Title:
LEARNING SYSTEM AND LEARNING METHOD FOR OPERATION INFERENCE LEARNING MODEL FOR CONTROLLING AUTOMATED DRIVING ROBOT
Document Type and Number:
WIPO Patent Application WO/2022/059484
Kind Code:
A1
Abstract:
[Problem] To provide a learning system and a learning method for an operation inference learning model which controls an automated driving robot (drive robot), capable of easily suppressing a deterioration in the learning accuracy of the operation inference learning model resulting from differences in processing times between a vehicle model and an actual vehicle, when performing machine learning of the operation inference learning model with the vehicle model as the object of operation execution. [Solution] A learning system 10 for performing machine learning of an operation inference learning model 70 which controls an automated driving robot 4 is provided with a real environment 1 provided with a vehicle 2 and the automated driving robot 4, mounted in the vehicle 2, and the operation inference learning model 70 for inferring operations of the vehicle 2 in such a way as to cause the vehicle 2 to travel in accordance with a prescribed command vehicle speed, on the basis of the traveling state of the vehicle 2 including the vehicle speed, wherein the automated driving robot 4 causes the vehicle 2 to travel on the basis of inferred operations inferred by the operation inference learning model 70, and wherein: the learning system 10 is provided with a virtual device model 50 provided with a vehicle model 52 which is set in such a way as to simulate the operations of the vehicle 2, and which outputs a simulated traveling state, being the traveling state imitating the vehicle 2, on the basis of the inferred operations; and the virtual device model 50 is provided with an adjusting unit 55 which adjusts the time from when the inferred operations are inferred until the same are used by the virtual device model 50, and from when the simulated traveling state is output until the same is applied to the operation inference learning model 70, on the basis of the actual time required from when the simulated traveling state is output by the vehicle model 52 on the basis of the inferred operations, upon input of the inferred operations, the operation inference learning model 70 is subjected to machine learning by the application of the simulated traveling state to the operation inference learning model 70, and the inferred operations are inferred in the real environment 1, until the inferred operations are input into the automated driving robot 4, the vehicle 2 is operated and travels, the traveling state of the vehicle 2 is acquired, and the traveling state is applied to the operation inference learning model 70.
More Like This:
Inventors:
KANAZASHI YASUHIRO (JP)
YOSHIDA KENTO (JP)
YOSHIDA KENTO (JP)
Application Number:
PCT/JP2021/032055
Publication Date:
March 24, 2022
Filing Date:
September 01, 2021
Export Citation:
Assignee:
MEIDENSHA ELECTRIC MFG CO LTD (JP)
International Classes:
F02D29/02; G01M17/007
Foreign References:
| JP2020056737A | 2020-04-09 | |||
| CN110795818A | 2020-02-14 | |||
| US20190278272A1 | 2019-09-12 | |||
| US20060155734A1 | 2006-07-13 |
Attorney, Agent or Firm:
SONODA & KOBAYASHI INTELLECTUAL PROPERTY LAW (JP)
Download PDF: