The Holy Grail of Multi-Robot Planning: Learning to Generate Online-Scalable Solutions from Offline-Optimal Experts

Abstract

Many multi-robot planning problems are burdened by the curse of dimensionality, which compounds the difficulty of applying solutions to large-scale problem instances. The use of learning-based methods in multi-robot planning holds great promise as it enables us to offload the online computational burden of expensive, yet optimal solvers, to an offline learning procedure. Simply put, the idea is to train a policy to copy an optimal pattern generated by a small-scale system, and then transfer that policy to much larger systems, in the hope that the learned strategy scales, while maintaining near-optimal performance. Yet, a number of issues impede us from leveraging this idea to its full potential. This blue-sky paper elaborates some of the key challenges that remain.

A Framework for Real-World Multi-Robot Systems Running Decentralized GNN-Based Policies

“…However, none of these systems utilize machine-learningbased policies, and only few learning-based methods have demonstrated real-world experiments [5]. Although numerous works at the intersection of machine-learning and multirobot control show remarkable performance [5], [6], [10], [25], [26], [27], little work has been done to show how these methods can be made practicable (i.e., in the real-world). Of particular interest to us is how explicit inter-robot communication [9], [10], [28] plays a role in accumulating information from other robots.…”