AV Perception State Identification Using Compliant Velocity Priors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous vehicle perception systems struggle to accurately forecast the state of road actors with unknown or uncertain measurements due to sensor noise, sensor disagreements, and obscured objects, leading to inappropriate assumptions and unexpected vehicle maneuvers.
Innovation Solution
The system employs a perception module that combines estimated tracking information with compliant priors, assuming road actors follow traffic rules unless evidence indicates otherwise, using sensor data, map information, and social features to generate and compare velocity distributions to estimate the state of uncertain road actors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional AV perception systems make assumptions about unknown states of road actors, then the system can operate in dynamic environments, but the accuracy of state forecasting deteriorates due to sensor noise, sensor disagreements, and obscured objects
Solution Approach 1:
The system performs preliminary actions by generating multiple velocity distributions (from sensor data, map information, and kinematic state of additional objects) before final state estimation. This allows the system to prepare multiple hypotheses about the unknown state and select the most probable one, improving forecasting accuracy while maintaining operational capability in dynamic environments
Solution Approach 2:
The patent introduces an intermediary mechanism - the velocity distribution comparison process - that mediates between uncertain sensor measurements and final state estimation. By comparing multiple velocity distributions and selecting the most consistent one, the system resolves uncertainties without making arbitrary assumptions, thereby improving measurement precision while adapting to dynamic conditions
2Measurement precision
If the system uses multiple sources of velocity distributions (sensor data, map information, kinematic state of additional objects), then the accuracy of velocity estimation improves, but the complexity of the perception system increases
Solution Approach 1:
The system segments the velocity estimation problem into three distinct sources: sensor data-based velocity distribution, map information-based velocity distribution, and kinematic state-based velocity distribution from additional objects. Each source is processed independently and then compared to determine the most consistent velocity estimate, improving accuracy while managing complexity through modular processing
Solution Approach 2:
The patent creates a universal velocity estimation framework that can process and compare velocity distributions from multiple different sources (sensors, maps, kinematic models). This multi-functional approach allows the same system architecture to handle diverse data types and sources, improving estimation accuracy without proportionally increasing system complexity through standardized processing pipelines
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for state identification for road actors with uncertain measurements based on compliant priors. The perception system of an autonomous vehicle (AV) may detect an object with an uncertain kinematic state based on sensor information received from a sensing device associated with the AV. A first, second, and third distribution of velocity values may be generated based on the sensor information, map information, and a kinematic state for each additional object of a plurality of additional objects in proximity to the detected object with the uncertain kinematic state, respectively. A velocity value for the detected object with the uncertain kinematic state may be generated based on a comparison of the first, second, and third distributions of velocity values. The AV may perform a driving maneuver based on the velocity value for the detected object.