question during creating a new environment

[simplekaiyan]

Hello,I met an error in the process of running the created MARL environment ,I guess it might be the wrong variable that I give to the wrapped algorithm or function,but I am not clear about the environment's running principle ,can you give me some advice or example about it ? Thanks in advance

the error appears when running "Agent.train(configs.running_steps // configs.parallels)"

[wenzhangliu]

It appears that the observation data type in your custom environment isn’t compatible with XuanCe. Please ensure that the observations returned from env.step() and env.reset() take the form of a dictionary, for example:

observations = {
'agent_0': np.array(...),
'agent_1': np.array(...),
...
}

[simplekaiyan]

Thanks for your reply!
I checked the data in function "reset" and "step" ,but the data type seem to be correct

actually,I set a breakpoint at the beginning of the function "step", but the error occurs before running the "step" function,is there anything wrong with my environment's structure?

[Golden-year]

Thanks for your reply! I checked the data in function "reset" and "step" ,but the data type seem to be correct

actually,I set a breakpoint at the beginning of the function "step", but the error occurs before running the "step" function,is there anything wrong with my environment's structure?

Check whether your reset() method is correct. In agents_marl.py, there is a try: self.envs.reset() except: pass. So if your reset() method has an error, it will not throw an exception.

[simplekaiyan]

Thanks,I checked the code and the problem does arise in the setup method of the "reset" function