using Q learning agent for continuous observation space

Hello,

I have a reinforcement learning problem where the observation is the error of closed loop feedback and it is continuous, and discrete action space.

I want to use Q learning as its document tells It can handle both discrete/continuous observations.

but Im a little bit confused about making critic using rlQValueRepresentation which its syntax mostly uses either a table or deep neural network,

and they are inappropriate for my work, as I didnt find any example like this in Mathworks website, Is there anyone who can help me on this?

Answer 1

[Stephan]

You also are allowed to write a custom critic function:

https://de.mathworks.com/help/reinforcement-learning/ref/rlqvaluerepresentation.html#mw_e269c528-2e43-4318-b3c1-dedae1321529

https://de.mathworks.com/help/reinforcement-learning/ref/rlqvaluerepresentation.html#mw_da601ea7-bc2c-4f1a-b3bc-05e142b4f764