CASS: Cooperative Autonomous, Safe and Scalable Multi-Agent Motion Planning

Abstract

One of the fundamental challenges preventing the widespread adoption of autonomous robots in daily life is the ability to guarantee their safety in various environments in a scalable manner. Safety-critical tasks such as deploying a large-scale fleet of fully autonomous robots, in indoor as well as outdoor environments, are challenging due to a wide variety of reasons. Many existing theoretical frameworks that guarantee safety, scale intractably with an increase in the number of agents or complexity in the environment. In this thesis, we propose CASS, a novel, safe and scalable framework for autonomous multi-agent motion planning in a cooperative setting with perfect information. We demonstrate the performance of our framework through extensive experimentation on quadrotors in the Flightmare quadcopter simulator.