Using Intelligent Agents to Build Navigation Meshes

We present a novel algorithm that allows agents to discover a navigation mesh for an environment as they move through the environment. The Navigation-Mesh Automated Discovery (NMAD) algorithm works by constructing its best guess for the navigation-mesh of a game level and then refines it when the agents moving through the world using the navigation mesh encounter unexpected or unknown obstacles. Using this algorithm, agents can enter a world in which they know nothing about, while still enjoying all of the advantages of a navigation mesh for path planning. We validated the effectiveness of this technique by showing that for both random and deliberative searches through multiple game worlds the error present in the best guess approximation the navigation mesh generated and maintained by NMAD converges to zero.