Lane-Topology Evaluation of Motion Prediction Model Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion prediction models for autonomous vehicles fail to accurately differentiate between longitudinal and lateral errors, leading to unsafe navigation due to unacceptably high lateral error tolerance, which can result in collisions.
Innovation Solution
A computing system evaluates motion prediction models by matching predicted trajectories to underlying lane topology, using distance and orientation-based assignment confidences to determine accuracy, reducing lateral error tolerance and enhancing safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion prediction models use traditional error evaluation methods, then longitudinal errors are treated equally with lateral errors, but this results in unacceptably high lateral error tolerance that compromises safety
Solution Approach 1:
The patent segments the error evaluation into two distinct components: longitudinal error and lateral error. By dividing the unified error metric into separate dimensions, the system can apply different tolerance thresholds and evaluation criteria to each direction, thereby preventing lateral errors from being masked by longitudinal errors and improving overall navigation safety
Solution Approach 2:
The patent applies different quality standards to different spatial directions. Lateral error tolerance is set to a stricter threshold compared to longitudinal error tolerance, recognizing that lateral deviations pose greater safety risks. This local differentiation of error tolerances ensures that safety-critical lateral predictions are evaluated more rigorously
2Reliability
If motion prediction models use lane topology matching, then lateral error tolerance is reduced, but this increases the complexity of the evaluation system
Solution Approach 1:
The patent performs preliminary actions by pre-establishing the lane topology structure and pre-defining error tolerance thresholds for different directions before evaluating motion predictions. This preparation work is done once and reused across multiple predictions, reducing the computational complexity of individual evaluation operations while maintaining high safety standards
Solution Approach 2:
The patent introduces lane topology as an intermediary reference framework that mediates between raw prediction data and safety evaluation. By matching predictions against the predefined lane topology, the system translates complex spatial relationships into simpler compliance checks, reducing evaluation complexity while improving reliability
3Device complexity
If motion prediction models evaluate all trajectories without lane context, then the evaluation process is simpler, but the system cannot differentiate between safe and unsafe lateral deviations
Solution Approach 1:
The patent changes the evaluation parameters from generic distance metrics to lane-specific parameters. By incorporating lane topology information, the system evaluates predictions based on their relationship to defined road boundaries and safe zones, enabling precise differentiation between safe and unsafe lateral deviations while maintaining manageable complexity through parameterized evaluation rules
Data Source
AI summary
A computing system can receive motion prediction data from a vehicle, where the motion prediction data is generated by a motion prediction model executing on the vehicle. Based on the motion prediction data, the system can determine predicted trajectories for a plurality of entities in a surrounding environment of the vehicle. The system can evaluate a prediction performance of the motion prediction model by (i) matching, for each respective entity of the plurality of entities, a predicted endpoint of each predicted trajectory of the set of predicted trajectories to one or more underlying lanes of the road segment, (ii) matching a ground truth future position of the entity to one or more underlying lanes of the underlying lane topology, and (iii) determining a distance along one or more lane segments between the lane(s) matched to the predicted endpoint and the lane(s) matched to the ground truth future position.