AV Perception Track Filtering for Trajectory-Relevant Collision Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional perception systems for automated vehicles (AVs) output a large number of tracks, including those that do not interact with the AV's planned trajectory, leading to inefficient use of computational resources and failure to identify critical collision avoidance tracks.
Innovation Solution
The perception system generates a trajectory for the AV based on a planned route and sensor information, identifies objects with trajectories intersecting the AV's trajectory, removes objects with kinematically unfeasible trajectories or intersections with other objects, and selects objects indicating potential collisions, assigning a severity ranking to each.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the perception system outputs all detectable tracks, then the system provides comprehensive observation coverage, but computational resources are unnecessarily taxed and critical collision avoidance tracks are not identified
Solution Approach 1:
The patent extracts and isolates only the critical subset of tracks that pose potential collision risks with the AV's planned trajectory. By filtering out non-critical tracks and focusing computational resources on objects with intersecting trajectories, the system achieves high reliability in collision detection without the computational burden of processing all detectable tracks.
Solution Approach 2:
The patent applies different processing quality levels to different tracks based on their relevance to the AV's trajectory. Critical tracks that intersect with the planned path receive near-perfect perception performance and intensive computational analysis, while non-critical tracks receive minimal or no processing, optimizing the allocation of computational resources.
2Reliability
If the perception system processes all tracks with near-perfect performance, then collision avoidance is maximized, but computational resources are exhausted and processing efficiency decreases
Solution Approach 1:
The patent segments the set of all detected tracks into distinct categories: critical tracks that intersect with the AV's planned trajectory and non-critical tracks that do not. This segmentation enables the system to apply near-perfect perception performance selectively to only the critical subset, maintaining high reliability while preserving processing efficiency by avoiding intensive analysis of non-critical tracks.
3Productivity
If the system filters out objects with non-intersecting trajectories, then computational resources are conserved, but the risk of missing potential collision risks increases
Solution Approach 1:
The patent performs preliminary analysis of object trajectories to predict future positions and determine potential intersections with the AV's planned trajectory before committing full computational resources. By evaluating kinematic feasibility and trajectory intersections in advance, the system confidently filters out non-critical objects while retaining all potential collision risks for detailed analysis, ensuring neither resource efficiency nor detection accuracy is compromised.
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for determining objects that are kinematically capable, even if non-compliant with rules-of-the-road, of affecting a trajectory of a vehicle. The computing system (e.g., perception system, etc.) of a vehicle may generate a trajectory for the vehicle and a respective trajectory for each object of a plurality of objects within a field of view (FOV) of the sensing device associated with the vehicle. The computing system may identify objects of the plurality of objects with trajectories that intersect the trajectory for the vehicle and remove from such objects, objects with trajectories that at least one of exit the FOV or intersect with other objects of the plurality of objects within the FOV. The computing system may select, from remaining objects with trajectories that intersect the trajectory for the vehicle, objects with trajectories that indicate a respective collision between the object and the vehicle and assign a severity of the respective collision.