FishGym: A High-Performance Physics-based Simulation Framework for Underwater Robot Learning

Bio-encouraged underwater robots have a good deal of advantages, such as robust maneuverability and propulsion performance. However, training their behaviors in genuine environments is tough because of high bodily charge. For that reason, a simulation system for schooling manage guidelines of these robots would be helpful for researchers.

Image credit:  arXiv:2206.01683 [cs.RO]

Picture credit rating: arXiv:2206.01683 [cs.RO]

A the latest paper on proposes a substantial-effectiveness simulation platform focusing on the two-way interaction dynamics concerning fish-like underwater robots and the surrounding fluid ecosystem. It makes use of A GPU-accelerated lattice Boltzmann solver to product significant-performance fluid-structure interaction in a regional moving body of reference.

Scientists suggest an algorithm customized for bionic underwater robots that can obtain normal and productive command insurance policies for swimming. Eventually, a collection of benchmark duties is provided to consider and compare diverse finding out approaches and control procedures.

Bionic underwater robots have demonstrated their superiority in quite a few purposes. Nevertheless, schooling their intelligence for a selection of responsibilities that mimic the behavior of underwater creatures poses a number of challenges in follow, primarily because of to deficiency of a significant amount of money of readily available education knowledge as well as the high price tag in actual bodily surroundings. Alternatively, simulation has been regarded as a practical and crucial tool for acquiring datasets in various environments, but it typically qualified rigid and gentle physique programs. There is at present dearth of work for a lot more complicated fluid units interacting with immersed solids that can be proficiently and precisely simulated for robot coaching purposes. In this paper, we propose a new platform termed “FishGym”, which can be applied to teach fish-like underwater robots. The framework is composed of a robotic fish modeling module making use of articulated entire body with skinning, a GPU-based mostly substantial-effectiveness localized two-way coupled fluid-structure interaction simulation module that handles both equally finite and infinitely huge domains, as perfectly as a reinforcement mastering module. We leveraged current teaching strategies with diversifications to underwater fish-like robots and acquired discovered regulate guidelines for many benchmark responsibilities. The training results are demonstrated with realistic motion trajectories, with comparisons and analyses to empirical models as properly as recognized true fish swimming behaviors to highlight the benefits of the proposed platform.

Research report: Liu, W., Bai, K., He, X., Track, S., Zheng, C., and Liu, X., “FishGym: A Higher-Functionality Physics-centered Simulation Framework for Underwater Robot Learning”, 2022. Website link: muscles/2206.01683