Ego Vehicle Behavior Planning With One-Shot Traffic Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search-based planning approaches for automated vehicles struggle with efficiently predicting meaningful maneuvers in complex traffic scenarios, leading to large and unrealistic search spaces due to separate implementation of prediction and planning components, which hinders real-time applications.
Innovation Solution
Integrate one-shot predictions for future developments of the traffic scenario into the search tree structure, associating each scenario representation with a one-shot prediction to focus planning on realistic outcomes, using deep learning to generate sampling distributions that consider participant behaviors and driving styles, reducing the need for pruning and enhancing computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate prediction and planning components are used, then comprehensive scenario analysis is achieved, but computational efficiency deteriorates and search space becomes unmanageably large
Solution Approach 1:
The patent combines separate prediction and planning components into an integrated system where the planning component directly generates a search tree of future scenario representations. This merging eliminates the need for separate prediction steps and reduces computational overhead while maintaining comprehensive scenario analysis through the tree structure that explores multiple possible future states.
Solution Approach 2:
The planning component performs preliminary generation of multiple future scenario representations in a tree structure before actual planning decisions are made. By pre-computing and organizing possible future scenarios in the search tree, the system prepares actionable information in advance, reducing real-time computational burden while ensuring comprehensive scenario coverage.
2Measurement precision
If comprehensive search space exploration is performed, then planning accuracy is improved, but computational burden increases excessively
Solution Approach 1:
The patent segments the comprehensive search space into a hierarchical tree structure of scenario representations, where each node represents a specific future state and branches represent possible transitions. This segmentation allows the system to explore comprehensive scenarios systematically while managing computational burden by processing scenarios in discrete, organized units rather than as a monolithic search space.
Solution Approach 2:
The system generates multiple future scenario representations (excessive action) to ensure comprehensive coverage, but evaluates and prunes scenarios based on relevance and probability (partial action). By generating more scenarios than strictly necessary and then filtering them, the system ensures high planning accuracy while managing computational resources through selective evaluation of the most promising paths in the search tree.
Data Source
AI summary
A computer-implemented method is for search-based behavior planning for an ego vehicle in a traffic scenario involving at least one further participant. A scenario representation of the traffic scenario is generated based on aggregated scenario-specific information in order to generate, using a deep learning based planning component, a tree structure including multiple sequences of scenario representations for N>1 consecutive planning time increments i, i∈{0, . . . , N}. At least one one-shot prediction is also generated for at least one possible development of the traffic scenario for M>1 consecutive prediction time increments in order to associate the individual sequences of the tree structure with at least one such one-shot prediction. The subsequent scenario representations are generated in individual planning time increments i, i∈{1, . . . , N}, each based on at least one such one-shot prediction.