Training neural networks using a neural network

Patent metadata

Field	Value
Publication number	US-12541951-B2
Application number	US-202017757834-A
Country	US
Kind code	B2
Filing date	Dec 15, 2020
Priority date	Dec 23, 2019
Publication date	Feb 3, 2026
Grant date	Feb 3, 2026

Abstract

The disclosure relates to a method for training a first neural network, in particular for generating training data for at least one second neural network, using a controller, wherein measurement data ascertained by at least one surroundings sensor or artificially generated data of initially ten traffic scenarios is received, the received measurement data is fed to the first neural network as input data in order to train the first neural network, and the first neural network which is trained on the basis of the input data is used to generate data of traffic scenarios which differ from the initial traffic scenarios. Furthermore, the disclosure relates to a method for training at least one second neural network, to a controller, to a computer program, and to a machine-readable storage medium.

First claim

The invention claimed is:

1 . A method for training a second neural network for a second environment sensor of a second vehicle, the method comprising:
receiving first measurement data of first traffic scenarios that are ascertained by at least one first environment sensor of a first vehicle, the at least one first environment sensor having at least one first mounting position and being of at least one first sensor type; transforming the first measurement data of the first traffic scenarios into a standardized world coordinate system; training the first neural network by feeding the transformed first measurement data of the first traffic scenarios to the first neural network as input data, in a manner depending on (i) a position of the first vehicle, (ii) the at least one first sensor type of the at least one first environment sensor, and (iii) the at least one first mounting position of the at least one first environment sensor; receiving second measurement data of second traffic scenarios which differ from the first traffic scenarios; transforming, using the first neural network, the second measurement data of the second traffic scenarios into an environment sensor coordinate system of the second environment sensor depending on both (i) a second mounting position of the second environment sensor and (ii) a second sensor type of the second environment sensor; and training the at least one second neural network by feeding the transformed second measurement data of the second traffic scenarios to the second neural network as input data.

2 . The method as claimed in claim 1 : wherein the first measurement data of the first traffic scenarios include traffic scenarios without traffic accidents.

3 . The method as claimed in claim 1 , wherein the second neural network is configured to operate as an object classifier.

4 . The method as claimed in claim 1 , the training the second neural network further comprising:
training the second neural network using the transformed second measurement data of the second traffic scenarios generated by the first neural network and using further measurement data ascertained by further environment sensors.

5 . The method as claimed in claim 4 , the training the at least one second neural network further comprising:
using the further measurement data ascertained by the further environment sensors to check reactions of the second neural network to traffic situations.

6 . The method as claimed in claim 1 , wherein the method is carried out by a computer program having instructions that, when carried out by one of a computer and a control device, cause the one of the computer and the control device to carry out the method.

7 . A non-transitory machine-readable storage medium that stores a computer program for training a second neural network, the computer program having instructions that, when carried out by one of a computer and a control device, cause the one of the computer and the control device to:
receive first measurement data of first traffic scenarios that are ascertained by at least one first environment sensor of a first vehicle, the at least one first environment sensor having at least one first mounting position and being of at least one first sensor type; transform the first measurement data of the first traffic scenarios into a standardized world coordinate system; train the first neural network by feeding the transformed first measurement data of the first traffic scenarios to the first neural network as input data, in a manner depending on (i) a position of the first vehicle, (ii) the at least one first sensor type of the at least one first environment sensor, and (iii) the at least one first mounting position of the at least one first environment sensor; receive second measurement data of second traffic scenarios which differ from the first traffic scenarios; transform, using the first neural network, the second measurement data of the second traffic scenarios into an environment sensor coordinate system of the second environment sensor depending on both (i) a second mounting position of the second environment sensor and (ii) a sensor type of the second environment sensor; and train the second neural network by feeding the transformed second measurement data of the second traffic scenarios to the second neural network as input data.

Assignees

• Bosch Gmbh Robert

Inventors

• Ewert Marlon Ramon

Classifications

• G06N3/088 Primary

  Non-supervised learning, e.g. competitive learning · CPC title

• G06V10/774 Primary

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/045

  Combinations of networks · CPC title