Task-Error Residual Learning for Real-Robot Five-Ball Juggling

2026-06-15

AI Summary

This work addresses the low sample efficiency and poor stability of residual learning in real-world robotic multi-ball juggling tasks by introducing a novel sample selection mechanism driven by directional task error supervision and a task error model. The proposed approach uniquely employs directional task error as a supervisory signal, integrating analytical priors to construct an efficient residual learning framework, and adopts a fixed Jacobian Newton update strategy. Experiments on a Barrett WAM manipulator demonstrate that the system achieves stable convergence from the second trial onward, with monotonically decreasing task error and no subsequent failures. The method successfully accomplishes three- to five-ball juggling—a highly dynamic task typically requiring years of human practice—thereby substantially improving both learning efficiency and robustness.

Abstract

For residual learning that refines existing behavior, sample efficiency depends on two things: how much information each rollout returns, and how efficiently the learner uses that information. Reinforcement learning's standard scalar reward carries far less information than the directional task error that defines the task. Random exploration further discards whatever information each rollout returns. Through residual learning with directional task-error supervision and a task error model that drives sample selection, we achieve stable three-, four-, and five-ball juggling on anthropomorphic Barrett WAM arms. Despite planning and controlling through a simple, idealized stack, the system converges from the second attempt. The first attempt drops, after which task error decreases monotonically without further failures. In comparison, five-ball juggling typically takes humans years of practice. We compare residual learners across two ternary axes, the directional information in the learning feedback and the commitment of the analytic prior, spanning Newton-style Jacobian updates, Composite Bayesian Optimization, and stochastic search methods. Both axes prove necessary: neither directional feedback nor an informative prior suffices alone, and the simplest method that combines them, a fixed-Jacobian Newton update, is the most reliable. The learned residual tolerates substantial prior misalignment and degraded joint tracking, affecting mainly convergence speed. The bottleneck for residual learning on real robots is therefore the information content of the supervision signal and how the learner uses it, not the accuracy of the surrounding stack. Video documentation of all experiments is available at https://kai-ploeger.com/residual-juggling.

Problem

• residual learning
• sample efficiency
• task error
• real-robot juggling
• directional feedback

Innovation

• task-error residual learning
• directional feedback
• sample-efficient robot learning
• Newton-style Jacobian update
• real-robot juggling