Title:
DEVICE AND METHOD FOR DETERMINING SAFE ACTION TO BE EXECUTED BY TECHNICAL SYSTEM
Document Type and Number:
Japanese Patent JP2023129360
Kind Code:
A
Abstract:
To provide a method for training a machine learning system for determining a control signal characterizing an action to be executed by a technical system, a training system, a control system, a computer program, and a storage medium.SOLUTION: In a training method, a step of obtaining a safe action to be executed by a technical system includes: obtaining a state signal characterizing a state of an environment (101); determining, by a parametrized policy module of a machine learning system, the distribution of potentially unsafe actions that may be executed by the technical system (102); sampling a potentially unsafe action from the distribution (103); and obtaining the safe action by a safety module of the machine learning system (104). The method includes the steps of: determining a loss value based on the state signal and the safe action (105); and training the machine learning system by updating parameters of the policy module according to the gradient of the loss value with respect to the parameters (106).SELECTED DRAWING: Figure 2
More Like This:
| JP6695490 | Learning data management device and learning data management method |
| JP2023128729 | ARITHMETIC PROCESSING UNIT |
| WO/2020/018394 | VEHICLE-DATA ANALYTICS |
Inventors:
PHILIPP GEIGER
CHRISTOPH-NIKOLAS STRAEHLE
CHRISTOPH-NIKOLAS STRAEHLE
Application Number:
JP2023031538A
Publication Date:
September 14, 2023
Filing Date:
March 02, 2023
Export Citation:
Assignee:
BOSCH GMBH ROBERT
International Classes:
G06N20/00; G05B9/02
Attorney, Agent or Firm:
Einzel Felix-Reinhard
Taku Morita
Junichi Maekawa
Hideo Nagashima
Ueshima class
Taku Morita
Junichi Maekawa
Hideo Nagashima
Ueshima class