# IRRS (Infinite   Realistic  Robotic Simulation)

- **Event:** [Hack FLUX: Beyond One](https://cerebralvalley.ai/e/bfl-hackathon)
- **When:** Nov 22 at 9:00 AM – Nov 23 at 5:00 PM (PST)
- **Where:** Shack15, San Francisco, CA
- **Placement:** 3rd Place
- **Team:** [Theo MICHEL](https://cerebralvalley.ai/u/theo_michel42), [David Dobas](https://cerebralvalley.ai/u/daviddobas)
- **GitHub:** https://github.com/theo-michel/sam-3d-objects https://github.com/DavidDobas/flux-robot-environments
- **Demo video:** https://youtu.be/nSmgWVGaywI
- **Gallery:** https://cerebralvalley.ai/e/bfl-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/bfl-hackathon/hackathon/gallery/24

Collecting high-quality data for robotics is hard.
You need accurate joint positions and high-resolution video or 3D data. Simulated data is usually too far from reality—physically, visually, and in terms of variation. Teleoperated data, on the other hand, is extremely expensive to collect.

We’re building a world simulator to bridge this gap.
Our goal is a prompt-to-3D-world system where a robot can act inside the generated environment, and we automatically produce high-quality frames of the key moments in its trajectory for training.

In our demo, we go from a prompt → Flux → SAM-3D.
The robot can then move in that 3D world and interact with objects. At the end, Flux generates high-quality data using screenshots from the 3D environment—plus the previously generated Flux images—to produce realistic, diverse scenes with accurate joint positions.

!THERE ARE TWO REPOS!

---

Markdown version of https://cerebralvalley.ai/e/bfl-hackathon/hackathon/gallery/24. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt