DeepMind’s robotic ballet: An AI for coordinating manufacturing robots (arstechnica.com)

DeepMind announced RoboBallet, an AI system that jointly automates task allocation, scheduling, and motion planning for teams of industrial robotic arms — replacing the manual, hours‑long choreography normally required to program manufacturing cells. The group framed the problem as a compounded combinatorial challenge (likened to a “traveling salesman problem on steroids”) where the system must decide which robot does which task in what order while ensuring collision‑free, feasible motions in cluttered workspaces. The team trained and tested RoboBallet in simulated “work cells” containing a workpiece and up to eight Franka Panda 7‑DOF arms tasked with completing as many as 40 subtasks. Each subtask required an end effector to reach within 2.5 cm of a precise location at the correct approach angle and hold position to simulate performing work. By sampling these realistic scenarios and solving allocation, scheduling, and motion constraints together, RoboBallet demonstrates a path to dramatically reduce setup time, improve throughput, and increase flexibility in factory automation — enabling more adaptive multi‑robot coordination without hand‑coding every motion.