Autonomous Driving Rule Switching for Stable Control Under Input Deviation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models in autonomous driving can produce unpredictable and unstable control when faced with input data significantly deviated from their learning data, posing a challenge for stable control in critical scenarios.
Innovation Solution
A control device that switches from a first control rule based on machine learning to a second fixed control rule upon detecting specific events, using sensor data and external communication to ensure stable operation, and optionally updates a common control rule based on collective vehicle data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a machine learning model is used for autonomous driving control, then optimum control based on learning data is achieved, but stable control deteriorates when input data deviates from learning data
Solution Approach 1:
The control device dynamically switches between machine learning-based control and fixed control rules based on the recognition result of the current situation. When the situation is recognized as a teaching case (deviated input), the system transitions from machine learning control to fixed control rules, ensuring stable and predictable behavior in critical scenarios while maintaining adaptability in normal conditions.
2Measurement precision
If machine learning control is used, then control precision is improved for learning data, but unpredictable operations occur for deviated input data
Solution Approach 1:
A situation recognition unit acts as an intermediary between the machine learning-based autonomous driving control unit and the fixed control rules. This intermediary evaluates whether the current input situation matches known patterns from teaching data, and based on this evaluation, determines whether to use machine learning control or switch to fixed control rules, ensuring predictable behavior when deviations are detected.
3Reliability
If fixed control rules are used, then control stability is improved, but adaptability to new situations deteriorates
Solution Approach 1:
The control system dynamically selects between fixed control rules and machine learning-based control based on the situation recognition result. In normal situations matching teaching data, the system uses machine learning control for high adaptability. When deviations are detected and situations are recognized as teaching cases, the system switches to fixed control rules for stability, thus dynamically balancing adaptability and stability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A control device (100) includes an event detection unit (110) and a control-rule change unit (120) . The event detection unit (110) determines whether or not an event to be a trigger of changing a control rule at the time of autonomous driving of a vehicle is detected while the vehicle is performing the autonomous driving using a first control rule based on machine learning. The control-rule change unit (120) changes the control rule at the time of the autonomous driving of the vehicle to a second control rule according to the event to be the trigger in a case where the event to be the trigger is detected by the event detection unit (110).