Driving Style Encoder Training for Human-Like Vehicle Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge is to ensure that an autonomous or semi-autonomous vehicle maintains a driving style similar to that of the human driver, thereby enhancing the comfort and satisfaction of the driver during automated driving.
Innovation Solution
A neural network is trained to extract an abstract driving style representation using a style encoder, which compresses sensory input data into a latent space, allowing the vehicle to mimic the driver's style by generating control commands that align with their driving habits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensory input data is compressed to an abstract driving style representation in latent space, then the driving style can be accurately captured and replicated, but the complexity of training the style encoder increases
Solution Approach 1:
The training process is segmented into multiple stages: first training the situation encoder to compress sensory input to driving situation representation, then training the style encoder separately to extract driving style features. This segmentation allows each encoder to be optimized independently, reducing overall training complexity while maintaining high representation accuracy.
Solution Approach 2:
The latent space serves as an intermediary between the sensory input data and the output control commands. By compressing data through the latent space representation, the system can efficiently capture essential driving style features without directly processing the full complexity of raw sensory data, thereby reducing training complexity while preserving accuracy.
2Device complexity
If the neural network uses preset variables to describe driving style, then the system complexity is reduced, but the ability to accurately capture individual driving styles is limited
Solution Approach 1:
The patent replaces traditional mechanical approaches using preset variables and manual parameterization with a neural network-based system. The style encoder automatically learns and extracts driving style features from sensory input data, eliminating the need for manually defined variables while achieving superior accuracy in capturing individual driving styles.
Solution Approach 2:
Instead of using fixed preset variables, the system dynamically changes parameters by learning optimal feature representations from data. The style encoder transforms sensory input into adaptive latent space representations that capture the essential characteristics of individual driving styles, allowing the system to accurately describe diverse driving behaviors without predefined constraints.
Data Source
AI summary
A method for training a style encoder of a neural network. Sensory input variables, which represent a movement of a system and surroundings of the system, are compressed to an abstract driving situation representation in at least one portion of a latent space of the neural network, using a trained situation encoder of the neural network. The sensory input variables are compressed to a driving style representation in at least one portion of the latent space, using the untrained style encoder. The driving style representation and the driving situation representation are decompressed from the latent space to output variables, using a style decoder of the neural network. A structure of the style encoder is changed to train the style encoder until the output variables of the style decoder represent the movement.