Reference Trajectory Generation by Clustering Natural Vehicle Paths

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing reference trajectory generation methods may produce inappropriate reference trajectories due to the inclusion of unnatural vehicle trajectories, which can affect autonomous driving systems, as these trajectories are not representative of normal vehicle travel patterns.

Innovation Solution

A reference trajectory generating device that classifies vehicle trajectories through a predetermined road section by clustering, selects the class with the most trajectories, and generates a reference trajectory by averaging individual trajectories within the selected class, excluding those with abnormal acceleration and prohibited area entries.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If all vehicle trajectories are used for generating reference trajectory, then more trajectory data is available for processing, but the reference trajectory becomes inappropriate due to inclusion of unnatural trajectories

Engineering Contradiction:
Improvenumber of trajectory dataVSAvoidquality of reference trajectory
Core Design Contradiction:
Quantity of substanceVSReliability

Solution Approach 1:

The patent extracts and removes unnatural trajectories from the dataset by calculating curvature values for each trajectory and comparing them against reference curvature values. Trajectories with curvature deviations exceeding a threshold are identified and excluded, leaving only natural trajectories for reference trajectory generation. This extraction process ensures that the reference trajectory is generated from high-quality data while maintaining sufficient data quantity.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different quality standards to different segments of trajectories by calculating local curvature values at multiple points along each trajectory. Each trajectory segment is evaluated individually against the reference curvature, allowing selective filtering of unnatural segments while preserving natural segments. This local quality assessment enables precise identification and removal of abnormal trajectory portions.

Inventive Principle:
Principle #3Local quality

2Reliability

If clustering is applied to classify trajectories, then natural trajectories can be separated from unnatural ones, but the processing complexity increases

Engineering Contradiction:
Improveaccuracy of trajectory classificationVSAvoidcomplexity of processing system
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent transforms trajectory data into a different parameter space by calculating curvature values instead of working directly with raw coordinate data. This parameter transformation simplifies the classification process by converting complex trajectory shapes into comparable curvature metrics. The curvature parameter serves as a feature that naturally separates natural from unnatural trajectories, reducing the complexity of the clustering algorithm needed.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4411320A1Reference trajectory generating device, method, and computer program for generating reference trajectory
Publication Date: 2024.08.07 WOVEN BY TOYOTA INC
  • EP4411320A1 patent drawingFigure 1
  • EP4411320A1 patent drawingFigure 2
  • EP4411320A1 patent drawingFigure 3

AI summary

A reference trajectory generating device includes a classification unit 41 that classifies a plurality of trajectories of travel of at least one vehicle 2 through a predetermined section of a road into a plurality of classes by clustering of the trajectories in the predetermined section; a selection unit 42 that selects a class including the most trajectories of the classes; and a reference trajectory generation unit 43 that generates a reference trajectory serving as a reference in the predetermined section by averaging individual trajectories included in the class selected from the classes.