Traffic Behavior Planning Using Epistemic Uncertainty Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional prediction methods for automated vehicles in traffic scenes struggle to accurately account for the epistemic uncertainty of other participants' behaviors, leading to potential safety issues due to perception-related shortcomings and insufficient performance of deep learning-based prediction components.
Innovation Solution
The method determines and evaluates the epistemic uncertainty of predicted behavior for each prediction step using a decision criterion, deciding whether to include the behavior in the planning for the first participant, thereby ensuring only reliable predictions are considered, and optionally terminating autoregressive predictions when uncertainty exceeds a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning-based prediction methods are used to predict behavior of other participants, then the ability to include extensive context information is improved, but epistemic uncertainty due to perception shortcomings and insufficient model performance increases
Solution Approach 1:
The patent introduces an uncertainty evaluation module as an intermediary between the deep learning prediction component and the behavior planning component. This module assesses the epistemic uncertainty of predictions and filters out unreliable predictions before they reach the planning stage, thereby resolving the contradiction by adding a mediating layer that handles the reliability issue without sacrificing the contextual information processing capability
Solution Approach 2:
The system implements feedback by evaluating prediction uncertainty and using this information to adjust which predictions are passed to the planning component. The uncertainty evaluation creates a feedback loop that improves reliability by continuously monitoring and filtering predictions based on their confidence levels, allowing the system to maintain high adaptability while improving reliability through selective acceptance of predictions
2Reliability
If all prediction steps are considered in behavior planning, then planning completeness is improved, but computing power consumption increases
Solution Approach 1:
The patent extracts only the reliable predictions from the set of all predictions by evaluating epistemic uncertainty. The uncertainty evaluation module identifies and separates unreliable predictions, which are then excluded from the behavior planning process. This extraction principle reduces computing power consumption by processing only the necessary subset of predictions while maintaining planning completeness for reliable cases
Solution Approach 2:
Instead of processing all prediction steps equally, the system applies partial action by selectively processing only those predictions that meet the reliability threshold. The uncertainty evaluation enables the system to perform partial processing - considering only the portion of predictions that are sufficiently reliable - thereby reducing computational energy consumption while maintaining adequate planning completeness
3Reliability
If perception equipment is enhanced to reduce epistemic uncertainty, then prediction reliability is improved, but device complexity and cost increase
Solution Approach 1:
The uncertainty evaluation module serves as an intermediary that compensates for perception limitations without requiring additional sensors. It processes the existing sensor data and quantifies the epistemic uncertainty, allowing the system to achieve better reliability through software-based uncertainty management rather than hardware expansion
Solution Approach 2:
The patent replaces the mechanical approach of adding more sensors with a computational approach - the uncertainty evaluation module uses algorithms to assess and filter predictions based on epistemic uncertainty. This substitution of mechanical sensor enhancement with computational uncertainty management reduces device complexity while improving prediction reliability
Data Source
AI summary
A computer-implemented method for behavior planning for a first participant in a traffic scene, in which a predicted behavior of at least one further participant in the traffic scene is taken into account, specifically depending on the epistemic uncertainty of the predicted behavior. For this purpose, the behavior of the further participant is predicted at regular time intervals, in each case for a prediction interval of a specified length, wherein each prediction interval includes a specified number of consecutive prediction steps. The epistemic uncertainty of the predicted behavior is determined for each prediction step of a prediction interval and is evaluated using a decision criterion to decide whether the behavior of the further participant predicted for the corresponding prediction step is taken into account in the behavior planning for the first participant.