Soft robots improve damage recovery with phase adaptation in nine trials
Nine Trials to Recover: A Reproducible Benchmark for Repertoire-Free Soft-Robot Damage Adaptation
Robotics
Summary
Soft robots can get hurt or damaged, which makes it hard for them to keep moving well. This paper presents a way to help soft robots quickly adjust and recover from damage without needing a large set of pre-made behaviors. The method uses smart testing combined with feedback to find good new control settings in just nine tries. When tested on many robot designs and different types of damage, this approach worked better than four other common methods. The results provide a reliable way to compare future techniques that help soft robots heal from damage on the fly.
What this means in practice
- •For soft robotics engineers: Improve robot controllers that quickly adapt actuator timing to regain functionality after physical damage without relying on pre-programmed behaviors.
- •For robotic control developers: Evaluate and compare damage recovery algorithms on diverse robot structures using a standard benchmark with fixed damage scenarios and controlled testing budgets.
Authors
Siyuan Zhang
Abstract
We present a repertoire-free benchmark for soft-robot damage recovery under nine online trials. The protocol pairs damage masks within each morphology, retains a measured nominal fallback, and separates method development from evaluation on new bodies. A Gaussian-process expected-improvement (GP-EI) reference controller adapts actuator phases using three initialization probes and six feedback-selected rollouts. Across two disjoint 75-body cohorts, it outperforms random search, Sobol, CEM, and CMA-ES under equal budgets. GP-EI improves the worst-mask gain on 69/75 confirmation bodies and exceeds these four baselines by 0.235-0.310 mean worst-mask reward. A TuRBO-style local GP is the closest comparator; the paired confidence interval includes zero. Post-confirmation fixed-controller replay finds 5.06 voxel widths of mean recovery together with a 0.0031 increase in worst-mask p99 geometric edge strain; a strict no-added-demand deployment gate retains 48.7% of mean gain. In a frozen development stress test that physically removes 10% of occupied voxels, GP-EI improves 71/75 bodies and exceeds official BoTorch TuRBO by 0.356 mean worst-mask gain. Together, the replayable controllers, body-level inference, and frozen-cohort evaluation provide a reference for measuring the added value of learned damage priors and future adaptation methods.