SimWorld releases SimWorld Studio with SimCoder, generating interactive 3D environments on Unreal Engine 5 that raise embodied navigation success rates from 50% to 90%
Includes GitHub repository, arXiv paper, and urban demo assets.
Scaling embodied AI starts with automating the environments.
Introducing SimWorld Studio: a self-evolving factory for endless interactive 3D environments where agents act, fail, and learn.
With coding-agent + embodied-agent co-evolution, navigation success improves from 50% → 90%.
1/
Why environment generation?
Digital agents scale through sandboxes: code, web, computer-use.
Embodied agents need the same but in 3D worlds.
Not just static scenes, but environments with physics, interaction, tasks, rewards, observations, and standard learning interfaces.
2/
Lianhui Qin@LianhuiqScaling embodied AI starts with automating the environments. Introducing SimWorld Studio: a self-evolving factory for endless interactive 3D environments where agents act, fail, and learn. With coding-agent + embodied-agent co-evolution, navigation success improves from 50% → 90%. 1/
How SimWorld Studio works:
Prompt → SimCoder writes UE5 code
Code → interactive 3D environment
Verifier → checks collisions, physics, prompt alignment, task validity
Fixes → become reusable skills
Environment → Gym-style task
Agent → enters and learns 3/
Lianhui Qin@LianhuiqWhy environment generation? Digital agents scale through sandboxes: code, web, computer-use. Embodied agents need the same but in 3D worlds. Not just static scenes, but environments with physics, interaction, tasks, rewards, observations, and standard learning interfaces. 2/
The key idea is co-evolution.
Agent succeeds → the next environment gets harder.
Agent fails → the next environment teaches what it missed.
The curriculum is no longer fixed.
Environments that don’t just get generated — Environments that grow with the agents inside them. 4/
Lianhui Qin@LianhuiqHow SimWorld Studio works: Prompt → SimCoder writes UE5 code Code → interactive 3D environment Verifier → checks collisions, physics, prompt alignment, task validity Fixes → become reusable skills Environment → Gym-style task Agent → enters and learns 3/
Github: https://github.com/SimWorld-AI/SimWorld-Studio Website: https://simworld.org/simworld-studio/ Arxiv paper: https://arxiv.org/abs/2605.09423 Huggingface paper: https://huggingface.co/papers/2605.09423 Youtube link: https://youtu.be/WPFZPWYqcFY
Lianhui Qin@LianhuiqThe key idea is co-evolution. Agent succeeds → the next environment gets harder. Agent fails → the next environment teaches what it missed. The curriculum is no longer fixed. Environments that don’t just get generated — Environments that grow with the agents inside them. 4/
#949Zhiting Hu@ZHITINGHU🔥🔥
SimWorld@simworld_aiEnvironment generation is the missing scaling axis for embodied AI. Introducing SimWorld Studio: a self-evolving factory for endless interactive 3D env where agents act, fail & learn. Env-agent co-evolvution improves navigation success 50% → 90%. From a prompt, our SimCoder writes code to automatically build an interactive world. Agents train inside it. And their performance shapes the next world.