Collaborative gym
A simulation benchmark for multi-robotic tasks
G. Ansaldo (TU Delft - Mechanical Engineering)
Frederik Schulte – Mentor (TU Delft - Transport Engineering and Logistics)
Babak Heydari – Mentor (Northeastern University)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
The design of multi-robot systems has gained increasing attention in recent years. The field of cooperative Multi-Agent Robot Systems (MARS) has shown the potential to provide reliable and cost-effective solutions to a wide range of automated applications. Communication and coordination between autonomous agents require robust and intelligent control systems in order to achieve high-quality performance. This paper presents Collaborative Gym, an open-source, physics-based simulation framework for multi-robot interaction. This simulation environment differs from existing robotic simulation environments in that it is designed to model the interaction between multiple robots. Despite the presence of a large number of single robotic environments, multi-robotic simulation environments for reinforcement learning are rare. Collaborative Gym contains four simulated tasks in which different commercial robots work in collaboration: poking, lifting, balancing, and passing. For each of the four tasks, baseline policies are presented for various combinations of commercial robots which have been trained using reinforcement learning. The study demonstrated that Collaborative Gym is a promising open-source framework for the development of multi-robotic collaborative robotic tasks.