Wen-Zhang Liu

Currently, the MASAC cannot deal with multi-agent environments with discrete action spaces.

The MASAC algorithm for discrete action spaces differs significantly from that for continuous action spaces. Currently, the MASAC implementation in Xuance does not support discrete action spaces directly, but we...

> Hope to add support for these types of action_space and obs_space. > > Currently, if I have to use these types of action_space and obs_space (Tuple or Dict), can...

For multi-robot warehouse (RWARE) environment, the version of gym is suggeted to be 0.21 according its the offcial instruction [(https://github.com/semitable/robotic-warehouse/blob/master/setup.py).](https://github.com/semitable/robotic-warehouse/blob/master/setup.py). Hence, when you run this environment with XuanCe, you should...

目前new_env_mas.py里给出的针对同质智能体（所有智能体的各种属性都一样）的示例。但是你提到的情况应该没有影响，只要每个智能体的obs_shape一样，在外部应该是不区分取值范围的。如果你需要在外部访问每个智能体的观测范围，不妨在这个文件中新增self.obs_space1，self.obs_space2这样的成员变量。

貌似是因为纵坐标显示范围太大

建议先看英文的文档，中文文档还没同步更新

自定义环境的教程在文档中有详细描述，请参考该链接：[https://xuance.readthedocs.io/en/latest/documents/usage/new_envs.html](https://xuance.readthedocs.io/en/latest/documents/usage/new_envs.html).

Thank you very much for your valuable suggestions. We will enhance the code by carefully addressing the issues you raised.

> At present, we can see in the documentation that this project has implemented a very rich agent model algorithm and the adaptation of different types of Env. However, it...