Driving Encounter Classification Using Paired Trajectory Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for classifying driving encounters between vehicles, particularly for autonomous vehicles, are limited as they primarily focus on individual trajectories rather than pairs, failing to fully represent all driving scenarios, which restricts their application in training autonomous vehicles effectively.
Innovation Solution
A method involving unsupervised learning using autoencoders to extract feature vectors from pairs of vehicle trajectories, followed by clustering using techniques like k-means, to categorize driving encounters into distinct groups, facilitating better decision-making for autonomous vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If clustering algorithms focus on individual trajectories, then computational complexity is reduced, but the ability to represent driving encounter scenarios is insufficient
Solution Approach 1:
The patent segments the trajectory data processing into two distinct levels: individual trajectory feature extraction and pair-wise encounter clustering. This segmentation allows the system to first process single trajectories independently (maintaining low computational complexity) and then combine them into encounter pairs (achieving comprehensive scenario representation). The feature extraction module processes each trajectory separately using GPS coordinates and temporal information, while the clustering module then groups these features into encounter scenarios.
Solution Approach 2:
The patent transitions from one-dimensional individual trajectory analysis to two-dimensional pair-wise encounter analysis by introducing a new dimension of interaction. Instead of merely clustering single trajectories in feature space, the system creates composite representations by pairing trajectories and clustering these pairs, thereby adding the dimension of inter-vehicle interaction to the analysis framework.
2Adaptability or versatility
If clustering algorithms process pairs of trajectories, then driving encounter scenarios are fully represented, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by extracting and storing feature vectors for individual trajectories before performing pair-wise clustering. This preprocessing step creates a library of standardized trajectory features that can be efficiently combined and compared later, avoiding the need to reprocess raw GPS data for each pair comparison and significantly reducing the computational burden of encounter-level clustering.
Solution Approach 2:
The system creates simplified copies of trajectory data in the form of feature vectors that capture essential characteristics (GPS coordinates, temporal information, kinematic parameters) without retaining the full complexity of raw trajectory datasets. These feature vector copies enable efficient comparison and clustering operations while preserving the meaningful patterns needed for scenario representation.
3Quantity of substance
If more trajectory data is collected from multiple vehicles, then the diversity of driving encounters is captured, but data processing difficulty increases
Solution Approach 1:
The patent extracts essential features from large volumes of trajectory data, isolating the most relevant characteristics (spatial coordinates, temporal stamps, velocity, acceleration) needed for encounter analysis. This extraction process filters out redundant information and focuses computational resources on the critical features that define driving scenarios, making processing of large datasets feasible.
Solution Approach 2:
The system transforms raw trajectory parameters into standardized feature vectors with consistent dimensions and scales. By normalizing GPS coordinates, time stamps, and kinematic parameters into uniform representations, the patent enables efficient comparison and clustering operations across diverse datasets from multiple vehicles, reducing processing difficulty while maintaining data integrity.
Data Source
AI summary
The present disclosure provides a method in a data processing system that includes at least one processor and at least one memory. The at least one memory includes instructions executed by the at least one processor to implement a driving encounter recognition system. The method includes receiving information, from one or more sensors coupled to a first vehicle, determining first trajectory information associated with the first vehicle and second trajectory information associated with a second vehicle, extracting a feature vector, providing the feature vector to a trained classifier, the classifier trained using unsupervised learning based on a plurality of feature vectors, and receiving, from the trained classifier, a classification of the current driving encounter in order to facilitate the first vehicle to perform a maneuver based on the current driving encounter.