Autonomous Vehicle Trajectory Safety Filtering Before ML Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for autonomous vehicles face inefficiencies in trajectory planning due to the computational expense of evaluating numerous possible trajectories, often resulting in the selection of unsafe paths, especially in complex environments, and lack sufficient training data to accurately determine safe trajectories.
Innovation Solution
A safety filter is implemented to identify and filter out unsafe trajectories by applying predefined assumptions about other vehicles and using trajectory modifiers, reducing the computational load and incorporating expert knowledge to ensure safer path selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of trajectories are evaluated to ensure safety, then the safety of the selected trajectory is improved, but the computational cost and time required increase significantly
Solution Approach 1:
The safety filter divides the trajectory evaluation process into distinct segments: generating multiple candidate trajectories, applying safety checks to filter unsafe ones, and selecting from the remaining safe trajectories. This segmentation allows the system to evaluate fewer trajectories while maintaining safety by ensuring that the selected trajectory passes through the filtering stage.
Solution Approach 2:
The safety filter performs preliminary safety checks on generated trajectories before they are selected by the machine learning planner. By pre-filtering unsafe trajectories based on predefined safety rules and constraints, the system reduces the computational burden on the planner while ensuring that only safe trajectories are considered for selection.
2Reliability
If expert knowledge and safety rules are incorporated into the planning system, then the safety of trajectory selection is improved, but the system complexity increases
Solution Approach 1:
The safety filter acts as an intermediary component between the trajectory generator and the machine learning planner. It implements safety rules and expert knowledge in a modular fashion, translating complex safety requirements into filterable constraints without directly complicating the planner's architecture.
Solution Approach 2:
The safety filter uses simplified representations of safety rules and constraints that can be efficiently evaluated without requiring the full complexity of the original expert knowledge systems. This allows safety checks to be performed with reduced computational overhead while maintaining essential safety guarantees.
Data Source
AI summary
Provided are methods for a safety filter for machine learning planners. Example methods can include applying a plurality of safety parameters to a plurality of trajectories generated for an ego vehicle, determining whether the plurality of trajectories are unsafe based at least on application of the plurality of safety parameters to the plurality of trajectories, filtering a trajectory from the plurality of trajectories based at least on determining the trajectory is unsafe, and providing the remaining trajectories from the plurality of trajectories to a machine learning model trained to generate a score for selection of a selected trajectory for the vehicle from the remaining trajectories. Systems and computer program products are also provided.