Autonomous Vehicle Motion Forecasting With Actor-Map Interaction Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion forecasting techniques for autonomous vehicles fail to accurately predict the motion of actors within a surrounding environment, particularly in highway scenarios, due to the use of single vectors that compress context information and ignore lane information, leading to unrealistic trajectory predictions and inefficient resource usage.
Innovation Solution
A graph-centric motion forecasting model framework that constructs actor-specific graphs with lane-specific nodes and edges, projecting these graphs to a global graph for interaction propagation, preserving map structure and enabling more accurate and efficient motion prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single vector is used to compress context information for motion forecasting, then computational resources are reduced, but prediction accuracy deteriorates due to loss of fine-grained information
Solution Approach 1:
The patent segments the context information into discrete lane segments that form a graph structure. Each lane segment becomes a node in the graph, preserving fine-grained spatial information while organizing it in a computationally efficient manner. This segmentation allows the system to process contextual information at multiple levels of detail rather than compressing everything into a single vector.
Solution Approach 2:
The patent transitions from a one-dimensional vector representation to a two-dimensional graph structure with nodes and edges. This dimensional change allows the system to preserve spatial relationships and fine-grained information while maintaining computational efficiency through the graph's structured organization.
2Device complexity
If lane information is ignored in motion forecasting, then model complexity is reduced, but trajectory realism deteriorates due to failure to capture map structure
Solution Approach 1:
The patent performs preliminary action by pre-processing map data into a graph structure that encodes lane topology and spatial relationships before the motion forecasting process. This pre-structured representation of lane information is then integrated with actor motion data, allowing the model to leverage map structure without adding significant complexity during the forecasting step.
Solution Approach 2:
The patent introduces a graph structure as an intermediary between raw map data and the motion forecasting model. This graph acts as a mediator that captures essential lane information and spatial relationships in a form that is computationally efficient to process while still providing rich contextual cues for realistic trajectory prediction.
3Device complexity
If a fully-connected graph among all actors is built without lane information, then interaction modeling is simplified, but map structure utilization is lost
Solution Approach 1:
The patent applies local quality by making the graph connections actor-specific rather than universally fully-connected. Each actor has a personalized graph that includes only the lane segments and map features locally relevant to its position and motion. This localized approach reduces overall graph complexity while preserving important map structure information where it matters for each individual actor.
Data Source
AI summary
Systems and methods are provided for forecasting the motion of actors within a surrounding environment of an autonomous platform. For example, a computing system of an autonomous platform can use machine-learned model(s) to generate actor-specific graphs with past motions of actors and the local map topology. The computing system can project the actor-specific graphs of all actors to a global graph. The global graph can allow the computing system to determine which actors may interact with one another by propagating information over the global graph. The computing system can distribute the interactions determined using the global graph to the individual actor-specific graphs. The computing system can then predict a motion trajectory for an actor based on the associated actor-specific graph, which captures the actor-to-actor interactions and actor-to-map relations.