Automated Training Data Extraction for Autonomous Driving Dynamic Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training dynamic models for autonomous driving vehicles face inefficiencies due to unbalanced training data and the time-consuming nature of manual data selection, which can lead to models with compromised performance and accuracy.
Innovation Solution
An automated closed-loop training process is implemented, where training data is extracted based on equally-spaced value ranges for each feature, and the dynamic model is iteratively retrained until it meets predetermined performance thresholds across controlled testing scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data is selected manually, then the quality and balance of training data can be improved, but the time and labor required for data preparation increases significantly
Solution Approach 1:
The system automatically selects and balances training data without human intervention. The automated data selection module analyzes driving scenarios, identifies underrepresented features, and retrieves additional training data autonomously, allowing the system to serve itself in the data preparation process while maintaining high model performance
Solution Approach 2:
The system implements a closed-loop feedback mechanism where model evaluation results are fed back to the data selection process. Performance metrics from evaluation scenarios automatically trigger targeted data retrieval for underperforming features, creating a self-correcting system that continuously improves model accuracy without manual intervention
2Productivity
If on-selected training data is used, then the data preparation process is faster, but the training data becomes unbalanced leading to compromised model performance
Solution Approach 1:
The system applies different data selection strategies to different features based on their specific needs. Instead of uniform random sampling, it identifies underrepresented features through evaluation feedback and retrieves targeted additional data for those specific features, ensuring each feature receives appropriate data balance while maintaining overall processing efficiency
Solution Approach 2:
The system dynamically adjusts data selection parameters based on model performance metrics. When evaluation reveals poor performance on specific features, it automatically modifies the data retrieval parameters to prioritize those features in subsequent training iterations, adapting the training process to actual model needs rather than using fixed sampling parameters
3Loss of time
If the dynamic model is trained with unbalanced data, then training speed is maintained, but the inference accuracy and robustness of the model deteriorates
Solution Approach 1:
The system implements dynamic data balancing where the training data composition changes based on real-time performance evaluation. The data selection process is not static but adapts across training iterations, automatically adjusting which features receive additional data based on current model performance, thereby maintaining training efficiency while progressively improving inference accuracy
Solution Approach 2:
The system performs preliminary evaluation of model performance on various driving scenarios before identifying data gaps. By evaluating the model on controlled test scenarios first, it can proactively determine which features need additional training data before full retraining begins, preventing accuracy deterioration rather than reacting to it after the fact
Data Source
AI summary
In one embodiment, a method of training dynamic models for autonomous driving vehicles includes the operations of receiving a first set of training data from a training data source, the first set of training data representing driving statistics for a first set of features; training a dynamic model based on the first set of training data for the first set of features; determining a second set of features as a subset of the first set of features based on evaluating the dynamic model, each of the second set of features representing a feature whose performance score is below a predetermined threshold. The method further includes the following operations for each of the second set of features: retrieving a second set of training data associated with the corresponding feature of the second set of features, and retraining the dynamic model using the second set of training data.