Autonomous Vehicle Model Validation Using Low-Discrepancy Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training and validating autonomous driving models face challenges in ensuring evenly distributed training data, leading to inadequate model performance in various driving scenarios due to variations in data density.
Innovation Solution
The use of low-discrepancy sequences to map data samples into a multidimensional space, allowing for the selection of a training data set that is evenly distributed across different scenarios, and the generation of synthetic data to fill gaps in the training set, ensuring comprehensive model training and validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional random sampling methods are used to select training data, then data selection is simple and fast, but the training data distribution is uneven leading to inadequate model performance in various driving scenarios
Solution Approach 1:
The patent transforms the data selection process by mapping training data to low-discrepancy sequences in a multidimensional space. This changes the selection criterion from random sampling to sequence-based selection, ensuring even distribution across driving scenarios while maintaining computational feasibility through the structured nature of low-discrepancy sequences
Solution Approach 2:
The patent replaces traditional random sampling mechanisms with a deterministic low-discrepancy sequence-based selection mechanism. This substitution eliminates the randomness that causes uneven distribution while maintaining computational efficiency, as low-discrepancy sequences can be generated and processed algorithmically without requiring complex optimization routines
2Reliability
If more training data is collected to cover all driving scenarios, then model performance improves, but data collection time and storage requirements increase
Solution Approach 1:
The patent extracts only the essential and representative training data needed for model training by using low-discrepancy sequences to identify and select diverse driving scenarios. This extraction approach avoids collecting redundant data while ensuring comprehensive scenario coverage, reducing both data collection time and storage requirements
Solution Approach 2:
The patent applies partial action by selecting a specific number of training data points corresponding to the length of the low-discrepancy sequence, rather than collecting exhaustive data for all possible scenarios. This partial sampling through low-discrepancy sequences achieves sufficient model performance without the time and storage costs of complete scenario coverage
Data Source
AI summary
Autonomous vehicle model training and validation using low-discrepancy sequences may include: generating a low-discrepancy sequence in a multidimensional space comprising a plurality of multidimensional points; mapping each sample of a plurality of samples of a data corpus to a corresponding entry in the low-discrepancy sequence, wherein each sample of the plurality of samples comprises one or more environmental descriptors for an environment relative to a vehicle and one or more state descriptors describing a state of the vehicle; selecting, from the data corpus, a training data set by selecting, for each multidimensional point of the low-discrepancy sequence having one or more mapped samples, a mapped sample for inclusion in the training data set; and training one or more models used to generate autonomous driving decisions of an autonomous vehicle based on the selected training data set.