Road Graph Neural Networks for Drivable Path Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in processing vast volumes of data for navigation, including image data, map data, and sensor data, which can limit their ability to safely and accurately navigate roadways, and traditional mapping technologies face daunting challenges in data storage and updates.
Innovation Solution
A system using a graph neural network to predict drivable paths by encoding actual vehicle trajectories and topographical representations as a grid of nodes, identifying drivable paths between nodes, and storing this information in navigational maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional mapping technology is used to navigate, then navigation functionality is provided, but the sheer volume of data needed to store and update the map poses daunting challenges
Solution Approach 1:
The patent segments the continuous road environment into discrete graph nodes and edges. Instead of storing comprehensive map data for entire road networks, the system represents roads as graphs where intersections and key points are nodes, and road segments are edges. This segmentation dramatically reduces the data volume required for navigation while maintaining essential routing functionality.
Solution Approach 2:
The patent extracts only the essential navigational elements from complete map data. By identifying and extracting key features such as drivable paths, obstacles, and road topology, the system creates a simplified graph representation that retains navigation capability while eliminating redundant data, thus reducing storage requirements.
2Measurement precision
If vast volumes of data are collected and analyzed for navigation, then navigation accuracy is improved, but the challenges limit or adversely affect autonomous navigation
Solution Approach 1:
The patent replaces traditional mechanical data processing systems with graph neural networks. Instead of using conventional algorithms to process vast volumes of sensor data, the system employs GNNs that naturally operate on graph structures. This substitution simplifies the processing architecture by aligning the data representation (graph) with the processing method (GNN), reducing computational complexity while maintaining high navigation precision.
Solution Approach 2:
The patent changes the parameter representation from continuous sensor data to discrete graph features. By transforming complex sensor inputs into graph node and edge attributes, the system reduces data dimensionality and simplifies processing. The graph representation captures essential spatial relationships with fewer parameters, making the navigation system more manageable while preserving accuracy.
3Adaptability or versatility
If complete map data is stored for all road segments, then comprehensive navigation coverage is achieved, but storage and update challenges increase
Solution Approach 1:
The patent performs preliminary action by pre-processing road data into graph structures during map creation. Instead of storing raw sensor data or detailed map information that requires extensive processing and updating, the system pre-computes the graph representation with nodes, edges, and essential attributes. This preliminary transformation simplifies subsequent updates and maintains comprehensive navigation coverage while reducing the effort required for map maintenance.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Systems and methods for predicting drivable paths relative to road segments are disclosed. In one implementation, a system includes a processor programmed to access topographical information associated with a road segment; generate a topographical representation of the road segment based on the topographical information; input the topographical representation of the road segment to a trained model, wherein the trained model includes a graph neural network and is configured to predict at least one drivable path relative to the road segment based on the topographical representation of the road segment; receive, from the trained model, information identifying the drivable path; and store the information identifying the drivable path in a map.