Action-Primitive Planning for Automated Driving Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated driving systems face challenges in decision-making in complex and uncertain situations involving dynamic objects, such as pedestrians and other vehicles, leading to unnatural behavior and potential collisions due to the inability to effectively plan trajectories that account for multiple uncertain factors and intentions.
Innovation Solution
A data-driven approach is introduced, utilizing an action-primitive planner that accounts for uncertainty by selecting a sequence of discrete action primitives, such as steering, braking, or honking, based on driving trends and intentions of dynamic obstacles, which are learned from a large-scale driving behavioral database, allowing for more natural and efficient motion planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a traditional trajectory planner is used to generate short-term paths, then the vehicle can respond to immediate obstacles, but the motion planning becomes intractable and time-consuming in complicated driving areas with high uncertainty and dynamic objects
Solution Approach 1:
The planning system is divided into two independent modules: a global route planner that determines the overall path to destination, and a local trajectory planner that generates short-term paths for immediate obstacles. This segmentation allows each module to operate independently with appropriate complexity level, avoiding intractable computation while maintaining responsive behavior.
2Adaptability or versatility
If a traditional trajectory planner frequently constructs alternative paths in response to uncertain environments, then the vehicle can adapt to dynamic obstacles, but the motion planning becomes intractable and time-consuming
Solution Approach 1:
The global route planner pre-computes the overall path to the destination before the vehicle encounters specific obstacles. This preliminary action provides a baseline trajectory that reduces the need for frequent alternative path construction, thereby decreasing planning time while maintaining adaptability through the local trajectory planner when immediate obstacles arise.
3Speed
If a purely short-term path planning approach is used, then the vehicle can respond quickly to immediate obstacles, but collisions may occur since it does not consider uncertainty in the intentions of dynamic obstacles
Solution Approach 1:
The system introduces an intermediary layer between the global route planner and local trajectory planner that incorporates uncertainty modeling of dynamic obstacles. This intermediary layer allows the short-term path planning to consider potential intentions of pedestrians and vehicles, improving collision avoidance while maintaining quick response capability.
Data Source
AI summary
A system and method are provided and include a subject vehicle having vehicle actuation systems. A driving context database stores driving trends associated with driving locations. A state estimation module determines a current location of the subject vehicle. A global route planner module determines a route to an inputted destination. An action primitive planning module retrieves a driving trend associated with the current location, selects a sequence of action primitives based on the driving trend associated with the current location and based on the determined route, and generates waypoints for the subject vehicle to travel to based on the sequence of action primitives, each waypoint including location coordinates and a direction. A trajectory planner module determines a trajectory for the subject vehicle based on the waypoints. A vehicle control module controls the vehicle actuation systems based on the determined trajectory.