US 12,475,710 B2
Generating verifiably realistic measurement data
Karim Said Mahmoud Barsim, Stuttgart (DE); and Melih Kandemir, Stuttgart (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Jun. 14, 2021, as Appl. No. 17/346,858.
Claims priority of application No. 20184909 (EP), filed on Jul. 9, 2020.
Prior Publication US 2022/0012597 A1, Jan. 13, 2022
Int. Cl. G06V 20/56 (2022.01); G06F 18/213 (2023.01); G06N 3/047 (2023.01); G06N 3/088 (2023.01); G06V 10/77 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06N 3/084 (2023.01)
CPC G06V 20/56 (2022.01) [G06F 18/213 (2023.01); G06N 3/047 (2023.01); G06N 3/088 (2013.01); G06V 10/7715 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06N 3/084 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A system comprising:
a generator for automated training of an image classifier of a machine for autonomous operation by generating one or more records of artificial measurement data, the generator comprising:
a processor programmed to:
implement a trained neural network to map each of a plurality of input vectors of a neural network latent space to a respective set of distribution parameters, wherein the set of distribution parameters characterize a random distribution of potential measurement data that is generatable from the respective input vector;
use a random or pseudo-random number generator as a source of randomness to perform a sampling to select samples from the random distributions;
generate the one or more records of measurement data from the input vectors using the selected samples of the random distributions; and
automatedly train the image classifier of the machine for autonomous operation using the one or more records; and
a processor system, wherein the processor system is configured to:
use the trained image classifier to process acquired sensor images of a physical environment of a vehicle or robot;
generate an actuation signal controlling the vehicle or robot based on a result of the processing of the acquired sensor images; and
use the generator to improve a perception system of the vehicle or robot by generating domain-specific synthetic sensor data that enhances the image classifier's ability to identify objects or events in the physical environment.