Learned Trajectory Tree Search for Scalable Vehicle Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Rule-based approaches for vehicle navigation are computationally expensive and scale poorly to new geographies, driving scenarios, and environments, often failing to provide ideal driving behavior.
Innovation Solution
Implement a machine-learning (ML) system to generate a tree structure using diverse sets of machine-learned trajectories, reducing the reliance on heuristic-based trajectories to optimize computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If rule-based approaches are used for vehicle navigation, then trajectories can be determined using hand-engineered driving rules, but computational resources are expensive and the system scales poorly to new geographies and driving scenarios
Solution Approach 1:
The patent replaces hand-engineered rule-based trajectory generation with machine learning models that learn driving behavior from data. The ML models substitute the mechanical process of manually coding driving rules with an automated learning system that adapts to different geographies and scenarios without requiring manual re-engineering of rules.
Solution Approach 2:
The system changes the fundamental parameter of trajectory generation from fixed hand-crafted rules to flexible learned parameters from ML models. This allows the system to adapt to new geographies and driving scenarios by loading different trained models or fine-tuning existing ones, rather than re-engineering rules for each scenario.
2Productivity
If heuristic-based trajectories are used to fill the tree structure, then trajectories can be generated, but the computational cost increases and the proportion of learned trajectories decreases
Solution Approach 1:
The patent performs preliminary action by pre-training ML models offline to generate diverse candidate trajectories. During runtime, these pre-learned trajectories are directly retrieved and used to populate the tree structure, eliminating the need for computationally expensive heuristic generation during critical decision-making moments.
Solution Approach 2:
The system uses copying by generating diverse candidate trajectories through the ML model and then copying these learned trajectories into the tree structure for evaluation. This replaces the need to generate new heuristic trajectories from scratch, reducing computational energy while maintaining trajectory diversity.
3Loss of energy
If diverse sets of machine-learned trajectories are used to generate the tree structure, then computational resources are reduced, but the system requires ML models to generate diverse candidate trajectories
Solution Approach 1:
The patent segments the trajectory generation system into distinct components: an ML model that generates diverse candidate trajectories, a tree structure component that organizes these candidates, and an evaluation component that selects the optimal trajectory. This segmentation allows each component to be optimized independently and facilitates the integration of ML with traditional trajectory evaluation.
Solution Approach 2:
The ML model serves multiple functions: it generates diverse candidate trajectories, captures uncertain dynamics, and adapts to different geographies and scenarios. This multi-functionality reduces the need for separate specialized components, thereby reducing overall system complexity while achieving the goal of lowering computational resource consumption.
Data Source
AI summary
Techniques for generating a tree structure based on multiple machine-learned trajectories are described herein. A planning component (“ML system”) within a vehicle may receive and encode various types of sensor and/or vehicle data. The ML system can provide the encoded data as input to multiple machine-learning models (“ML models”), each of which may be trained to output a unique candidate trajectory for the vehicle follow. In some examples, each ML model may be trained to output a unique type of learned trajectory that causes the vehicle to perform a certain type of action. Using the learned candidate trajectories, the ML system may generate a tree structure that includes some or all of the candidate trajectories. The vehicle may determine a control trajectory based on the generation and traversal of the tree structure using a tree search algorithm, and may follow the control trajectory within the environment.