Autonomous Perception Fault Isolation via Neural Network Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous and active safety vehicles face challenges in detecting and isolating faults in their perception systems, which can impact their operational efficiency and safety, particularly in complex environments where sensor data processing and object detection are critical.
Innovation Solution
A method and system that utilize neural network models to process sensor data, analyze potential faults such as false positives, false negatives, processing time issues, temporal changes, and spatial variance, and take vehicle control actions, including relinquishing automated functions if faults are detected, to ensure safe operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If perception systems process sensor data to detect objects in complex environments, then navigation capability is improved, but fault detection and isolation capability deteriorates
Solution Approach 1:
The patent segments the perception system into multiple independent analysis components, each dedicated to detecting specific fault types (false positives, false negatives, processing time faults, temporal change faults, spatial variance faults). This segmentation allows each component to specialize in detecting its specific fault type without being overwhelmed by the complexity of the entire perception system, thereby improving fault detection capability while maintaining navigation capability.
2Measurement precision
If multiple fault analysis techniques are applied to perception results, then fault detection accuracy is improved, but system complexity increases
Solution Approach 1:
The patent implements a universal fault isolation system that handles multiple fault types through a standardized multi-technique analysis framework. The same processor and analysis infrastructure are used across all five fault detection techniques (bounding box correlation, class correlation, processing time analysis, temporal change detection, spatial variance detection), allowing the system to achieve high fault detection accuracy without proportionally increasing overall system complexity.
3Reliability
If real-time fault analysis is performed on perception results, then vehicle safety is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary fault analysis by continuously monitoring perception results in real-time using multiple specialized techniques. The system proactively identifies faults such as false positives, false negatives, and processing delays before they compromise vehicle safety, allowing for preventive control actions rather than reactive responses, thereby improving safety while managing processing time through early detection.
4Reliability
If automated control functions are relinquished in response to faults, then safety is improved, but operational efficiency deteriorates
Solution Approach 1:
The patent implements a feedback-based fault isolation system that continuously monitors perception results and automatically adjusts vehicle control functions based on detected faults. When faults are identified through the multi-technique analysis, the system provides feedback to relinquish or adjust automated control functions, ensuring safety while minimizing unnecessary interruptions to operational efficiency by only disengaging automation when actually needed.
Data Source
AI summary
In various embodiments, method, systems, and vehicles are provided that include: obtaining, via one or more sensors, sensor data pertaining including one or more images of one or more detected objects in proximity to a vehicle; processing the one or more images via a neural network model of a perception system of the vehicle, generating perception results for the one or more images; analyzing, via a processor, a plurality of potential faults in the perception system, based on the perception results, using a respective different technique for each of the plurality of potential faults, for the neural network model; determining a fault of the plurality of potential faults, via the processor, based on the analyzing; and taking a vehicle control action with respect to the vehicle, based on the fault, via instructions provided by the processor.