Resumen de Multi-robot persistent coverage in complex environments
Recent advances in mobile robotics and an increasing development of a ordable autonomous mobile robots have motivated an extensive research in multi-robot systems. The complexity of these systems resides in the design of communication, coordination and control strategies to perform complex tasks that a single robot can not. A particularly interesting task is that of persistent coverage, that aims to maintain covered over time a given environment with a team of robotic agents. This problem is of interest in many applications such as vacuuming, cleaning a place where dust is continuously settling, lawn mowing or environmental monitoring. More recently, the apparition of useful unmanned aerial vehicles (UAVs) has encouraged the application of the coverage problem to surveillance and monitoring.
This thesis focuses on the problem of persistently covering a continuous environment in increasingly more dicult settings. At rst, we propose a receding-horizon optimal solution for a centralized system in a convex environment using dynamic programming. Then we look for distributed solutions, which are more robust, scalable and ecient. To deal with the lack of global information, we present a communication-e ective distributed estimation algorithm that allows the robots to have an accurate estimate of the coverage of the environment even when they can not exchange information with all the members of the team. Using this estimation, we propose two di erent solutions based on coverage goals, which are the points of the environment in which the coverage can be improved the most. The rst method is a motion controller, that combines a gradient term with a term that drives the robots to the goals, and which performs well in convex environments. For environments with some obstacles, the second method plans open paths to the goals that are optimal in terms of coverage. Finally, for complex, non-convex environments we propose a distributed algorithm to nd equitable partitions for the robots, i.e., with an amount of work proportional to their capabilities. To cover this region, each robot plans optimal, nite-horizon paths through a graph of sweep-like paths.
The nal part of the thesis is devoted to discrete environment, in which only a nite set of points has to be covered. We propose a divide-and-conquer strategy to separate the problem to reduce its complexity into three smaller subproblem, which can be optimally solved. We rst plan closed paths through the points, then calculate the optimal coverage times and actions to periodically satisfy the coverage required by the points, and nally join together the individual plans of the robots into a collision-free team plan that minimizes simultaneous motions. This solution is eventually used for a novel application that is domestic induction heating with mobile inductors. We adapt it to the particular setting of a domestic hob and demonstrate that it performs really well in a real prototype.
