US 12,361,322 B2
Device and method for training a gaussian process state space model
Hon Sum Alec Yu, London (GB); Dingling Yao, Rottenburg am Neckar (DE); Christoph Zimmer, Korntal (DE); and The Duy Nguyen-Tuong, Calw (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Jan. 14, 2022, as Appl. No. 17/648,069.
Claims priority of application No. 10 2021 200 569.1 (DE), filed on Jan. 22, 2021.
Prior Publication US 2022/0245521 A1, Aug. 4, 2022
Int. Cl. G06N 20/00 (2019.01); B25J 9/16 (2006.01); G06N 5/04 (2023.01)
CPC G06N 20/00 (2019.01) [B25J 9/163 (2013.01); G06N 5/04 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A method for training a Gaussian process state space model, the Gaussian process state space model describing a correlation between selected control parameters of a plurality of control parameters for controlling a robotic device and measured output variables of the robotic device assigned in each case, the Gaussian process state space model including a transition function and an output variable prediction function, the transition function mapping an input state of the robotic device and each control parameter of the plurality of control parameters for controlling the robotic device according to a transitional normal distribution assigned to the control parameter and to the input state onto a predicted output state, and the output variable prediction function mapping the predicted output state according to an output variable normal distribution assigned to the predicted output state onto an output variable of the robotic device, the method comprising:
for each control parameter of the plurality of control parameters:
ascertaining the transitional normal distribution assigned to the respective control parameter and to a present input state of the robotic device,
ascertaining the output variable normal distribution assigned to the respective predicted output state, and
ascertaining a respective value of a piece of mutual information between the ascertained output variable normal distribution and the ascertained transitional normal distribution;
selecting a control parameter of the plurality of control parameters having a highest value of mutual information as a new control parameter;
controlling the robotic device using the new control parameter and measuring an output variable of the robotic device assigned to the new control parameter; and
training the Gaussian process state space model using the new control parameter and the assigned measured output variable of the robotic device in such a way that a difference between the measured output variable of the robotic device and a mean value of the output variable normal distribution ascertained for the new control parameter is reduced.