Actor Motion Forecasting Using Graph Neural Networks for AVs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately predicting the complex interactions between various third-party actors in their environment, such as vehicles, pedestrians, and cyclists, which are crucial for safe navigation.
Innovation Solution
A computing system utilizing a graph neural network to model the spatial relationships and interactions between actors, incorporating sensor and map data to generate motion forecast data, and iteratively update node states to predict future trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional processing techniques are used to predict future circumstances, then the system can operate with simpler computational resources, but the measurement precision and reliability of predicting complex actor interactions deteriorates
Solution Approach 1:
The system segments the prediction task into two distinct stages: (1) an object detection model that identifies and locates actors in the environment, and (2) a graph neural network that specifically models interactions between detected actors. This segmentation allows each component to specialize in one aspect of the prediction task, improving overall accuracy while managing computational complexity through modular design.
Solution Approach 2:
The patent introduces an intermediary representation - the graph structure with nodes and edges - that mediates between raw sensor data and final motion forecasts. The graph neural network uses this intermediate structure to efficiently capture and process interaction relationships between actors, transforming complex multi-body interactions into a structured computational framework that improves prediction reliability.
2Reliability
If complex models are used to accurately model actor interactions, then the reliability of interaction prediction improves, but the computational resources and processing time increase
Solution Approach 1:
The system performs preliminary object detection and actor identification before engaging the computationally intensive graph neural network. By pre-processing the sensor data to extract only relevant actor information and relationships, the system reduces the input complexity for the interaction prediction model, thereby lowering computational resource consumption while maintaining reliable interaction predictions.
Solution Approach 2:
The graph neural network focuses computational resources only on modeling interactions between detected actors rather than processing all environmental data. This partial action approach applies complex computational models only where necessary (between actors), while simpler processing is used for non-actor elements, optimizing the balance between prediction reliability and computational efficiency.
Data Source
AI summary
Systems and methods for generating motion forecast data for actors with respect to an autonomous vehicle and training a machine learned model for the same are disclosed. The computing system can include an object detection model and a graph neural network including a plurality of nodes and a plurality of edges. The computing system can be configured to input sensor data into the object detection model; receive object detection data describing the location of the plurality of the actors relative to the autonomous vehicle as an output of the object detection model; input the object detection data into the graph neural network; iteratively update a plurality of node states respectively associated with the plurality of nodes; and receive, as an output of the graph neural network, the motion forecast data with respect to the plurality of actors.