Multileader and Role-Based Time-Varying Formation Using GP Inference and Sliding-Mode Control

This article presents a novel approach to time-varying formation for a multiagent system using leader–follower dynamics. A team consisting of two-wheeled mobile robots is considered. A high level optimization method for leaders was implemented and low level sliding-mode control for followers was used simultaneously for scale, density transparency, and fault tolerance. Trajectories and collaboration for the leaders of the system are determined using a Gaussian process (GP) inference, and followers are controlled via a terminal sliding-mode controller. This allows agents navigate an environment while maintaining a formation from arbitrary initial positions. All agents utilize sensor data to locally detect collisions and avoid obstacles, employing an artificial potential field process. This approach can be extended to accommodate a large number of agents. Simulations are conducted in various environments, demonstrating the capability of our agents to successfully navigate the given environments. Experimental results of a team of four TurtleBots validate the effectiveness of the proposed approach.