Question about MASAC Algorithm in XuanCe and Discrete Action Space Support

[SunSkyLH]

Dear XuanCe Development Team,

Hello!

Thank you for your contributions to the field of multi-agent reinforcement learning! I noticed that the MASAC algorithm in the XuanCe project references the paper titled "Decomposed Soft Actor Critic Method for Cooperative Multi-Agent Reinforcement Learning". This paper explicitly states that its proposed mSAC method supports discrete action spaces. However, it seems that the current implementation of MASAC in XuanCe only supports continuous action spaces.

Could you kindly clarify whether this is due to technical considerations or development priorities? Are there any plans to support discrete action spaces in the future? I am very interested in this topic and look forward to your response.

Best regards！

[wenzhangliu]

Thank you for your interest in XuanCe! During our development of the SAC algorithm, we discovered significant differences between its implementation in discrete action spaces and continuous action spaces. As a result, the development of MASAC for discrete action spaces has been delayed. However, we plan to support this feature very soon. Please stay tuned, and thank you for your patience!