Autonomous Vehicle Sign Recognition via Fleet Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional perception subsystems in autonomous vehicles are unable to handle the full array of variable road signage due to limited training data, resulting in uncertain and unsafe navigation when encountering less common signs.
Innovation Solution
An autonomous vehicle system that detects unrecognized signs, sends sensor data to a remote operations center for guidance, and updates a central machine learning model based on assessment data from multiple vehicles, enabling continuous learning and improved recognition of variable signage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a local machine learning model is used for sign recognition in autonomous vehicles, then the system can operate independently with limited computing resources, but the recognition accuracy decreases for less common or variable signage
Solution Approach 1:
The patent combines local machine learning models in each vehicle with a centralized operations center that aggregates data from multiple vehicles. This merging allows the system to maintain independent local operation while accessing collective fleet knowledge through the central server, thereby improving recognition accuracy for variable signage without compromising local autonomy.
Solution Approach 2:
The system implements a feedback loop where unrecognized signs are captured by sensor data, sent to the operations center, and used to retrain the central machine learning model. The updated model is then distributed back to vehicles, creating continuous improvement of recognition accuracy through iterative feedback from real-world encounters.
2Reliability
If sensor data is sent to a remote operations center for unrecognized signs, then recognition accuracy improves through centralized learning, but system complexity and communication requirements increase
Solution Approach 1:
The operations center serves as an intermediary between individual vehicles and the central machine learning model. It receives sensor data from vehicles with unrecognized signs, processes the information, and distributes updated models back to the fleet, thereby managing system complexity through a dedicated intermediary component.
Solution Approach 2:
The system enables self-service through automated processes where vehicles autonomously identify unrecognized signs, submit sensor data to the operations center, and receive updated models without human intervention. This automation reduces operational complexity while maintaining high navigation safety.
3Adaptability or versatility
If the central machine learning model is continuously updated with fleet data, then the system adapts to new signage variations, but data processing time and computational resources increase
Solution Approach 1:
The system performs preliminary actions by continuously collecting and preprocessing sign data from the fleet in the background. When new signage patterns are detected, the central model is retrained proactively before they become critical issues, ensuring the system is prepared for upcoming challenges without causing operational delays.
Solution Approach 2:
The machine learning model retraining is implemented as a periodic action rather than a continuous process. The system aggregates data over time and performs scheduled retraining cycles, which reduces computational overhead and time loss compared to real-time updates, while still maintaining adaptability to new signage variations.
Data Source
AI summary
Methods, systems, and non-transitory computer readable media are configured to perform operations comprising determining an occurrence of a sign in an environment of a vehicle; determining that at least a portion of the sign is unrecognized by a local machine learning model of the vehicle; and providing sensor data associated with the at least one portion of the sign to an operations center remote from the vehicle.