TUM & MBZUAI teach robot arm to juggle balls with optimal control

A team from the Technical University of Munich and MBZUAI Abu Dhabi has introduced an optimal control approach for non-prehensile ball juggling using a 7-degree-of-freedom (DoF) manipulator. The key challenge in robotic juggling is dynamically stabilizing an underactuated object—since the ball cannot self-correct, stability depends entirely on forces applied by the robot. Timing is critical to continuously counteract deviations. The researchers developed a two-stage optimal control framework: first, offline computation of feasible juggling trajectories, then a real-time error correction mechanism that avoids solving optimal control problems online. The approach was validated on a Franka Emika Panda robot arm in both simulation and physical experiments.

The method handles the hybrid dynamics of impact events and continuous motion, enabling high-speed periodic juggling. The framework generalizes to other non-prehensile tasks like balancing or sliding objects. Accepted at ICRA 2026, this work pushes robotic dexterity for applications requiring precise dynamic interactions, from industrial part handling to collaborative robotics.