Event

Learning-based inverse design framework turns biological curves into soft robot configurations

1 reports · 1 sources

Dixiang Wang and co-authors propose a learning-based real-to-sim-to-real inverse design framework that converts biological curves into target shape configurations of cable-driven modular soft robots through surrogate-assisted inverse design and uncertainty-guided correction. According to the paper, the learning-based forward surrogate reduces deformation prediction time from 1.53×10^5 ms to 0.0008 ms (approximately 1.9×10^8-fold speedup), and inverse design time from 1.99×10^6 ms to 4.18 ms. On 5000 test samples generated from previously unseen threading patterns, the method achieves a 93% shape reconstruction success rate under surrogate-based evaluation; the authors also demonstrated four biomimetic robots physically, showing that biological curves can be converted into realizable robot shapes.