Lane Change Prediction Model Using Merged Driving and Map Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lane change prediction models for automated driving vehicles face challenges due to limited driving data size and low quality predictions, often resulting in incorrect decisions, especially with high expense and legal issues associated with data collection.
Innovation Solution
A computer-implemented method and system for lane change prediction that collects raw driving data, extracts feature sets, automatically labels them with corresponding lane change information, and trains a prediction model using a combination of data from automated driving cars and map providers, enhancing data precision and reliability through LSTM model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If driving data is collected from HAD cars, then prediction accuracy can be improved, but data size remains limited and collection cost increases
Solution Approach 1:
The patent combines driving data from multiple HAD cars with map data from map providers to create a comprehensive training dataset. This merging of data sources increases the overall data size available for training the prediction model, directly addressing the limitation of limited data from individual HAD cars while maintaining high prediction accuracy
2Reliability
If driving data is collected from HAD cars, then prediction quality can be improved, but legal issues and collection expenses increase
Solution Approach 1:
The patent merges HAD car driving data with map provider data to achieve reliable lane change predictions. By combining these data sources, the system reduces dependence on expensive and legally complex HAD car data collection alone, while maintaining high prediction quality through the complementary information from map data
3Ease of operation
If experience models are used for lane change prediction, then implementation can be simplified, but prediction accuracy deteriorates
Solution Approach 1:
The patent changes the parameters used for lane change prediction from simple experience-based parameters (distance to lane border, yaw direction) to comprehensive features extracted from both HAD car driving data and map data. This parameter transformation enables the model to achieve high prediction accuracy while maintaining implementation feasibility through automated feature extraction and labeling
Data Source
AI summary
A method and a system for lane change prediction. The method includes collecting raw driving data, extracting a plurality of feature sets from the collected raw driving data, and obtaining corresponding lane change information. The lane change information indicates a status of lane change of a vehicle under each of the extracted feature sets. The method further includes automatically labeling each of the extracted plurality sets of features with the obtained corresponding lane change information. The method further includes training a lane change prediction model with the labeled plurality sets of features. Examples of the present disclosure further describe methods, systems, and vehicles for applying the lane change prediction model.