Reset Terminal States in Grid World example using RL Toolbox

Using the Grid World example in the Reinforcement Learning Toolbox, I would like to reset the Terminal States at the end of each episode as well as the Current State (starting location). Would this be possible using the toolbox and if so, how would I code this?

Answer 1

[Asvin Kumar]

There are a couple in different ways in which custom environments can be defined in MATLAB. Have a look at https://www.mathworks.com/help/reinforcement-learning/ug/create-matlab-environments-for-reinforcement-learning.html?s_tid=answers_rc2-1_p4_BOTH#mw_c167fd11-8f6f-496b-a43b-67328b10568d

Among those three ways, you’re looking either for the second or third option.

For an example on the second approach visit: https://www.mathworks.com/help/reinforcement-learning/ug/create-custom-reinforcement-learning-environment-in-matlab.html#CreateMATLABEnvironmentUsingCustomFunctionsExample-3

For an example on the third approach visit: https://www.mathworks.com/help/reinforcement-learning/ug/create-custom-matlab-environment-from-template.html#mw_188b5ed0-c7bd-46c4-a0c5-25973cebf5c0

Both these examples are written for the cart pole environment. Since your requirement is to modify the reset behaviour in a Grid World environment, you will need to modify the properties accordingly. A list of properties of the Grid World is given at https://www.mathworks.com/help/reinforcement-learning/ug/create-custom-grid-world-environments.html. As you will see on the page, the ‘CurrentState’ and ‘TerminalStates’ property are not Read-Only and can therefore be modified in your custom reset function.