Multi-Arm Robot Table Tennis

Research project (with Rohan Paleja, Purdue SCALE Robotics Lab) investigating whether multiple low-cost robot arms can collaboratively perform a high-speed task like table tennis, without relying on expensive sensors or hardware.

The system perceives the ball with two high-speed cameras running at 200Hz, localizes it via motion segmentation and epipolar geometry, and predicts its flight trajectory with a polynomial fit under an air-drag model. Multi-robot trajectories are generated with collocation MPC, and task allocation is learned by a small neural network that assigns the strike to whichever arm is best positioned while the idle arm avoids the active one.

Presented as a poster at Purdue CS.