Vehicle Trajectory Prediction Using RNN-GNN Interaction Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current trajectory prediction methods for autonomous vehicles are limited in their ability to accurately predict future vehicle locations, especially in complex interactive driving scenarios with varying numbers of surrounding vehicles, as they often rely on fixed numbers of predicted motions and do not effectively integrate the dynamics of individual vehicles and road structures.
Innovation Solution
A graph-neural-network-based (GNN) deep learning system that uses recurrent neural networks (RNNs) to extract dynamic features from historical vehicle data and represents inter-vehicular interactions as a graph, allowing for the prediction of multiple, variable future trajectories by considering the dynamics and interactions of all vehicles and the road structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed number of predicted motions are used, then the model structure is simplified and computation is faster, but the generalizability to complex map geometries deteriorates
Solution Approach 1:
The patent applies dynamics by making the number of predicted trajectories variable rather than fixed. The model adapts the number of output trajectories based on the complexity of the map geometry and traffic situation, allowing it to generalize better to complex scenarios while maintaining computational efficiency in simpler cases. This is achieved through a dynamic trajectory generation mechanism that adjusts prediction quantity based on environmental complexity.
Solution Approach 2:
The patent changes the parameter of trajectory count from a fixed value to a variable parameter that can be adjusted based on map geometry complexity and traffic conditions. This allows the model to optimize between computational efficiency and prediction accuracy dynamically, resolving the contradiction between simplified structure and generalizability.
2Measurement precision
If interaction-aware methods are used to consider interactions among many objects, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the computational task by processing each object's trajectory independently through RNN encoders, then combining them through a structured interaction model. This segmentation allows the system to handle multiple objects efficiently by breaking down the complex interaction problem into manageable individual processing units that can be computed in parallel, reducing overall computational complexity while maintaining prediction accuracy.
Solution Approach 2:
The patent employs universal RNN encoder and decoder modules that can process any number of objects with the same architecture. These multi-functional components handle both individual object dynamics and inter-object interactions within a unified framework, eliminating the need for separate processing pipelines and reducing computational overhead while maintaining high prediction accuracy.
3Device complexity
If physics-based methods assuming constant velocity or acceleration are used, then computational simplicity is maintained, but accuracy in complex interactive driving scenarios deteriorates
Solution Approach 1:
The patent replaces traditional physics-based mechanical models (constant velocity/acceleration assumptions) with data-driven RNN encoders that learn complex motion patterns from historical trajectory data. This substitution allows the system to capture realistic driving behaviors including acceleration, deceleration, lane changes, and interaction responses without requiring complex explicit physics calculations, achieving both accuracy and computational efficiency.
Data Source
AI summary
A method of determining a predicted trajectory of a moving object. The method comprises obtaining historical trajectory data for the moving object and for one or more neighbouring objects; passing the historical trajectory data to a RNN encoder to generate dynamic features for the moving object and the one or more neighbouring objects; constructing a graph representing interactions between the moving object and the one or more neighbouring objects, wherein each node of the graph repre-sents one of the moving object or one of neighbouring objects, and comprises the respective dynamic features of the moving object or the one or more neighbouring objects, and each edge represents an effect of the moving object on a neighbouring object or vice versa, or an effect of a neighbouring object on another neighbouring object; passing the graph and the dynamic features to a GNN encoder to generate a plurality of interaction features; and passing the dynamic features and the interaction features to a RNN decoder to generate the predicted trajectory.