US 12,444,175 B2
Method and device for training a style encoder of a neural network and method for generating a driving style representation representing a driving style of a driver
Peter Lakatos, Nőtincs (HU); Gabor Kovacs, Budapest (HU); and Laszlo Szoke, Budapest (HU)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Mar. 29, 2022, as Appl. No. 17/656,941.
Claims priority of application No. 10 2021 203 587.6 (DE), filed on Apr. 12, 2021.
Prior Publication US 2022/0327812 A1, Oct. 13, 2022
Int. Cl. G06V 10/774 (2022.01); G06N 3/0455 (2023.01); G06N 3/088 (2023.01); G06V 10/77 (2022.01); G06V 10/82 (2022.01)
CPC G06V 10/7747 (2022.01) [G06N 3/0455 (2023.01); G06N 3/088 (2013.01); G06V 10/7715 (2022.01); G06V 10/82 (2022.01)] 10 Claims
OG exemplary drawing
 
1. A computer-implemented method performed by a processor for training a style encoder of a neural network, the processor receiving sensory data produced by a sensor, the processor performing the following steps:
compressing sensory input variables, which represent a movement of a system and surroundings of the system, 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;
compressing the sensory input variables into a driving style representation in at least one portion of the latent space, using an untrained style encoder;
decompressing the driving style representation and the driving situation representation from the latent space to output variables, using a style decoder of the neural network; and
changing a structure of the style encoder to train the style encoder until the output variables of the style decoder represent the movement;
wherein, in a preceding step for training the situation encoder, the sensory input variables are compressed to the driving situation representation in at least one portion of the latent space, using the untrained situation encoder, and the driving situation representation is decompressed from the latent space to the output variables using a situation decoder of the neural network, and a driver classifier decompresses the driving situation representation from the latent space to a piece of driver information including a driver identity based on driver criteria, a structure of the situation encoder being changed to train the situation encoder until the driver classifier is unable to decompress any driver information from the driving situation representation, and wherein the output variables of the situation decoder represent the movement and the surroundings; and
wherein the driving situation representation is decompressed from the latent space to the output variables using an untrained situation decoder of the neural network, a structure of the situation decoder being changed to train the situation decoder until the output variables of the situation decoder represent the movement and the surroundings.