Boundary-Aware Object Trajectory Prediction Near Road Edges
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in predicting the future locations of objects in complex environments, particularly in congested areas with multiple moving vehicles and static obstacles, where existing methods often result in unrealistic trajectories and increased computational resources.
Innovation Solution
The use of a graph neural network (GNN) that combines sensor data and map data to predict object positions, velocities, and trajectories by encoding features into nodes and edges, allowing for efficient sampling of distribution data and focusing on realistic scenarios that adhere to environmental rules and interactions between objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing prediction methods are used to predict object positions in congested areas, then predictive coverage is achieved, but unrealistic trajectories are generated and computational resources increase
Solution Approach 1:
The patent segments the prediction task by separating boundary objects from non-boundary objects. Boundary objects (those near environment boundaries like road edges) are handled with specialized boundary-aware prediction logic, while non-boundary objects use standard prediction methods. This segmentation prevents unrealistic trajectories for boundary objects without requiring complex computations for all objects, thus improving trajectory realism while maintaining computational efficiency.
Solution Approach 2:
The patent applies local quality by using different prediction strategies for different spatial locations. Objects near environment boundaries receive boundary-aware prediction that incorporates environmental constraints, while objects in open spaces use conventional prediction. This localized approach ensures realistic trajectories only where needed (at boundaries) rather than applying computationally intensive methods uniformly across all objects.
2Reliability
If boundary-aware prediction is implemented to improve trajectory realism, then environmental constraints are satisfied, but system complexity increases
Solution Approach 1:
The prediction system is segmented into boundary-aware and non-boundary components. The system identifies which objects are near environmental boundaries and applies boundary-aware prediction only to those objects. This segmentation reduces overall system complexity by limiting the application of complex boundary-aware logic to only the necessary subset of objects, rather than implementing it for all objects in the scene.
Solution Approach 2:
The patent introduces an intermediary boundary awareness mechanism that mediates between standard prediction outputs and environmental constraints. This intermediary layer checks predicted trajectories for boundary objects and adjusts them to satisfy environmental constraints without requiring complete redesign of the prediction system, thus managing complexity while improving constraint satisfaction.
3Productivity
If comprehensive object prediction is performed without boundary awareness, then all objects are tracked, but trajectories violate environmental constraints
Solution Approach 1:
The patent segments objects into boundary and non-boundary categories, applying appropriate prediction methods to each segment. This ensures comprehensive tracking of all objects while maintaining constraint adherence for boundary objects through specialized boundary-aware prediction logic.
Solution Approach 2:
The patent applies local quality by enhancing prediction accuracy and constraint adherence specifically for boundary objects where environmental constraints are most relevant, while using standard prediction for non-boundary objects. This localized enhancement maintains overall prediction coverage while ensuring constraint satisfaction where needed.
Data Source
AI summary
Techniques for determining unified futures of objects in an environment are discussed herein. Techniques may include determining a first feature associated with an object in an environment and a second feature associated with the environment and based on a position of the object in the environment, updating a graph neural network (GNN) to encode the first feature and second feature into a graph node representing the object and encode relative positions of additional objects in the environment into one or more edges attached to the node. The GNN may be decoded to determine a first predicted position of the object. The first predicted position may be determined to be outside of a bounded area of the environment. Based on this determination, a second predicted position of the object may be determined using map data associated with the object.