Automated Driving Reliability Monitoring With Two-Stage Driver Alerts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated driving control using machine learning models faces challenges in determining normal operation due to probabilistic inference results, making it difficult to ensure continuous control.
Innovation Solution
A control apparatus and method that calculates reliability of inference results and notifies operators to check the driving environment and take hands-on control when reliability thresholds are met, effectively monitoring and operating automated driving control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated driving control is performed using a machine learning model, then control performance is improved, but it becomes difficult to determine whether the control can be continued normally due to probabilistic inference results
Solution Approach 1:
The system calculates reliability metrics for machine learning inference results and uses this feedback to determine whether automated driving control should continue. The reliability calculation provides quantitative feedback about the quality of inference results, enabling the system to assess whether control can be continued normally despite the probabilistic nature of ML outputs.
Solution Approach 2:
The system changes the parameter of reliability calculation to transform the probabilistic inference output into a deterministic reliability metric. By calculating reliability as a separate parameter and comparing it against thresholds, the system converts the uncertain probabilistic output into a actionable determination about whether to continue control.
2Reliability
If reliability calculation and notification systems are added to monitor automated driving control, then safety is improved, but system complexity increases
Solution Approach 1:
The monitoring system is segmented into distinct functional modules: reliability calculation unit, first notification unit, and second notification unit. Each module has a specific function, allowing the complex safety monitoring task to be divided into manageable components that can be implemented and maintained separately.
Solution Approach 2:
The system performs preliminary reliability calculation before making control continuation decisions. By calculating reliability in advance and establishing notification thresholds beforehand, the system prepares the necessary safety assessment mechanisms before they are needed, reducing the complexity of real-time decision-making.
Data Source
AI summary
A control apparatus for a vehicle comprises one or more processors configured to perform automated driving control of the vehicle by using a machine learning model. The control apparatus calculates a reliability regarding an inference result by the machine learning model while performing the automated driving control. When a first condition indicating that the reliability has decreased is met, the control apparatus performs notification for prompting an operator to check a driving environment of the vehicle. When a second condition indicating that the reliability has further decreased in addition to the first condition is met, the control apparatus performs notification for requesting hands-on to the operator.