ABSTRACT
Driver San Francisco was an open-world driving game set in a fictionalized version of the city of San Francisco. The game's version of the city was very large, and in some areas had very dense traffic. This chapter presents the path planning solution Driver San Francisco used: a three-tier path optimization approach that provided locally optimal paths. It describes three stages of the process—route finding, mid-level path planning, and low-level path optimization. Driver San Francisco's AI used state-of-the-art driving AI techniques that allowed the game to run a simulation with thousands of vehicles that could navigate the game's version of the city with a very high level of quality, as our path generation and following used the same physics model as any player-driven vehicle. Nontraffic vehicles were controlled by the more complex Active Life AI system. The system was composed of multiple levels of detail, each of which updated vehicle paths only when required to do so.
