Fleet-Based Label Vectors for Autonomous Driving Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for autonomous vehicles lack reliable and efficient methods to determine uncertainty measures during training, relying on deterministic labels from unknown data distributions, leading to poor intrinsic uncertainty measures.
Innovation Solution
A method for generating a labeled data set using environmental data from a vehicle fleet, where label vectors with uncertainty measures are combined to form an aggregated label vector, incorporating Dempster-Shafer theory to represent uncertainty through probability masses, and storing these vectors in a database for training machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If deterministic labels from unknown data distributions are used during training, then the training process is simple, but the intrinsic uncertainty measures are poor
Solution Approach 1:
The patent transforms deterministic labels into probabilistic label vectors that incorporate uncertainty measures. By changing the parameter representation from single-value labels to multi-element probability distributions, the system maintains training simplicity while significantly improving uncertainty estimation quality through the aggregated uncertainty measures from multiple vehicles.
2Measurement precision
If additional ML algorithms are used to determine uncertainty measures, then the uncertainty measurement quality improves, but the computational complexity and latency increase
Solution Approach 1:
The patent performs uncertainty aggregation in advance during the data collection and labeling phase, before the actual machine learning training and inference. By pre-computing aggregated uncertainty measures from multiple vehicles' label vectors and storing them with training data, the system eliminates the need for additional complex uncertainty algorithms during real-time operation, thus improving uncertainty quality without increasing runtime computational complexity.
3Device complexity
If deterministic labels are used, then the data structure is simple, but the information about data distribution uncertainty is lost
Solution Approach 1:
The patent merges label vectors from multiple vehicles by linearly combining their probability mass distributions. This consolidation approach integrates uncertainty information from diverse sources while maintaining a unified data structure. The aggregated label vector preserves and synthesizes uncertainty information that would be lost in deterministic labels, creating a richer yet still manageable data representation.
Data Source
Figure 1
Figure 2
Figure 2
AI summary
The present invention relates to a method for generating a labeled data set for an environment model. Environmental data is acquired (11) by several vehicles (F1, F2, FX) of a vehicle fleet, wherein the environmental data (UD1, UD2, UDx) is acquired by sensors (K1, K2, Kx) of the respective vehicles and relates to environmental objects and/or environmental parameters in the vehicle environment of the respective vehicles. Label vectors (LV1, LV2, LVx) are generated (12) for data points in the respectively acquired environmental data (UD1, UD2, UDx), wherein a label vector for classifying the data points into n different classes has a set of 2^n possible elements resulting from the possible combinations of the classes, and wherein a acquired data point is provided with an uncertainty measure by assigning values for probability masses to one or more of the possible elements.The label vectors (LV1, LV2, LVx) are combined into an aggregated label vector for a data point by linearly combining the transmitted label vectors (14). The aggregated label vector or information derived from the aggregated label vector is stored together with the corresponding data point in a database (DB) (15).