Motion control strategies for networked robot teams in environments with obstacles.

摘要

Communication in multi-robot teams has, historically, been a means to improve control and perception. Recent advances in embedded processor technology have made it possible to equip every robot with inexpensive off-the-shelf wireless communication capabilities. These advances have also given robots the ability to monitor and respond to changes in the quality of their communication links. As such, progress in multi-agent robotics and sensor networks, and particularly the convergence of the two, will inevitably engender problems at the intersection of communication, control and perception.;While control is necessary for successful mission execution, reliable communication is essential for coordination and cooperation in multi-robot teams. For example, in applications such as perimeter surveillance or the cordoning off of hazardous regions, robots must be capable of forming complex shapes in the plane while maintaining the quality of the communication network. Thus, motion control strategies that do not require inter-agent communication can often be beneficial since they preserve limited bandwidth for the transmission of critical data. This is especially relevant in teams composed of large a number of small, resource constrained agents where bandwidth often becomes the limiting factor in agents' abilities to communicate.;Towards this end, this thesis considers scalable motion control strategies for networked robot teams that can be implemented with no inter-agent communication. Experimental studies of strategies for maintaining end-to-end communication links for tasks like surveillance, reconnaissance, and target search and identification are discussed in the first part of the thesis. This then motivates the work presented in the second part: the synthesis of decentralized controllers for robot teams to form complex patterns in two dimensions. These decentralized controllers do not require the explicit communication of robots' state information. Rather, agents are assumed to be equipped with appropriate sensors, enabling them to infer relative position and bearing information of their neighbors. The stability and convergence properties of the controllers are presented, and the feasibility of the proposed methods is verified via computer simulations and experimental results using two multi-robot testbeds.