Autonomous Driving Model Update Verification Across Vehicle Fleets

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing autonomous driving systems lack effective verification methods for ensuring that updated machine learning models are functioning correctly, which can impact the reliability of autonomous driving functions.

Innovation Solution

A method and system where a management server transmits update data to vehicles, collects verification data from multiple vehicles using the updated models, and verifies the autonomous driving control functions to ensure the models are functioning as intended.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If trained models are updated in autonomous driving systems, then the performance and accuracy of autonomous driving control can be improved, but the reliability and safety of the system cannot be ensured without proper verification

Engineering Contradiction:

Improvereliability of autonomous driving functionVSAvoidcomplexity of verification system

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements a feedback mechanism where verification results from multiple vehicles are collected and used to determine whether model updates should be deployed. The management server receives verification data from vehicles, evaluates whether the updated model performs as intended, and only distributes the update if verification succeeds. This closed-loop feedback system ensures reliability while maintaining controlled complexity.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent applies preliminary action by performing verification tests before widespread deployment of updated models. The management server distributes updated models to a subset of vehicles for verification, collects performance data, and evaluates results before deciding on full deployment. This preliminary testing phase prevents unreliable updates from reaching the entire fleet.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If verification data is collected from multiple vehicles, then the accuracy of model validation can be improved, but the time and resources required for verification increase

Engineering Contradiction:

Improveaccuracy of model validationVSAvoidtime for verification process

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the verification process by dividing the vehicle fleet into multiple groups that verify different aspects or versions of the updated model. The management server distributes verification tasks to multiple vehicles simultaneously, collecting data in parallel rather than sequentially. This segmentation enables accurate validation across diverse driving conditions while reducing total verification time through concurrent processing.

Inventive Principle:

Principle #1Segmentation

3Adaptability or versatility

If updated trained models are deployed to vehicles, then the functionality of autonomous driving control can be enhanced, but the risk of incorrect or harmful behavior increases without verification

Engineering Contradiction:

Improvefunctionality of autonomous driving controlVSAvoidrisk of incorrect model behavior

Core Design Contradiction:

Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

The patent applies preliminary anti-action by implementing a verification mechanism that proactively identifies and prevents harmful behaviors before they affect the entire fleet. The management server monitors verification data from test vehicles to detect any incorrect or harmful model behavior, and can halt deployment or request model corrections if issues are found. This preemptive approach counteracts potential harm before it spreads.

Inventive Principle:

Principle #9Preliminary anti-action

4Reliability

If verification processes are implemented for updated models, then the safety and correctness of autonomous driving functions can be ensured, but the complexity of the management system increases

Engineering Contradiction:

Improvesafety of autonomous driving functionVSAvoidcomplexity of management server system

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies universality by designing the management server to perform multiple functions: distributing model updates, collecting verification data from vehicles, evaluating verification results, and controlling deployment decisions. This multi-functional approach consolidates verification complexity into a single centralized system rather than requiring complex verification mechanisms in each vehicle, thereby ensuring safety while managing overall system complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20240391492A1Method and device for managing vehicle data

Publication Date: 2024.11.28 TOYOTA JIDOSHA KK

AI summary

A management server transmits data for updating indicating data to update a trained model to at least two vehicles on which autonomous driving control by using the trained model is performed. The trained model indicates a model generated by a machine learning. The management server collects data for verification from each of the at least two vehicles to which the data for updating have been transmitted. The data for verification indicates data obtained or generated in connection with the performance with the autonomous driving control by using an updated trained model indicating the trained model updated by the data for updating. The management server verifies the function of the autonomous driving control in which the updated trained model is used, by using the data for verification that have been collected from the at least two vehicles.