Autonomous Vehicle Motion Planning With Hierarchical Interaction Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion planning methods for autonomous vehicles, particularly interaction-aware models, face computational complexity issues that make them intractable for real-time applications on vehicle-embedded systems.
Innovation Solution
A hierarchical interaction-aware model that orders traffic participants from most to least determinative, allowing recursive computation of states and sorting based on road rules, reducing computational complexity by decomposing the problem into N computation problems, where the η-th participant interacts with η−1 participants only.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If interaction-aware motion models are used to improve prediction accuracy, then prediction reliability is improved, but computational complexity increases making real-time processing infeasible
Solution Approach 1:
The patent segments the computation of traffic participant states by introducing a hierarchical structure where participants are divided into groups based on their determinative relationships. Instead of computing all N participants simultaneously with full interaction awareness, the system computes them in a hierarchical sequence, reducing the computational burden from O(N²) to O(N) while preserving interaction-awareness for critical participants.
Solution Approach 2:
The patent applies local quality by differentiating the level of interaction-awareness applied to different traffic participants. The hierarchical list assigns different computation depths to different participants based on their determinative importance. High-priority participants (those that are more determinative) receive full interaction-aware computation, while lower-priority participants use simplified models, optimizing the trade-off between accuracy and computational cost.
2Measurement precision
If full interaction-aware computation is performed for all traffic participants, then prediction accuracy is improved, but processing time increases beyond real-time capabilities
Solution Approach 1:
The patent performs preliminary action by pre-sorting traffic participants into a hierarchical list based on their determinative relationships before computation. This sorting establishes a computation sequence that allows the system to process high-priority participants first with full interaction-awareness, while lower-priority participants can be processed more quickly using simplified models, ensuring real-time performance while maintaining accuracy for critical participants.
3Productivity
If hierarchical decomposition is applied to reduce computational complexity, then processing speed is improved, but model accuracy may be reduced by simplifying interactions
Solution Approach 1:
The patent applies local quality by differentiating the level of interaction-awareness applied to different traffic participants. The hierarchical list assigns different computation depths to different participants based on their determinative importance. High-priority participants (those that are more determinative) receive full interaction-aware computation, while lower-priority participants use simplified models, optimizing the trade-off between accuracy and computational cost.
Data Source
AI summary
The invention is notably directed to computer-implemented method of planning motion of a vehicle. The method comprises receiving (S20) real-time signals as to positions of N traffic participants. At each time step of multiple time steps, the method plans (S50) motion for the vehicle by computing (S30) states of each of the N traffic participants according to the signals received (S20). Said states include current states, which are estimated for each of the N traffic participants, as well as future states of each of the participants, wherein the future states are predicted over a prediction horizon T. This is achieved using an interaction-aware model of the N traffic participants. This model is designed to cause the method to recursively compute (S32-S38), at said each time step, the states of each of the N traffic participants according to a hierarchical list. The N traffic participants are ordered in the list from a most determinative one to a least determinative one of the N traffic participants. As a result, the states of the η-th participant in the list are computed based on the states of each of the η−1 participants in the list, ∀η∈[2, . . . , N]. This model is further designed to cause the method to update and sort (S40) the hierarchical list based on rules of the road, wherein the rules are evaluated based on the states of each of the N traffic participants as computed at said each time step. The invention is further directed to related computerized systems, vehicles, and computer program products.