Computing possible driving corridors for automated vehicles

Motion planing in dynamic traffic scenes is a challenging problem. In particular, since it is unknown during planning whether a certain decision, such as passing another traffic participant on the left or right, will result in a safe and comfortable motion. Exhaustive exploration of all principle driving paths is computationally expensive, so that one typically reverts to heuristics - this, however can be unsatisfactory in situations when the heuristics fail to find a solution although it exists. We address this problem by computing the union of all possible motions for a sequence of high-level decisions (e.g. overtake vehicle on the left and then another one on the right), which we refer to as a driving corridor. Our proposed algorithm is over-approximative, i.e. the union of driving corridors provably encloses all possible motions. Thus, if the set of reachable positions within a driving corridor becomes empty, the corresponding sequence of high-level decisions is infeasible and can be discarded by the motion planner. Driving corridors also facilitate selecting high-level plans: Large driving corridors should be preferred since they provide more opportunities for optimizing motions and are more robust towards unpredicted changes. Numerical examples demonstrate the usefulness of our approach.