Our work on optimized swimmers and simulation accepted at ICML 2022

How can we leverage advances in Physics-Informed Neural Networks to design soft robotic swimmers that perform well in real water? At the Soft Robotics Lab we developed a novel method to integrate soft body and hydrodynamics simulations that are both lightning fast and fully differentiable, with potential great impact on computational controller and shape design for swimmers and beyond.

Specifically, we combine differentiable Finite Element Method simulation of the swimmer's soft body with a neural network-based surrogate model of the fluid medium which is fully learned in a self-supervised manner, and as a result obtain a powerful, sufficiently general and fast simulator that can be used for design tasks where previous computationally intensive and non-differentiable methods could not be effectively employed. We demonstrate the computational efficiency and differentiability of our hybrid approach by finding the optimal swimming frequency of a simulated 2D soft body swimmer through gradient-based optimization, but exciting future applications of the technique could involve full 3D shape optimization, real world robotic fabrication, and the training of neural network based controllers without expensive Reinforcement Learning.

You can consult the preprint external page here. If you are interested in working on this or similar projects, please contact us at rkk at ethz dot com !

"Fast Aquatic Swimmer Optimization with Differentiable Projective Dynamics and Neural Network Hydrodynamic Models", 2022, International Conference of Machine Learning, E. Nava, J. Z. Zhang, M. Y. Michelis, T. Du, P. Ma, B. F. Grewe, W. Matusik, R. K. Katzschmann. external page https://arxiv.org/abs/2204.12584