Title:
TRAINING SYSTEM, TRAINING METHOD, TRAINING PROGRAM, AND AUTONOMOUS CONTROL DEVICE
Document Type and Number:
WIPO Patent Application WO/2024/053615
Kind Code:
A1
Abstract:
Provided is a training system comprising a processing unit for the reinforcement training of an agent that controls an object to be controlled, wherein the processing unit repeatedly executes steps up to updating a policy as evaluation steps on the basis of remuneration obtained by the agent in a prescribed state by executing a prescribed action in a prescribed environment, the updating of the evaluation steps being defined as event-driven, where a transition of the object being controlled from a first environment state to a predefined second environment state when the agent executes a prescribed action is employed as an event.
Inventors:
KARAKAMA SOTARO (JP)
MATSUNAMI NATSUKI (JP)
ITO MASAYUKI (JP)
MATSUNAMI NATSUKI (JP)
ITO MASAYUKI (JP)
Application Number:
PCT/JP2023/032264
Publication Date:
March 14, 2024
Filing Date:
September 04, 2023
Export Citation:
Assignee:
MITSUBISHI HEAVY IND LTD (JP)
International Classes:
G06N20/00
Foreign References:
| JP2020190853A | 2020-11-26 | |||
| JP2022033570A | 2022-03-02 |
Attorney, Agent or Firm:
SAKAI INTERNATIONAL PATENT OFFICE (JP)
Download PDF: