Efficient Cooperative Path-Planning

Even after lots of effort made by today game companies, recent state of the art real-time strategy (RTS) games (e.g. Age of Empires, Warcraft III) lack an efficient robust path planner for mass multi-agent path planning. The main reason for this failure is that current methods do not try to solve the whole path planning problem as one task. One recent work that addresses this issue is Cooperative Pathfinding [D. Silver, 2006]. The main idea behind his technique is to extend the map into time dimension and save it in memory. Each agent would use a path finding method in order to find its goal and reserve the path in this memory. This will help the next agent to consider the previous agents’ paths and positions in each time step in his path planning and avoid collision. The main drawback of this method is that in large maps, the memory requirement is extensively high. In our new approach the same method would be applied to limited areas of the map. These areas will be identified using a simple path finding for each agent and then finding the collision points. Our main idea for these extension areas is to consider a bound around each collision point. This way we don’t keep the whole map information in each time step and just keep the information around collision spots. therefore, we save a great deal of memory in our method. The test-bed of this project would be Hierarchal Open Graph (HOG)[N. Sturtevant]. Since we haven’t been working with this system before, one of the initial steps of the project is to get familiar with this environment.