Autonomous Vehicle Policy Planning with Reactive Behavior Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous vehicle systems face challenges in determining drivable paths in complex environments due to conservative motion planning that ignores new information and lacks multistage reasoning, leading to inefficient trajectory planning and collision avoidance.
Innovation Solution
The system employs policy planning using behavior models that generate multistage trajectories and scenario trees based on future reactive behaviors of other objects, allowing for dynamic navigation and improved collision avoidance through deep-learning prediction models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional systems use single-stage reasoning to determine trajectories, then the system complexity is reduced, but the navigation safety and efficiency deteriorate due to inability to account for future reactive behaviors
Solution Approach 1:
The patent segments the trajectory planning into multiple stages: first generating candidate trajectories, then for each candidate generating scenario trees that model future reactive behaviors of other objects, and finally evaluating and selecting trajectories based on these multi-stage predictions. This segmentation allows the system to account for future reactive behaviors while maintaining manageable computational complexity through structured decomposition.
Solution Approach 2:
The patent performs preliminary generation of scenario trees that predict future reactive behaviors of other objects before final trajectory selection. By pre-computing these behavior models for multiple potential futures, the system prepares information in advance that enables safer navigation decisions without increasing the complexity of the final selection process.
2Measurement precision
If conventional systems use simple differentiable prediction models, then the computational efficiency is improved, but the accuracy of predicting other objects' behaviors deteriorates
Solution Approach 1:
The patent employs dynamic scenario trees that are generated recursively to model future reactive behaviors at multiple time steps. Rather than using static simple prediction models, the system dynamically constructs behavior models that adapt to different candidate trajectories and time horizons, achieving higher prediction accuracy while managing computational load through selective generation.
Solution Approach 2:
The patent replaces simple differentiable prediction models with deep-learning prediction models that leverage neural networks to predict future behaviors of other objects. These AI-based models capture complex behavioral patterns and interactions more accurately than traditional mechanical or physics-based models, achieving superior prediction precision despite higher computational requirements.
3Productivity
If conventional systems determine a single trajectory minimizing expected cost, then the computational process is simplified, but the navigation becomes overly conservative and less efficient
Solution Approach 1:
The patent generates multiple candidate trajectories and evaluates each against multiple scenario trees representing different future behaviors, rather than computing a single optimal trajectory. This partial exploration of multiple possibilities allows the system to identify more efficient navigation paths without being overly conservative, while managing complexity through structured evaluation criteria.
Solution Approach 2:
The patent changes the planning approach from minimizing expected cost for a single trajectory to evaluating multiple trajectories against multiple future scenarios. By transforming the objective function and evaluation parameters to consider diverse future possibilities, the system achieves more efficient and adaptive navigation while maintaining tractable computation through parameterized scenario generation.
Data Source
AI summary
In various examples, policy planning using behavior models for autonomous and semi-autonomous systems and applications is described herein. Systems and methods are disclosed that determine a policy for navigating a vehicle, such as a semi-autonomous vehicle or an autonomous vehicle (or other machine), where the policy allows for multistage reasoning that leverages future reactive behaviors of one or more other objects. For instance, a first behavior model (e.g., a trajectory tree) may be generated that represents candidate trajectories for the vehicle and one or more second behavior models (e.g., one or more scenario trees) may be generated that respectively represent future behaviors of the other object(s). The first behavior model and the second behavior model(s) may then be processed, such as in a closed-loop simulation based on a realistic data-driven traffic model, to determine the policy for navigating the vehicle.