# MultiAgent env wrong structures

**URL:** <https://discuss.ray.io/t/multiagent-env-wrong-structures/20782>\
**Category:** RLlib\
**Created:** [November 27, 2024, 2:18pm UTC](https://discuss.ray.io/t/multiagent-env-wrong-structures/20782 "2024-11-27T14:18:32Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Adrielram](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/adrielram/32/7356_2.png) [@Adrielram](https://discuss.ray.io/u/Adrielram)\
**Post date:** [November 27, 2024, 2:18pm UTC](https://discuss.ray.io/t/multiagent-env-wrong-structures/20782/1 "2024-11-27T14:18:32Z")

</div>

**How severe does this issue affect your experience of using Ray?**

- High: It blocks me to complete my task.

Hi everyone!  
I’m new using Ray, I chose it because I believe it’s a great option for making RL projects.  
I’m now trying to train a multiagent enviroment, like simple\_tag from pettingzoo, but I cant make it work.  
I always got this error ‘substructures are different size or nested’ and when I look at them, one of them is an array of (for example) 16 elements, and the other one is a dict, with keys ‘agent\_0’, ‘agent\_1’ and an array of 16 elements for each key.  
I think the first structure is the one that the enviroment is giving, and larger one, the one ray is expecting.

This is the config:  
config = {  
“env”: “simple\_spread”,  
“framework”: “torch”, # o “tf” si usas TensorFlow  
“multiagent”: {  
“policies”: {  
“agent\_0”: (None, env\_creator({}).observation\_space, env\_creator({}).action\_space, {}),  
“agent\_1”: (None, env\_creator({}).observation\_space, env\_creator({}).action\_space, {}),  
“agent\_2”: (None, env\_creator({}).observation\_space, env\_creator({}).action\_space, {}),  
},  
“policy\_mapping\_fn”: policy\_mapping\_fn,  
},  
“num\_workers”:1,  
}

What can I do? is that an enviroment problem?  
Thanks!

---

<div class="post-metadata">

**Author:** ![Adrielram](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/adrielram/32/7356_2.png) [@Adrielram](https://discuss.ray.io/u/Adrielram)\
**Post date:** [November 28, 2024, 8:34pm UTC](https://discuss.ray.io/t/multiagent-env-wrong-structures/20782/2 "2024-11-28T20:34:21Z")

</div>

I will give some more information:  
More specifically: Substructure “type=OrderedDict str=OrderedDict([(‘adversary\_0’, array([ 0.01255419, -0.6857402 , 3.5040925 , -0.5908824 , -0.28089884,  
(PPO pid=13468) -0.42432716, -0.98887783, -0.8802794 , -0.559584 , -0.19044599,  
(PPO pid=13468) -0.71634007, -0.06078899, -0.43675077, -0.15962653, 0.9411768 ,  
(PPO pid=13468) -2.162692 ], dtype=float32)), (‘adversary\_1’, array([-1.5311705 , 1.3489779 , 0.42647207, -1.2930255 , 0.34891367,  
(PPO pid=13468) 0.35973787, 0.9409657 , 0.31296352, -1.4111371 , -0.66821265,  
(PPO pid=13468) 1.8022188 , 0.07645752, 0.5687924 , -0.36956072, 0.8243999 ,  
(PPO pid=13468) 0.09302095], dtype=float32)), (‘adversary\_2’, array([ 1.3076917 , 1.1348339 , -0.41443637,  
1.0638967 , 0.05962287,  
(PPO pid=13468) -0.8662066 , -0.15366012, -1.1731756 , 0.46935457, -1.4058455 ,  
(PPO pid=13468) 0.27503854, 0.27102682, -0.6621483 , -0.7810565 , -0.58224857,  
(PPO pid=13468) 0.39142346], dtype=float32)), (‘agent\_0’, array([ 0.27529287, 1.212923 , -1.0894041 , -0.28833956, -0.4597624 ,  
(PPO pid=13468) 0.19036998, 1.4900975 , 0.06139584, 1.9097888 , 0.9566886 ,  
(PPO pid=13468) -0.8215733 , 0.16532217, -1.4654043 , -0.7305452 ], dtype=float32))])” is a sequence, while substructure “type=ndarray str=[ 0. 0. -0.89107406 0.09816181 1.5538627 -0.97209406  
(PPO pid=13468) 0.43329492 -0.93592656 0.16501105 -0.51356584 -0.07702168 -0.28925264  
(PPO pid=13468) 0.70538473 0.14393562 0. 0. ]” is not