Map Updating via Auto-Labeled Sensor Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for creating three-dimensional maps for autonomous vehicle navigation are time-consuming and labor-intensive due to the need for manual labeling of features, making it impractical to generate and update maps efficiently, especially in unmapped sections and varying environmental conditions.
Innovation Solution
A system and method for automatically creating training data for machine learning networks to segment and classify sensor-generated data, using existing map data to project features onto sensor data, allowing for the creation of updated maps with minimal human intervention, and enabling the generation of semantic labels for objects and features like lanes, traffic signs, and road geometry.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling methods are used to create training data for segmentation networks, then classification accuracy may be improved, but the time and labor required becomes impractically large
Solution Approach 1:
The patent uses existing map data as a template or copy to generate synthetic training data. Instead of manually labeling real sensor data, the system projects map features onto sensor data to create labeled training examples automatically, significantly reducing manual labeling time while maintaining classification accuracy
Solution Approach 2:
The system performs self-labeling by using its own map data to automatically generate training labels. The map data serves as ground truth to automatically annotate sensor data, eliminating the need for external manual labeling while maintaining consistent and accurate classifications
2Manufacturing precision
If more manually labeled training data is collected to improve segmentation of difficult features, then segmentation quality improves, but computational resources and memory capacity are excessively consumed
Solution Approach 1:
The patent generates synthetic training data by copying and projecting map features onto sensor data. This creates unlimited labeled training examples without requiring proportional increases in computational resources, as the labeling is done through geometric projection rather than intensive manual annotation
Solution Approach 2:
The system performs preliminary actions by pre-processing sensor data with projected map features before actual segmentation. This preliminary projection creates pre-labeled data that guides the segmentation process, improving quality while reducing the computational burden during actual segmentation operations
3Adaptability or versatility
If manual creation of training data for different environmental conditions is performed, then adaptability to various conditions improves, but the process becomes extremely time-consuming and resource-intensive
Solution Approach 1:
The patent creates a universal training data generation system that works across all environmental conditions using the same map projection methodology. The system generates condition-specific training data on-demand by applying the same underlying principle of projecting map features onto sensor data, maintaining productivity while achieving broad adaptability
Solution Approach 2:
The system dynamically generates training data adapted to specific environmental conditions when needed, rather than statically pre-collecting all possible condition data. This allows the system to maintain high productivity by generating only the specific training data required for current operational conditions
Data Source
AI summary
A system may automatically create training datasets for training a segmentation model to recognize features such as lanes on a road. The system may receive sensor data representative of a portion of an environment and map data from a map data store including existing map data for the portion of the environment that includes features present in that portion of the environment. The system may project or overlay the features onto the sensor data to create training datasets for training the segmentation model, which may be a neural network. The training datasets may be communicated to the segmentation model to train the segmentation model to segment data associated with similar features present in different sensor data. The trained segmentation model may be used to update the map data store, and may be used to segment sensor data obtained from other portions of the environment, such as portions not previously mapped.