Path Planning for Efficient Deployment and Collection of a Marsupial Robot Team

Abstract

This dissertation presents time-efficient approaches to path planning for initial deployment and collection of a heterogeneous marsupial robot team consists of a large-scale carrier robot and multiple mobile robots. Although much research has been conducted about task allocation and path planning of multi-robot systems, the path planning problems for deployment and collection of a marsupial robot team have not been fully addressed. The objectives of the problems are to minimize the duration that mobile robots require to reach their assigned task locations and return from those locations. Taking the small mobile robot's energy constraint into account, a large-scale carrier robot, which is faster than a mobile robot, is considered for efficient deployment and collection. The carrier robot oversees transporting, deploying, and retrieving of mobile robots, whereas the mobile robots are responsible for carrying out given tasks such as reconnaissance and search and rescue. The path planning methods are introduced in both an open space without obstacles and a roadmap graph which avoids obstacles. For the both cases, determining optimal path requires enormous search space whose computational complexity is equivalent to solving a combinatorial optimization problem. To reduce the amount of computation, both task locations and mobile robots to be collected are divided into a number of clusters according to their geographical adjacency and their energies. Next, the cluster are sorted based on the location of the carrier robot. Then, an efficient path for the carrier robot can be generated by traveling to each deploying and loading location relevant to each cluster. The feasibility of the proposed algorithms is demonstrated through several simulations in various environments including three-dimensional space and dynamic task environment. Finally, the performance of the proposed algorithms is also demonstrated by comparing with other simple methods.