https://github.com/futureagi/xenoverse
Xenoverse is a collection of randomized RL, Language, and general-purpose simulation environments, designed for training General-Purpose Learning Agents (GLAs).
https://github.com/futureagi/xenoverse
agi ai artificial-general-intelligence artificial-intelligence general-purpose-in-context-learning general-purpose-learning-agents in-context-learning in-context-reinforcement-learning meta-learning meta-reinforcement-learning meta-rl reinforcement-learning rl rl-environment rl-envs simulation simulator
Last synced: 3 months ago
JSON representation
Xenoverse is a collection of randomized RL, Language, and general-purpose simulation environments, designed for training General-Purpose Learning Agents (GLAs).
- Host: GitHub
- URL: https://github.com/futureagi/xenoverse
- Owner: FutureAGI
- License: apache-2.0
- Created: 2023-02-23T09:54:29.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2025-09-26T09:04:29.000Z (4 months ago)
- Last Synced: 2025-09-26T11:18:07.355Z (4 months ago)
- Topics: agi, ai, artificial-general-intelligence, artificial-intelligence, general-purpose-in-context-learning, general-purpose-learning-agents, in-context-learning, in-context-reinforcement-learning, meta-learning, meta-reinforcement-learning, meta-rl, reinforcement-learning, rl, rl-environment, rl-envs, simulation, simulator
- Language: Python
- Homepage: https://futureagi.github.io/
- Size: 1.94 MB
- Stars: 18
- Watchers: 2
- Forks: 7
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Xenoverse: Toward Training General-Purpose Learning Agents (GLA) with Randomized Worlds
## xenoverse instead of a single universe
The recent research indicates that the generalization ability of learning agents is primarily dependent on the diversity of training environments. However, the real-world poses a significant limitation on the diversity itself, e.g., physical laws, the gravitational constant is almost constant. We believe this limitation is serious bottleneck to incentivize artificial general intelligence (AGI).
Xenoverse is a collection of extremely diverse worlds by procedural generation based on completely random parameters. We propose that AGI should not be trained and adapted in a single universe, but in xenoverse.
## collection of xenoverse environments
- [AnyMDP](xenoverse/anymdp): Procedurally generated unlimited general-purpose Markov Decision Processes (MDP) in discrete spaces.
- [AnyHVAC](xenoverse/anyhvac): Procedurally generated random room and equipments for Heating, Ventilation, and Air Conditioning (HVAC) control
- [MetaLanguage](xenoverse/metalang): Pseudo-language generated from randomized neural networks, benchmarking in-context language learning (ICLL).
- [MazeWorld](xenoverse/mazeworld): Procedurally generated immersed 3D mazes with diverse maze structures.
- [MazeControl](xenoverse/metcontrol): Randomized environments for classic control and locomotions.
# Installation
```bash
pip install xenoverse
```
# Reference
Related works
```bibtex
@article{wang2024benchmarking,
title={Benchmarking General Purpose In-Context Learning},
author={Wang, Fan and Lin, Chuan and Cao, Yang and Kang, Yu},
journal={arXiv preprint arXiv:2405.17234},
year={2024}
}
@article{wang2025towards,
title={Towards Large-Scale In-Context Reinforcement Learning by Meta-Training in Randomized Worlds},
author={Wang, Fan and Shao, Pengtao and Zhang, Yiming and Yu, Bo and Liu, Shaoshan and Ding, Ning and Cao, Yang and Kang, Yu and Wang, Haifeng},
journal={arXiv preprint arXiv:2502.02869},
year={2025}
}
```