Driving Recommendation System Using Graph Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current connected-vehicle technologies face challenges in accurately predicting the movement of surrounding vehicles, especially at a distance, which affects the effectiveness of driving recommendations for both human drivers and autonomous systems.
Innovation Solution
A system utilizing graph convolutional networks and recurrent neural networks to construct spatial relationships among vehicles and predict acceleration, incorporating probabilistic mixture models to enhance prediction accuracy, allowing for timely and relevant driving recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If connected-vehicle technology is used to provide driving recommendations, then drivers receive information about surrounding vehicles, but the accuracy of predicting vehicle movement especially at a distance deteriorates
Solution Approach 1:
The patent introduces an intermediary prediction system that processes vehicle data through multiple computational layers. The system uses a first prediction model to predict movements of nearby vehicles and a second prediction model for distant vehicles, with a determination module that selectively applies the appropriate model based on distance thresholds. This intermediary processing layer resolves the contradiction by adapting the prediction approach to the specific distance context, maintaining information flow while improving prediction accuracy for different spatial zones.
2Ease of operation
If the system predicts acceleration of surrounding vehicles, then driving recommendations can be generated, but the complexity of the prediction system increases
Solution Approach 1:
The patent segments the prediction system into distinct functional components: a first prediction model for nearby vehicles, a second prediction model for distant vehicles, and a determination module that routes between them. This segmentation allows each component to be optimized independently and simplifies the overall control logic by using clear distance-based thresholds to determine which model to apply, thereby managing complexity while enabling comprehensive driving recommendations.
Solution Approach 2:
The system changes the parameter of prediction accuracy by selecting different models based on the distance parameter. When the distance to a vehicle exceeds a threshold, the system switches from the first prediction model to the second prediction model. This parameter-based model selection resolves the complexity issue by providing a simple, rule-based decision mechanism that adapts the system's behavior to the operational context without requiring complex real-time analysis.
Data Source
AI summary
Systems and methods for providing driving recommendations are disclosed herein. One embodiment receives, at an ego vehicle, first vehicle data and first encoded information from one or more other vehicles; constructs, from the first vehicle data, graph data representing how the ego vehicle and the one or more other vehicles are spatially related; inputs the graph data, the first vehicle data, second vehicle data pertaining to the ego vehicle, and the first encoded information to a graph convolutional network that outputs second encoded information; inputs the second encoded information and previously stored encoded information to a recurrent neural network that outputs a set of parameters to a mixture model; predicts acceleration of the one or more other vehicles using the mixture model; and generates a driving recommendation for the ego vehicle based, at least in part, on the predicted acceleration of the one or more other vehicles.