Autonomous DNN Training With HD Map-Based Ground Truth Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches for training deep neural networks (DNNs) in autonomous vehicles rely heavily on manual generation of ground truth data, which is costly and prone to human error, leading to inaccurate performance when HD maps are unavailable or outdated.
Innovation Solution
Leverage high-definition (HD) map information to automatically generate training labels and annotations for DNNs, using localization techniques to correlate sensor data with map data, reducing manual effort and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to generate ground truth data, then the DNNs can be trained with accurate annotations, but the process requires substantial manual effort and time
Solution Approach 1:
The patent uses HD map data as a template or copy to automatically generate ground truth annotations for training images. Instead of manually creating annotations from scratch, the system copies relevant spatial and semantic information from the pre-existing HD map structure (lanes, road boundaries, intersections) and applies it to the corresponding sensor data, dramatically reducing labeling time while maintaining accuracy
Solution Approach 2:
The patent introduces HD map data as an intermediary between the raw sensor data and the ground truth annotations. The HD map serves as a mediator that provides structured environmental information, which is then correlated with sensor data through localization techniques to generate accurate annotations without requiring direct manual labeling of each image
2Measurement precision
If manual labeling is used to generate ground truth data, then the DNNs can be trained with accurate annotations, but the cost and development time increase significantly
Solution Approach 1:
The system copies structured environmental information from HD maps to generate training annotations automatically. This approach leverages the existing investment in HD map creation and reuses it for training purposes, eliminating the need for separate, costly manual labeling processes while maintaining high annotation quality
Solution Approach 2:
The system enables self-service annotation generation by automatically correlating sensor data with HD map information through localization. The process serves itself by using the HD map data structure to generate its own training annotations without requiring external manual intervention, reducing both time and cost
3Productivity
If HD map information is used to generate ground truth data, then the process becomes automated and faster, but the system requires accurate localization and map data
Solution Approach 1:
The patent performs preliminary localization of the vehicle within the HD map framework before generating ground truth data. By establishing the vehicle's position and orientation in advance relative to the map, the system creates a reliable reference frame that enables accurate automatic annotation generation without requiring complex real-time alignment during the labeling process
Data Source
AI summary
In various examples, training sensor data generated by one or more sensors of autonomous machines may be localized to high definition (HD) map data to augment and/or generate ground truth data—e.g., automatically, in embodiments. The ground truth data may be associated with the training sensor data for training one or more deep neural networks (DNNs) to compute outputs corresponding to autonomous machine operations-such as object or feature detection, road feature detection and classification, wait condition identification and classification, etc. As a result, the HD map data may be leveraged during training such that the DNNs—in deployment—may aid autonomous machines in navigating environments safely without relying on HD map data to do so.