Situation Prediction And Reaction Control (SPARC)

Approaches to automated driving typically hand over vehicle control to specialized modules for intersection handling, parking, obstacle avoidance etc., depending on the perceived traffic situation. This paper proposes a continuous-state alternative that allows to take all modeled goals and influences into account simultaneously, similarly to how a human driver would behave. A dynamic map of the environment is analyzed in real time for traffic rules and obstacles. The behavior of dynamic objects is predicted into the near future. This information is used to generate a d penalty map over space and time. An optimal trajectory is found based on these penalties as well as on penalties for internal control parameters. This holistic approach considers all relevant goals as well as the dynamic limits of the ego vehicle simultaneously when planning the trajectory, and requires no sharp state transitions during operation.