Autonomous Vehicle Motion Forecasting With Joint Scenario Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle forecasting systems fail to accurately predict diverse future traffic scenarios due to their focus on marginal distributions of individual agent trajectories, leading to scene inconsistencies and high computational resource usage.
Innovation Solution
The proposed system models the joint distribution of actor trajectories using a graph neural network to determine a compact set of diverse future scenarios, incorporating a contingency planner that optimizes motion planning with a diversity objective and reduces resource usage by employing a shared noise mapping across latent variables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte-Carlo sampling of latent variables is used to accurately reflect the state of prediction, then prediction accuracy is improved, but computational resource usage increases prohibitively and latency increases
Solution Approach 1:
The patent applies partial action by using a limited number of carefully selected diverse samples rather than exhaustive Monte-Carlo sampling. The system generates a compact set of diverse future scenarios that cover the most important possibilities without requiring prohibitively large numbers of samples, achieving adequate prediction accuracy with reduced computational resources.
Solution Approach 2:
The patent changes the approach from sampling-based estimation to direct optimization of diversity objectives. By formulating the problem as an optimization task that directly maximizes scenario diversity while maintaining accuracy, the system avoids the computational burden of extensive Monte-Carlo sampling while preserving prediction quality.
2Device complexity
If marginal distributions of individual agent trajectories are modeled, then computational complexity is reduced, but scene consistency deteriorates and interaction predictions become inaccurate
Solution Approach 1:
The patent merges the modeling of multiple agent trajectories into a joint distribution framework. Instead of treating each agent independently as in marginal distribution approaches, the system models the combined behavior of all agents together, capturing their interactions and maintaining scene consistency while managing computational complexity through efficient optimization techniques.
Solution Approach 2:
The patent introduces latent variables as intermediaries that encode unobserved scene dynamics. These latent variables serve as a bridge between individual agent observations and joint trajectory predictions, enabling the model to capture interactions and maintain scene consistency without directly computing the full joint distribution, thus managing computational complexity.
3Adaptability or versatility
If a large number of samples are generated to cover diverse future scenarios, then scenario diversity is improved, but computational resource usage increases and latency increases
Solution Approach 1:
The patent generates a compact set of diverse scenarios that covers the essential range of future possibilities without requiring a large number of samples. By focusing on quality and diversity of scenarios rather than quantity, the system achieves adequate scenario coverage with reduced computational resources and lower latency.
Solution Approach 2:
The patent performs preliminary optimization to identify and generate only the most diverse and relevant future scenarios before they are needed for planning. By pre-selecting and optimizing the scenario set to maximize diversity objectives, the system reduces the number of scenarios that need to be processed later, thereby reducing latency without sacrificing scenario diversity.
Data Source
AI summary
Systems and methods are disclosed for motion forecasting and planning for autonomous vehicles. For example, a plurality of future traffic scenarios are determined by modeling a joint distribution of actor trajectories for a plurality of actors, as opposed to an approach that models actors individually. As another example, a diversity objective is evaluated that rewards sampling of the future traffic scenarios that require distinct reactions from the autonomous vehicle. An estimated probability for the plurality of future traffic scenarios can be determined and used to generate a contingency plan for motion of the autonomous vehicle. The contingency plan can include at least one initial short-term trajectory intended for immediate action of the AV and a plurality of subsequent long-term trajectories associated with the plurality of future traffic scenarios.