NeuroMesh: A Unified Neural Inference Framework for Decentralized Multi-Robot Collaborative Learning

Yang Zhou¹, Yash Shetye¹, Long Quang^1,3, Devon Super², Jesse Milzman³, Manohari Goarin¹, Aditya Azad¹, Devang Sunil Dhake¹, Jeffery Mao¹, Carlos Nieto-Granda³, Giuseppe Loianno¹,

Conferance name and year
¹The authors are with New York University, New York, USA. email: {yangzhou, ys5153, lq2146, mg7363, aa10878, dsd9855, jm7752, loiannog}@nyu.edu.
²The author is with Vanderbilt University, Tennessee, USA. email: devon.a.super@vanderbilt.edu.
³The authors are with U.S. Army Combat Capabilities Development Command, Army Research Laboratory, Adelphi, MD 20783, USA. email: {long.p.quang.civ, jesse.m.milzman.civ, carlos.p.nieto2.civ}@army.mil.

Paper Supplementary Code arXiv

The NeuroMesh framework allows multiple robots to collaborate on any number of tasks, including depth perception.

Abstract

NeuroMesh is a general, modular, and decentralized deployment framework that enables robot teams to collaborate through information-sharing. The framework enables flexible and efficient deployment of multi-robot collaborative learning algorithms in various domains, including perception, control, planning. We evaluate the proposed approach on perception and control tasks in real-world deployment and the results show the proposed framework can adapt to different problem domains, satisfying real-robot deployment needs. We plan to release NeuroMesh as an open-source framework to the robotics community.