Experiments in robot formation control with an emphasis on decentralized control algorithms

Abstract

In this thesis, several algorithms and experiments involving the control of robot formations are presented. The algorithms used were focused on decentralized control. The experiments were implemented on two different experimental testbeds consisting of teams of wheeled mobile robots. The robots used are described along with their sensors and supporting hardware. Also, there is a discussion of the programming framework used to build the control software. The first control algorithm and experiment uses a robust consensus tracking algorithm to control a formation of robots to track a desired trajectory. The robots must maintain the correct formation shape while the formation follows the trajectory. This task is complicated by limited communication between the robots, and disturbances applied to the information exchange. Additionally, only a subset of the robots have access to the reference trajectory. In the second experiment, the same algorithm was re-implemented in a decentralized way, which more effectively demonstrated the goals of the algorithm. The second algorithm involves controlling a formation of robots without a global reference frame. In order to accomplish this, the formation description is reformulated into variables describing the relative positions of the robots, and vision-based measurements are used for control. A homography-based technique is used to determine the relative positions of the robots using a single camera. Then a consensus tracking controller similar to the one used previously is used to distribute the measured information to all of the robots. This is done despite the fact that different parts of the information are measured by different agents.