Control switching device for abnormality prevention in multiple terminals
There is provided a control device that prevents the occurrence of an abnormality in control using artificial intelligence in the case where an abnormality occurs in another terminal that performs the control using the artificial intelligence. A control means (a first control unit) performs the control using artificial intelligence. The control means (second control unit) performs control different from the artificial intelligence. A control switching unit (a switching determination means, a switch) is configured to, when another terminal having the control means (the first control unit) is normal, cause the own control means (the first control unit) to perform the control using artificial intelligence, and when an abnormality occurs in another terminal having the control means (the first control unit), cause the control means (the second control unit) to perform the control different from artificial intelligence.
1. A control device comprising:
a plurality of terminals, each of the plurality of terminals including
a first control algorithm that performs control using machine learning;
a second control algorithm that performs a different control, which is different from the machine learning; and
a control switch that,
during control execution using the first control algorithm on any of the plurality of terminals,
when an abnormality occurs during the control execution
controlling, via the control switch, a terminal of the plurality of terminals where the abnormality occurred,
while switching from the first control algorithm to the second control algorithm in the terminal where the abnormality occurred,
controlling any other terminals of the plurality of terminals where the abnormality has not occurred, and
switching in the any other terminals from the first control algorithm to the second control algorithm.
2. The control device according to claim 1 , wherein the plurality of terminals further comprise
an input/output device that receives an abnormality notification indicating an abnormality in the other terminal, wherein
the control switch is configured to, when the abnormality notification is not received, cause the first control algorithm to perform the control using the machine learning, and when the abnormality notification is received, cause the second control algorithm to perform the different control.
3. The control device according to claim 2 , wherein
the first control algorithm
learns control parameter values indicating parameters of a function for determining an output value of the control using the machine learning from an input value of the control using the machine learning, and
calculates the output value based on the input value and the learned control parameter values.
4. The control device according to claim 1 , wherein
the different control is PID control or manual control.
5. The control device according to claim 3 , wherein the plurality of terminals further comprise
a third control algorithm that, when a state in which at least one of the control parameter values is not within a predetermined range lasts for a predetermined time, determines that the first control algorithm is abnormal, wherein
the control switch is configured to,
when the abnormality notification is not received and it is not determined that the first control algorithm is abnormal, cause the first control algorithm to perform the control using the machine learning, and
when the abnormality notification is received or it is determined that the first control algorithm is abnormal, cause the second control algorithm to perform the different control.
6. The control device according to claim 3 , wherein the plurality of terminals further comprise
a memory that stores the control parameter values learned by the first control algorithm, wherein
the input/output device receives the control parameter values learned by the server, and
the first control algorithm updates the control parameter values stored in the memory to the control parameter values learned by the server in a period during which the second control algorithm performs different control.
7. The control device according to claim 6 , wherein
the control switch switches from the different control to the control using the machine learning after the updating of the control parameter values.
8. The control device according to claim 5 , wherein the plurality of terminals further comprise the input/output device that, when it is determined that the first control algorithm is abnormal, transmits the abnormality notification.
9. The control device according to claim 1 , wherein
a target of the control using the machine learning and the different control is a machine.
10. The control device according to claim 9 , wherein
the machine is a robot, an autonomous vehicle, or a flying object.
11. A control system comprising the control device according to claim 8 and a server, wherein
the abnormality notification is provided via the server to the plurality of terminals in which it is not determined that the first processor is abnormal.
12. A control system including the control device according to claim 8 , wherein
the abnormality notification is provided through inter-terminal communication to the plurality of terminals in which it is not determined that the first control algorithm is abnormal.
13. The control system according to claim 11 , wherein
the server transmits to each of the plurality of terminals a switching synchronization flag for synchronizing a timing for switching from the control using the machine learning to the different control,
the input/output device of each of the plurality of terminals receives the switching synchronization flag, and
the control switch of each of the plurality of terminals is configured to, when the abnormality notification is received and the switching synchronization flag is received, switch from the control using the machine learning-to the different control.
14. The control system according to claim 11 , wherein
the server transmits to each of the plurality of terminals a period signal indicating a periodic signal,
the input/output device of each of the plurality of terminals receives the period signal, and
the control switch of each of the plurality of terminals is configured to, even when the period signal is not received, switch from the control using the machine learning to the different control.