Adaptive Lane-Keeping Control Using Driver-Specific Offset Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lane keeping assist (LKA) systems often overcorrect or undercorrect vehicle positioning, leading to unpleasant motion sensations and a lack of driver confidence, especially for experienced drivers, particularly on challenging road conditions.
Innovation Solution
A neural network module that learns driver-specific driving styles and environmental parameters to generate adaptive lane assist offset values, adjusting steering angle control signals based on individual driver preferences and real-time conditions, thereby enhancing the LKA system's responsiveness and comfort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a predetermined lane offset value is used for LKA correction, then the system is simple to operate, but the system lacks adaptability to individual driver preferences and road conditions
Solution Approach 1:
The system performs preliminary actions by collecting driver data and training neural network models during a calibration phase before actual LKA operation. Driver-specific models are pre-trained using historical steering wheel data, pressure sensor data, and vehicle telemetry to capture individual driving styles. This preliminary training enables the system to adapt to each driver's preferences without adding complexity during real-time operation, as the adaptive behavior is prepared in advance.
Solution Approach 2:
The system creates copies of driver behavior patterns through neural network models that replicate each driver's unique steering characteristics. Multiple driver profiles are stored as separate models, allowing the system to copy and apply the appropriate driver's preferred lane positioning behavior. This copying approach enables adaptability to individual drivers while keeping the core LKA system architecture relatively simple, as each driver essentially has their own pre-computed behavioral model.
2Ease of operation
If conventional LKA systems apply uniform corrective steering, then the system is easy to manufacture, but the system causes unpleasant motion sensations and driver discomfort
Solution Approach 1:
The system applies local quality by customizing the LKA correction behavior for each individual driver based on their unique steering patterns and preferences. Instead of a uniform correction applied to all drivers, the system adjusts the corrective steering angle, rate of change, and magnitude according to each driver's specific characteristics captured during calibration. This localized adaptation improves driver comfort by making the corrections feel natural to each driver while maintaining a relatively simple overall system architecture through the use of driver-specific models.
3Adaptability or versatility
If the LKA system uses a fixed lane offset value, then the system is reliable and consistent, but the system cannot adapt to different driving styles and environmental conditions
Solution Approach 1:
The system transitions from a static, fixed lane offset value to a dynamic, adaptive offset that changes based on the identified driver and environmental conditions. The neural network models continuously adjust the lane offset parameter in real-time based on inputs from cameras, sensors, and the identified driver's profile. This dynamic adaptation maintains reliability by ensuring consistent performance for each driver while improving versatility to handle different road conditions, weather, and traffic scenarios.
4Adaptability or versatility
If driver-specific models are trained using neural networks, then the system achieves high adaptability to individual drivers, but the device complexity and computational requirements increase
Solution Approach 1:
The system performs the computationally intensive neural network training in advance during a calibration phase, collecting driver data from steering wheel sensors, pressure sensors, and vehicle telemetry. The models are pre-trained and stored for later use, avoiding the need for real-time training during actual LKA operation. This preliminary action reduces computational complexity during operation while maintaining high adaptability to individual drivers.
Solution Approach 2:
The system uses feedback from driver behavior data to continuously refine and update the neural network models during the calibration phase. By collecting and analyzing actual driver steering patterns, pressure application, and vehicle responses, the system iteratively improves the accuracy of driver-specific models. This feedback mechanism enables high adaptability while managing computational complexity through focused offline training rather than continuous real-time computation.
Data Source
AI summary
A system includes a neural network module, a lane keeping assist (LKA) module, and a control module. The neural network module is configured to receive driver data, identify the driver, receive non-driver data, and generate a lane assist offset value specific to the identified driver based on the non-driver data and the driver data. The LKA module is configured to receive the lane assist offset value, receive vehicle data associated with one or more parameters of the vehicle while the vehicle is moving, the one or more parameters including a steering angle, a velocity, and a vehicle position between lane markings, and generate a steering angle control signal based on the received lane assist offset value and vehicle data. The control module is configured to control steering of the vehicle based on the generated steering angle control signal. Other example systems and methods are also disclosed.