Safe Navigation on Path-Following Tasks: A Study of MPC-based Collision Avoidance Schemes in Distributed Robot Systems

Abstract

This work aims to enable distributed robot systems to follow time-varying paths safely. Artificial Vector Fields offer a viable alternative for addressing path-following challenges, yet collision avoidance among agents guided by such fields remains an open problem. To address this, we have designed a Model Predictive Control (MPC) setup that integrates an Artificial Vector Field reference with prominent collision avoidance methods, such as Optimal Reciprocal Collision Avoidance (ORCA) and Control Barrier Functions (CBF), to produce real-time, safe solutions. Our work involves a direct comparison between different MPC-based collision avoidance methods, and we have obtained results from various simulation scenarios as well as experiments on real robotic systems (Crazyflie 2.1). We aim to assess the applicability and limitations of these techniques through extracted metrics and insights.