finish physical tasks to precision and tolerance.
Watch the explainer
The idea behind this lab in ninety seconds — then come back and drive it.
How the work happens.
The methods behind the research.
Beyond pick-and-place
Policies for insertion, assembly, and tool use, tasks defined by forces and tolerances, not just positions.
Feeling, not just seeing
Touch as a first-class sensor: slip detection, contact geometry, and force feedback closing the loop where vision can't reach.
Two hands, one task
Coordinated control for what a single manipulator can't do alone: bracing, reorienting, and handing off.
Hands worth controlling
Co-designing grippers and actuation with the policies that drive them, so capability isn't capped by the mechanism.
Open problems we're pursuing.
What's being pursued now.
Force-controlled assembly
Policies that hit tight tolerances under uncertainty. This is the gap between a demo and a production cell.
Tactile-driven policy learning
How much does touch add over vision alone, and how do you train with it at scale?
Manipulation sim-to-real
Contact dynamics are the hardest thing to simulate faithfully. How far does randomization actually carry?
Insertion by feel, not by aim.
The contact-rich task in one demo: a peg-in-hole where stiff position control jams on the smallest misalignment, while force control feels the chamfer and slides in. Manipulation is defined by forces and tolerances, not positions.
Open the full view ↗Drag the misalignment past the clearance and switch controllers: position jams, force-control self-aligns down the chamfer. An illustrative contact model.
Work in this area.
Open positions, including the Physical AI Investigator Program, are listed on Careers. Research here can also spin out into a company.