Data-driven Distributed Multi-Robot Coordination through Neural-based Potential Functions

In this work, we propose a data-driven potential-based control framework for multi-robot coordination. Each robot is equipped with an interaction device, i.e., a device that allows a robot to communicate with or perceive another robot, such as a camera or an antenna, which can be used to gather data regarding the quality of the interaction. This data is leveraged to build data-driven pairwise potential functions, which may be difficult or impossible to derive in a closed analytical form as typical in potential-based control frameworks. To achieve this, we model a pairwise potential function using a neural network, where, to ensure system stability for multi-robot coordination, we impose the necessary properties drawn from analytical potential-based multi-robot frameworks. Simulation results with different interaction devices and laboratory experimental results validate the proposed approach.