Adaptive Control Model Selection for Low-Intervention Mobile Bodies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems often require frequent user intervention due to incompatibility between control models and user preferences, leading to reduced system usage and increased wear on operation tools.
Innovation Solution
A control device that includes a storage for control models and a selection model, which selects appropriate models based on user intervention records to minimize user intervention by adapting control commands to user preferences, using a combination of trained machine learning and rule-based models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed control model is used for autonomous control, then the control system is simple and reliable, but user intervention frequency increases due to incompatibility with user preferences
Solution Approach 1:
The system dynamically switches between multiple control models based on situation patterns and user preferences. The selection model adapts the control strategy in real-time by selecting appropriate control models from a plurality of available models, making the control system flexible and adaptable to different user preferences while maintaining reliability through model-based control.
Solution Approach 2:
The system changes the parameters of the control system by selecting different control models with different characteristics. Each control model has distinct control parameters and strategies, and the selection model adjusts which model is active based on the situation pattern and learned user preferences, effectively changing the control behavior without altering the underlying control architecture.
2Ease of operation
If multiple control models are used to adapt to user preferences, then user intervention frequency decreases, but device complexity increases
Solution Approach 1:
The selection model acts as an intermediary between the multiple control models and the execution system. It receives situation patterns as input, determines the most appropriate control model based on learned user preferences, and outputs the selected model for execution. This intermediary layer manages the complexity by providing a unified interface while handling the diversity of control models internally.
Solution Approach 2:
The control system is segmented into distinct functional modules: multiple specialized control models for different control scenarios, a selection model for model choice, and an execution unit for implementing the selected model. This segmentation allows each component to be optimized independently while working together as an integrated system, managing overall complexity through modular architecture.
3Ease of operation
If control models are frequently switched to match user preferences, then ease of operation improves, but loss of time for model selection increases
Solution Approach 1:
The system performs preliminary actions by pre-training the selection model with situation patterns and user preference data before actual autonomous operation. The selection model learns to quickly identify the most appropriate control model for given situations in advance, reducing the time required for model selection during actual operation. User preferences are also learned and stored for future reference, avoiding repeated analysis.
4Adaptability or versatility
If user intervention records are stored and analyzed, then adaptability to user preferences improves, but loss of information privacy increases
Solution Approach 1:
The system extracts only the essential pattern information from user intervention records that is necessary for control model selection, while excluding personally identifiable information and sensitive personal data. The selection model learns from aggregated pattern data rather than storing complete user records, extracting only the control-relevant features while leaving private information behind.
Data Source
AI summary
The control device according to one aspect of the present disclosure selects one or more control models from a plurality of control models by a selection model and derives a control command of a mobile body by using one or more selected control models. In the absence of an intervention operation, the control device controls the operation of the mobile body according to the derived control commands. When there is an intervention operation, the control device discards the derived control command or overlaps the intervention operation with the derived control command and controls the operation of the mobile body according to the operation of the intervention by the user. The selection model is configured to select one or more control models to avoid the occurrence of intervention from the intervention archival record of the user.