Using time as a negative reward in RL toolbox

1 次查看（过去 30 天）
Amin Moradi
Amin Moradi 2022-2-24
回答： Kartik Saxena 2023-11-30
I want to use RL toolbox to train a DQN agent. Right now, i'm using the related step_function to implement the reward function. The problem is I don't know how to punish the agent for taking too long to do the objective. How should I add time to my reward function in this toolbox? Your help is appreciated.
function [NextObs,Reward,IsDone,LoggedSignals] = WW6_StepFunction_genloss(Action,LoggedSignals)
a = Action;
obj=4;
d=[1 2];
state = LoggedSignals.State;
[next_state, ~, genloss]=attack_eff_WW6(state, a, d);
LoggedSignals.State = next_state;
NextObs = LoggedSignals.State;
Down=nnz(~next_state);
IsDone = Down==11;
Reward=genloss;
end

请先登录，再回答此问题。

回答（1 个）

Kartik Saxena
Kartik Saxena 2023-11-30
Hi,
I understand that you want to add time penalty in the reward function to punish it for taking too long.
The example given below in the MathWorks documentation would be useful for this purpose:
https://www.mathworks.com/help/reinforcement-learning/ug/create-matlab-environments-using-custom-functions.html
You can refer to it and introduce penalty in your reward function by deducting from the reward as per your requirements, instead of adding '1'.
I hope this resolves your issue.

类别

在 Help Center 和 File Exchange 中查找有关 Introduction to Installation and Licensing 的更多信息

标签

产品


版本

R2021b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by