Reinformence Learning Tool box - where to get Action?
1 view (last 30 days)
Show older comments
Hello,
I have been following the tutorial on this example to customize the environment. for setting up the step function, I noticed that in the following statement:
function [NextObs,Reward,IsDone,LoggedSignals] = myStepFunction(Action,LoggedSignals)
the input "Action" was never mentioned elsewhere in the tutorial on how it be generated. My question is that I have a continuous action space; should I call in main.m to first generate "Action"? or it is a default output from the RL toolbox?
Now I treat it as a default output from the RL training, but I kept on having the issue of action output saturating at the max and min value, although I put a tanh and scaling layer at the end of the actor Network and adjusted the error. If I test the actions using getAction(agent,rand(obs_dim)) they are within the defined boundary without saturation; but when I check the simulation.Action.Action.data from simulation result everything goes to 1e10 or so. My observation space is also continuous.
Thank you!
0 Comments
Answers (0)
See Also
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!