rlNumericSpec and rlFiniteSetSpec multi channell environment + Qtable

I'm seeking to develop a custom environment for a (partially observable) Q learning problem. In particular I'm interested in been able to use in my environment a DT markov chain and some differential equations ( which estimante the state of the Markov chain). Something like have observations which are rlNumericSpec and rlFiniteSetSpec. In other words I would need my observations to have multiple channels and make use of a Qtable

Is there some best practice/ideas on how to accomplish this?

Cheers

Answer 1

[Aditya]

Developing a custom environment for a partially observable Q-learning problem with a discrete-time (DT) Markov chain and differential equations is a complex task. However, MATLAB provides the tools necessary to create such an environment. Here are some best practices and ideas to help you accomplish this:

1. Define the Environment interface
2. Create the Custom Environment
3. Create the Q-Table
4. Implement the Learning Algorithm
5. Create the MATLAB Environment Object