Safe-Path Graph Generation for Path Planning in Urban Environments

To serve as a platform for multi-agent research, the BYU MAGICC Lab has created a version of the game of capturethe-flag played in a maze environment that can be configured to model a variety of indoor and urban settings. Each competing team has mobile robots on the ground, a simulated UAV flying overhead, and a remote human commander. Since simultaneous low-level control of all robots is not possible, directives from the human operator must be at a suitably high level. In the context of our game, most important high-level directives have to do with point-to-point movement through the maze. To support this functionality, robots need to employ fast path planning algorithms that create short, safe paths. Such path planners consist of two main parts: one that generates and maintains some type of graph representing the world, and a second part that searches the graph to create a path. This paper focuses on the problem of generating a graph that provides a good representation of the navigable space in a maze. We identify criteria to judge the quality of a graph, including run-time and memory requirements, scalability, and the quality of paths generated using the graph. Alternative methods of generating graphs are examined and their weaknesses are discussed. A new graph generation algorithm is described that offers several important advantages, and results are presented that demonstrate the significant advantages of the new algorithm.