Siamese Neural Network for Adaptive Driving Style Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users often disable advanced driving features in vehicles due to mismatch between automated driving styles and their personal preferences, leading to frequent takeovers or disabling of automation systems.
Innovation Solution
A Siamese neural network (SNN) based system that uses multi-modal sensors such as heart rate, gaze, grip force, and CAN signals to predict and adapt driving styles, allowing for personalized driving preferences to be learned and applied in real-time without the need for retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If automated driving features are provided with fixed control behavior and driving styles, then the system is simple to implement, but user acceptance and trust decrease when the driving style does not match user preferences
Solution Approach 1:
The system performs preliminary action by training the neural network model offline using aggregated sensor data from multiple users before deployment. During real-time operation, the pre-trained model directly predicts user preferences without requiring complex online retraining, thus achieving adaptability while maintaining system simplicity
Solution Approach 2:
The system creates a virtual model (neural network) that copies and learns from aggregated sensor data patterns of multiple users. This virtual model can then predict and simulate individual user preferences without requiring physical modifications or complex hardware changes to the actual driving system
2Measurement precision
If multiple sensors are used to capture user preferences, then measurement precision of driving style preferences is improved, but device complexity increases
Solution Approach 1:
The system merges data from multiple different sensor types (eye tracking, physiological sensors, steering inputs, pedal inputs) into a unified neural network model. By combining these diverse data streams and processing them through a single predictive model, the system achieves high measurement precision while avoiding the complexity of separate processing systems for each sensor type
Solution Approach 2:
The neural network model serves as a universal processing system that handles multiple sensor types and multiple user preferences simultaneously. This multi-functional model can process various sensor inputs (gaze, heart rate, grip force, pedal position) and predict different aspects of driving style preferences, reducing the need for specialized processing systems
3Measurement precision
If traditional machine learning models are used, then training requires large amounts of data per user, but model performance in detecting individual preferences is insufficient
Solution Approach 1:
The system introduces an intermediary approach by using aggregated sensor data from multiple users as a intermediate training set. This intermediary data pool allows the neural network to learn general driving behavior patterns that can then be applied to predict individual user preferences with limited personal data, achieving high accuracy without requiring large amounts of per-user training data
Solution Approach 2:
The system changes the training parameter approach by shifting from requiring large quantities of per-user data to using a hybrid approach: aggregated data from multiple users for general pattern learning, combined with small amounts of individual user data for personalization. This parameter change in data quantity and composition enables effective individual preference detection
Data Source
AI summary
Siamese neural network (SNN) based adaptive driving style prediction may be achieved by calculating a first distance between the input data and a first class of a set of anchor data using a trained SNN, calculating a second distance between the input data and a second class of the set of anchor data using the trained SNN, and generating an adaptive driving style prediction based on the first distance and the second distance. The trained SNN may be trained based on two or more sensor signals received during a training phase, a distance-based loss for the two or more sensor signals from the training phase, and by back-propagating the distance-based loss.