NousResearch/atropos
1
0
Fork 0
mirror of https://github.com/NousResearch/atropos.git synced 2026-04-19 12:57:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Update README.md

Browse source
This commit is contained in:
hjc-puro 2025-05-08 15:12:44 -04:00 • committed by GitHub
parent 7b0028d0ba
commit 515c4cb6ab
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
atroposlib/envs/README.md
View file

@ -6,7 +6,7 @@ The `BaseEnv` class (located in `trajectoryhandler/envs/base.py`) provides a fou
Every environment in Atropos is a microservice that generates rollout data async from whatever trainer you attach to it. Environments (possibly many at once) send data to the [Atropos API server](https://github.com/NousResearch/atropos/tree/main/atroposlib/api) which sequesters rollout data. Your trainer of choice grabs batches of data from the API and backpropagates. Every environment in Atropos is a microservice that generates rollout data async from whatever trainer you attach to it. Environments (possibly many at once) send data to the [Atropos API server](https://github.com/NousResearch/atropos/tree/main/atroposlib/api) which sequesters rollout data. Your trainer of choice grabs batches of data from the API and backpropagates.
![image](https://github.com/user-attachments/assets/171d21f2-bbd8-41cb-a56a-35ec30ee4efc) ![image](https://github.com/user-attachments/assets/9fb3fcdf-2932-4fde-acb7-3a1523e96f49)
Unlike other popular alternatives like Gymnasium which model environments as [MDPs](https://arxiv.org/abs/2412.05265), we think about environments as dataloaders and do not make any assumptions about how a trajectory is produced. For multi-agent for example, this means our design is agnostic to [AEC](https://pettingzoo.farama.org/api/aec/) vs. [POSG](https://pettingzoo.farama.org/api/parallel/) - both are supported out of the box. Unlike other popular alternatives like Gymnasium which model environments as [MDPs](https://arxiv.org/abs/2412.05265), we think about environments as dataloaders and do not make any assumptions about how a trajectory is produced. For multi-agent for example, this means our design is agnostic to [AEC](https://pettingzoo.farama.org/api/aec/) vs. [POSG](https://pettingzoo.farama.org/api/parallel/) - both are supported out of the box.
To achieve this generality, our environment abstraction deviates from other open source alternatives in several key ways. To achieve this generality, our environment abstraction deviates from other open source alternatives in several key ways.