Mix action channels are not supported in DDPG

5 次查看（过去 30 天）
Sania Gul
Sania Gul 2024-9-18
评论： Sania Gul 2024-9-18
My action space is mixed. Except one, all variables are continuous. So, I have defined the action apace as shown in the code below for a DDPG agent. But I get the error as shown in the attached image.
actInfoCont = rlNumericSpec([255 1],"UpperLimit",1,"LowerLimit",0); % (0 to 1 probabilistic mask)
actInfoDIsc = rlFiniteSetSpec([0 3.141592654]);
actInfo = [actInfoCont actInfoDIsc];

请先登录，再回答此问题。

采纳的回答

Ayush Aniket
Ayush Aniket 2024-9-18
编辑：Ayush Aniket 2024-9-18
Hi Sania,
Currently, Reinforcement Learning Toolbox does not support mixed action spaces. As an alternative you could consider training two agents, one for the continuous action and another for the discrete. Refer to the following MATLAB Answer: https://www.mathworks.com/matlabcentral/answers/2015776-reinforcement-learning-agent-for-mixed-action-space

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Reinforcement Learning 的更多信息

标签

Community Treasure Hunt

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

Start Hunting!

Translated by