maddpg训练问题

[2568754202]

我用maddpg训练一个unity应用，配置了多个大脑，输出action都是连续的，但是运行一开始就报错，list has no attributes is_continuous。

[BlueFisher]

@2568754202 we'll check and work on it.

[2568754202]

I found the problem , the behavior name must like '2#Brain'. But there is a new bug that too many values to unpack (expected 5).

[2568754202]

s, visual_s, _, _, _ = env.reset()
s, visual_s = map(data_change_func, [s, visual_s])
we must set reset_config={.....}
how to set?

[StepNeverStop]

@2568754202 This repo is now under deep reconstruction, and maddpg has lots of details out-of-date. We highly recommend you not use this repo to training multi-agent environments for now.

[2568754202]

Thank you! But there is only a Multi-Agent training algorithms in your RLS.

[StepNeverStop]

@2568754202 yes, my main research direction is single-agent RL. So the progress of multi-agents algorithms is slow and priority is low. But I'll reconstruct/optimize MADDPG asap. :)

[2568754202]

Your work is great! Thank you so much