Reinforcement Learning Agents generating zero episode

Hello Matlab community,

I faced an issue for training a multiple agent problem in Matlab/Simulink. I tried to solve a very simple problem; however, training stops at episode 1.

Suppose that we have three discrete variables A=[1 2 3], B=[1 2 3], C=[1 2 3].

Reward function = A*B*C;

Observation= A+B+C;

I tried different parameters but it didn’t work. I attached the sample file for the reference. Very appreciate it if you can suggest potential solution to solve convergence issue.

Thanks for your time 😊

Answer 1

[Ari Biswas]

There is an issue with the way you specified the reset function. Your function resetRobots should return a Simulink.SimulationInput object, which is also an input argument to the function. So for e.g. the correct function signature should be:

function in = resetRobots(in, var1, var2, var3)
% write reset code
end

See this example to see how reset functions can be defined.

https://www.mathworks.com/help/reinforcement-learning/ug/water-tank-reinforcement-learning-environment-model.html#WaterTankEnvironmentModelExample-5

[Hamid Fazeli]

Hi Ari,

Thank you so much for the suggestion. The problem was reset function. It was fixed :)