Remote Assistance Rewind Images for Low-Confidence AV Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately identifying objects with low detection confidence, leading to potential safety issues due to uncertainty in recognizing unusual or unfamiliar environmental elements, such as unusual traffic directions or partially obscured signs.
Innovation Solution
A remote assistance system that receives image data from autonomous vehicles, determines objects with low detection confidence, and provides previously-stored image data to operators for verification, allowing them to input instructions back to the vehicle via a network, enabling correction of object identification and vehicle control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If autonomous vehicle systems rely solely on automated object detection, then operational speed and efficiency are improved, but detection accuracy and safety deteriorate when encountering unusual or unfamiliar environmental elements
Solution Approach 1:
The patent introduces a remote assistance system that acts as an intermediary between the autonomous vehicle and human operators. When the detection system encounters low-confidence objects, the system automatically requests remote assistance, allowing human operators to verify and correct detections. This mediator approach maintains automated operation for routine tasks while providing human expertise for uncertain situations, resolving the contradiction between operational speed and detection accuracy.
Solution Approach 2:
The system performs preliminary automated detection for all objects, then selectively escalates only low-confidence detections to remote human operators. This preliminary action allows the system to maintain high operational speed for routine detections while ensuring accuracy for uncertain cases, avoiding the need for continuous human oversight and preserving productivity.
2Reliability
If the system requests remote assistance for all low-confidence detections, then detection accuracy is improved, but system complexity and response time increase
Solution Approach 1:
The patent applies local quality by differentiating the level of assistance needed for different detections. Instead of uniform human oversight, the system selectively engages remote assistance only for low-confidence detections, maintaining automated processing for high-confidence cases. This localized approach to quality control improves detection accuracy where needed while avoiding unnecessary system complexity for routine detections.
Solution Approach 2:
The system performs partial human verification only for low-confidence detections rather than requiring complete human review of all detections. This partial action approach provides sufficient detection accuracy for uncertain cases while avoiding the excessive complexity and response time delays that would result from comprehensive human oversight of all detections.
3Measurement precision
If pre-stored image data is provided to operators, then verification accuracy is improved, but data transmission time and memory requirements increase
Solution Approach 1:
The system performs preliminary capture and storage of image data at the source (autonomous vehicle) before remote verification is needed. This preliminary action ensures that high-quality verification data is already available when remote assistance is requested, improving verification accuracy without requiring time-consuming data transmission from the vehicle to the remote system.
Solution Approach 2:
The patent uses copying by transmitting only the necessary image data portions to remote operators for verification, rather than transmitting entire datasets. This selective copying approach provides sufficient verification accuracy for low-confidence detections while minimizing data transmission time and memory requirements at the remote system.
Data Source
AI summary
Examples described may enable provision of remote assistance for an autonomous vehicle. An example method includes a computing system operating in a rewind mode. In the rewind mode, the system may be configured to provide information to a remote assistance operated based on a remote-assistance triggering criteria being met, such as a detected object having a low detection confidence. When the triggering criteria is met, the remote assistance system may provide data from the time leading up to when the remote-assistance triggering criteria was met that was capture of the environment of autonomous vehicle to the remote assistance operator. Based on viewing the data, the remote assistance operator may provide and input to the system that causes a command to be issued to the autonomous vehicle.