Intelligent Planning

The purpose of the intelligent planning system is to use the scene estimator’s probabilistic estimates of the environment to determine vehicle actions. The system is designed from three basic requirements. First, the planner must be expandable such that additional behaviors can easily be added as the system matures. Second, the planner must accommodate asynchronous operation, so low level control loops can run much faster than high level trajectory planning. Finally, the system must be stable and robust despite potentially unpredictable actions of other vehicles in the environment.

The behavioral layer functions as a global situation analyzer. First a traditional search technique is used to determine the minimum time global path to the next mission checkpoint, taking into account a priori RNDF information as well as semi-permanent obstacle locations and even transient traffic jams. Once the desired global route is chosen, a situational analysis is performed to determine how to plan a trajectory over the portion of the global environment nearest to the vehicle.

The tactical layer is broken up into an ever-expanding set of components, each designed to mimic a single behavior for a specific DUC scenario. Many of these behaviors can be considered human-like, such as passing a vehicle, moving around an obstacle, and stopping at an intersection. These components interpret obstacle data to generate a piece of a trajectory appropriate for the situation presented. Three of the most commonly-used tactical components, are lane reasoning, intersection reasoning, and zone reasoning.