Training an agent of reinforcement learning as a motor's controller, but Matlab doesn't not do training at all?

1 view (last 30 days)
Hi, I'm trying to use reinforcement learning to train an agent as a controller of a motor.
My motor's transfer function is:
% The input and output of this transfer function are current and speed, respectively.
69.4/(s+69.4)
After reading this example, I only made some minor modifications:
1. Replace the watertank model with my motor's transfer function like below:
1.png
2. Set the sample time Ts and simulation time Tf to 1/8000 and 10, respectively.
2.png
However, the result is:
3.png
It doesn't work :(
The reward is always zero, and I don't know why.
Could anybody give me advices on how to deal with this problem?
Very thankful in advance.

Answers (2)

Emmanouil Tzorakoleftherakis
Edited: Emmanouil Tzorakoleftherakis on 16 Sep 2019
Hello,
It is hard to pinpoint the problem exactly without a repro model, but sounds like training stops prematurely. Can you remove the "Stop Simulation" subsystem from your model? I suspect that this flag is always set to true for the new plant.

Smita Sadhukhan
Smita Sadhukhan on 14 Jul 2022
Edited: Smita Sadhukhan on 14 Jul 2022
I have tried removing the stop simulation block but the it is not giving proper reward to train the model.After traing the output is not even going close to the set point. Can you please look into this problem and help me.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!