Vehicle Recognition Confidence Alerts for Autonomous Driving

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In autonomous driving systems, supervisors may overrely on AI outputs, leading to a lack of appropriate understanding of the confidence level in recognition tasks, which can impact safe vehicle operation.

Innovation Solution

A recognition state notification device is implemented within the autonomous driving system, featuring a confidence level calculation unit and a notification unit. This device calculates the confidence level of recognized targets and notifies the supervisor when the confidence level falls below a reference value, allowing for differentiated notification modes based on confidence levels.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If the autonomous driving system uses AI for recognition tasks, then the automation level is improved, but the supervisor's understanding of confidence level deteriorates

Engineering Contradiction:
Improveautonomous driving automationVSAvoidconfidence level information
Core Design Contradiction:
Extent of automationVSLoss of information

Solution Approach 1:

The notification unit provides feedback to the supervisor about the confidence level of AI recognition results. When the confidence level falls below a reference value, the system notifies the supervisor, creating a feedback loop that informs the supervisor about the reliability of AI recognition without reducing automation.

Inventive Principle:
Principle #23Feedback

2Loss of information

If the system notifies the supervisor of all recognition targets, then the information completeness is improved, but the notification burden increases

Engineering Contradiction:
Improverecognition information completenessVSAvoidsupervisor operation burden
Core Design Contradiction:
Loss of informationVSEase of operation

Solution Approach 1:

The notification unit applies different notification strategies based on the confidence level of each recognition target. High-confidence targets are not notified, while low-confidence targets trigger notifications. This local differentiation reduces overall notification burden while maintaining information completeness for critical cases.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system changes the notification parameter based on confidence level thresholds. By comparing confidence levels against reference values and adjusting notification behavior accordingly, the system maintains information completeness for uncertain recognitions while avoiding unnecessary notifications for high-confidence targets.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4506224A1Recognition state notification device
Publication Date: 2025.02.12 TOYOTA JIDOSHA KK
  • EP4506224A1 patent drawingFigure 1
  • EP4506224A1 patent drawingFigure 2
  • EP4506224A1 patent drawing

AI summary

A recognition state notification device includes: a confidence level calculation unit configured to calculate a confidence level in recognition of a recognition target recognized by a recognition unit; and a notification unit configured to notify a supervisor of a vehicle of a recognition target whose confidence level is equal to or lower than a reference value.