Vehicle Motion Scoring Using Variation Metrics for Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training autonomous driving models face challenges due to varying vehicle motion among individual drivers and conditions, and manual scoring of vehicle motion is labor-intensive.
Innovation Solution
A vehicle motion scoring device that calculates the degree of variations and maximum of motion indices such as acceleration, deceleration, and travel direction changes to automatically assess the suitability of vehicle motion information for training, setting lower scores for higher variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If vehicle motion information from individual drivers is used for training, then the model can learn diverse driving characteristics, but the variation in motion data reduces training quality
Solution Approach 1:
The patent changes the parameter of motion data by calculating statistical metrics (variation degree, maximum values) to transform raw motion information into scored data that reflects quality, enabling the system to select appropriate training samples based on quantitative thresholds
Solution Approach 2:
The system implements feedback by using the calculated score to determine whether motion information should be used for training, creating a closed-loop mechanism where motion data is evaluated and then selected or rejected based on the evaluation results
2Measurement precision
If manual scoring of vehicle motion information is performed, then accurate assessment of training suitability is achieved, but countless man-hours are required
Solution Approach 1:
The patent replaces the mechanical manual scoring process with an automated computational system that calculates variation degrees and maximum values of motion indices, substituting human labor with algorithmic processing to achieve both accuracy and efficiency
Solution Approach 2:
The system performs self-service by automatically evaluating its own motion data through calculated metrics, enabling the vehicle motion scoring device to assess training suitability without external human intervention
3Quantity of substance
If all vehicle motion information is used for training, then sufficient training data is available, but unsuitable motion data reduces model performance
Solution Approach 1:
The patent applies local quality by differentiating between suitable and unsuitable motion data within the overall dataset, assigning different qualities (scores) to different portions of the data based on local characteristics such as variation degree and maximum values
Data Source
AI summary
A vehicle motion scoring device includes a processor configured to calculate at least one of a degree of variations in a distribution of a motion index or a maximum of the motion index, based on vehicle motion information indicating motion of a vehicle traveling along a predetermined road section, and set a lower score to the vehicle motion information as a calculated value of the degree of variations in the distribution of the motion index or the maximum of the motion index is greater. The motion index includes at least one of acceleration or deceleration of the vehicle, an amount of change in the acceleration or deceleration of the vehicle per unit time, an amount of change in a travel direction of the vehicle, or an amount of change in the travel direction of the vehicle per unit time.