How can I have several actions for a DQN in the Reinforcement Learning Toolbox?

I'm trying to define the output of a DQN agent with a custom environment, and can't use the actionInfo = rlFiniteSetSpec() correctly.

I'm trying to control 3 actuators that will receive commands 0 and 1.

I always get number of actions = 1.

And the documentation is not clear as it's a new toolbox.

Any suggestions?

[Huzaifah Shamim]

Where you able to get this to work?

[Rusczak]

Yes,

I followed Emmanouil's suggestion:

actInfo = rlFiniteSetSpec({[0; 0; 0]; [1; 1; 1]; ...

[1; 0; 0]; [0; 1; 0]; [0; 0; 1]; ...

[1; 1; 0]; [0; 1; 1]; [1; 0; 1]})

[Huzaifah Shamim]

Oh nice okok. What custom environment where you trying to make?

Also how should I approach it if i have 3 agents (like your three actuators) but 12 actions could be applied to them?

Answer 1

[Emmanouil Tzorakoleftherakis]

If you type

help rlFiniteSetSpec

the second example is

spec = rlFiniteSetSpec({[0,1];[1,1];[1,2];[1,3]})

If you define all possible combinations of the discrete inputs in a cell array as above, that should work (think of a single action as one possible combination of your 3 actuator commands).

I hope this helps.