Steering Torque AI Detection for Vehicle Hands-Off Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting hands-off conditions in vehicles, such as those using capacitive distance sensors and machine learning approaches, face challenges in generating and validating training data, leading to unreliable and precise detection due to complex labeling of steering torque curves and lack of uniform evaluation criteria.
Innovation Solution
A process for generating labeled steering torque data by recording steering torque and distance data from sensors while the vehicle is moving, using a threshold value to automatically detect hands-off situations and synchronize them with time stamps, which can be used to train an adaptive algorithm like a neural network for improved detection without requiring a distance sensor.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If capacitive distance sensors are used to detect hands-off condition directly, then measurement precision is improved, but device complexity and technological problems increase
Solution Approach 1:
The patent uses capacitive distance sensors during the training phase to create labeled training data (a copy of ground truth), but the actual detection system only needs the trained machine learning model and steering torque sensors. This separates the high-precision measurement need from the operational system complexity.
Solution Approach 2:
The patent performs preliminary data collection and labeling using distance sensors before deploying the detection system. The training data is pre-labeled with hands-off condition information, so the operational system doesn't need distance sensors during actual detection, reducing complexity while maintaining precision through the trained model.
2Adaptability or versatility
If machine learning approaches are used with manual labeling of steering torque curves, then adaptability is improved, but ease of manufacture and data generation complexity worsen
Solution Approach 1:
The patent implements an automated labeling system where the machine learning model predicts hands-off conditions from steering torque data, and these predictions are automatically used as labels for training data. This self-service approach eliminates manual labeling while maintaining adaptability through iterative training.
Solution Approach 2:
The patent uses a feedback loop where the machine learning model makes predictions, these predictions are compared with actual distance sensor data (ground truth), and the model is retrained with this feedback. This automated feedback mechanism improves adaptability while eliminating manual intervention.
3Measurement precision
If a distance sensor is used during vehicle operation, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The distance sensor is used only during the preliminary training phase to create labeled training data. After training, the sensor is removed from the operational system, which relies solely on the trained machine learning model and existing steering torque sensors for detection.
Solution Approach 2:
The training data created with distance sensors serves as a permanent copy of ground truth that enables the model to achieve high precision without requiring the actual distance sensor during operation. The information is captured and stored in the trained model parameters.
Data Source
AI summary
Technologies and techniques for automatically generating labeled steering torque data, with which an artificial intelligence (AI) unit is trained to detect hands-off conditions when the vehicle is being operated.