US 12,456,048 B2
Determining an output signal by aggregating parent instances
Martin Schiegg, Korntal-Muenchingen (DE); and Xiahan Shi, Renningen (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Oct. 26, 2020, as Appl. No. 17/080,365.
Claims priority of application No. 19206119 (EP), filed on Oct. 30, 2019; and application No. 19206484 (EP), filed on Oct. 31, 2019.
Prior Publication US 2021/0133567 A1, May 6, 2021
Int. Cl. G06N 3/08 (2023.01); G05B 13/02 (2006.01)
CPC G06N 3/08 (2013.01) [G05B 13/027 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A computer-implemented method of training a function for use in controlling an autonomous or semiautonomous vehicle operating in an environment, the function mapping an input instance including one or more sensor measurements to an output signal for use in the controlling, the function being parameterized by a set of parameters, the set of parameters including representations of multiple reference instances, the method comprising the following steps:
accessing a training dataset for training the function, the training dataset including multiple training input instances and corresponding training output signals;
training the function by:
determining an output signal of the function for a training input instance of the training input instances, including:
identifying a number of reference instances of the multiple reference instances as parent instances of the training input instance based on a similarity between the training input instance and the multiple reference instances determined according to a similarity function,
determining an aggregate latent representation for the training input instance based on aggregating representations or reference output signals of the identified parent instances using an aggregation function, and
determining the output signal based on the aggregate latent representation for the training input instance using an output function;
deriving a training signal by comparing the determined output signal for the training input instance to the corresponding training output signal for the training input instance;
adjusting at least a representation of a reference instance of the reference instances according to the training signal to obtain a reference instance not included in the training dataset; and
determining a confidence value of the function for the input instance based on (i) sampling multiple output signals or (ii) determining a probability distribution of the output signal;
wherein the autonomous or semi-autonomous vehicle is controlled based on the output signal and the confidence value.