HD Map Generation by Aggregating Vehicle Trails in Grid Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating and maintaining high-definition (HD) maps for advanced driver assistance systems (ADAS) and autonomous driving (AD) are costly, time-consuming, and lack scalability, with conventional approaches often relying on high-cost sensors, manual annotation, or requiring annotated HD map data for training deep neural networks, and are not suitable for complex scenarios.
Innovation Solution
A method involving ADAS vehicles aggregating individual trails of road users using sensor data to generate HD maps, including classifying and associating trails with density maps, extracting lane centers, and representing them in a directed graph, which can be updated dynamically through a cloud server.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation methods are used to generate HD maps, then map accuracy is improved, but cost and time consumption increase significantly
Solution Approach 1:
The system enables vehicles to automatically generate and update HD maps using their own sensor data and localization information without requiring manual annotation. The map data is crowd-sourced from multiple vehicles traversing the same road segments, allowing the system to self-update and maintain accuracy autonomously
Solution Approach 2:
Instead of manually creating map data, the system copies and aggregates trajectory information from multiple vehicles that have already traversed the road segments. This copying approach allows rapid replication of accurate map data across the fleet without repeating manual annotation work
2Measurement precision
If multiple surveillance vehicles with high-cost sensors are deployed to map roads, then map detail and accuracy are improved, but system cost increases
Solution Approach 1:
The system makes vehicle trajectory data serve multiple purposes: it is used for navigation, for generating HD maps, and for updating map data. The same sensor data that vehicles collect for their primary navigation function is reused to contribute to collective map generation, eliminating the need for dedicated surveillance vehicles
Solution Approach 2:
The system copies trajectory data from vehicles performing their primary navigation function and uses this copied data for map generation. This avoids the need for separate high-cost surveillance vehicles, as the map data is derived from copying existing operational data
3Reliability
If HD maps are updated frequently to reflect environmental changes, then map currency is improved, but data processing complexity increases
Solution Approach 1:
The system continuously updates HD maps as vehicles traverse road segments in real-time. Rather than periodic batch updates, the map data is continuously refined with each new trajectory observation, ensuring maps remain current without requiring complex scheduled processing cycles
Solution Approach 2:
The system uses feedback from multiple vehicle trajectories to automatically detect and correct map discrepancies. When new trajectory data conflicts with existing map data, the system processes this feedback to update the map, maintaining currency through automated feedback loops rather than complex manual review processes
4Productivity
If crowd-sourced GPS trajectory data is used to generate maps, then scalability is improved, but map precision deteriorates due to lack of detailed sensor information
Solution Approach 1:
The system merges multiple data sources including GPS trajectories, sensor data from ADAS vehicles (cameras, radars, LiDARs), and localization information into a unified HD map. This combination preserves the scalability of crowd-sourced GPS data while enhancing precision through additional sensor measurements from vehicles already equipped with these sensors for their primary functions
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Provided is a computer-implemented method of generating a map, comprising the steps of: acquiring sensor-based data of a sensor unit of at least one vehicle and localization data related to a localization unit of the at least one vehicle; determining a first trail of the at least one vehicle using the localization data and a plurality of second trails of other vehicles using the sensor-based data; and aggregating the first trail and/or the plurality of second trails to generate aggregated trails in a grid map representation of the map.