Multiple Hypothesis Planning for Vehicle Control

Abstract

One of the most difficult aspects of planning the motion of autonomous vehicles is the uncertainty regarding the future behavior of other road users. Ensuring a safe motion often requires preparing for multiple plausible outcomes of a current road situation. To achieve this, we propose an optimization-based trajectory generation method capable of planning an efficient motion while ensuring a collision-free behavior with respect to several conflicting hypotheses. The method generates multiple control trajectories that overlap for a certain time period, ensuring that the initial part of the trajectory can be continued with a safe motion regardless of which hypothesis will prove to be true. One notable application of the proposed method is an implementation of a Fail-Safe Planning concept, where the vehicle’s trajectory is planned based on a most plausible prediction of the other road users’ behaviors while simultaneously a fail-safe trajectory is generated based on a worst-case prediction, ensuring the existence of a collision-free maneuver that can be used in emergency situations.