# Bravo

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Juwon Lee](https://cerebralvalley.ai/u/juwonlee)
- **GitHub:** https://github.com/NasDoctorFish/ESD-Guard-AI
- **Demo video:** https://youtu.be/1ycFpVw9au0
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1640

ESD-GUARD AI is a simulated lunar robot arm (Franka Panda in MuJoCo, lunar gravity) that recognises what an object is made of and changes how it handles the object to avoid electrostatic discharge. A camera image goes to Meta's Perception Encoder, which runs locally, and Muse Spark adds a second opinion. That gives a material estimate with an uncertainty flag. A transparent ESD risk model, built from cited NASA and peer-reviewed lunar charging data, turns the estimate plus the environment (sunlit plain, shadow, sun/shadow boundary, solar storm, or an Earth factory) into a handling policy: speed, grip, touching a ground rod first, slow release, and the grounding pad instead of the bin. Muse Spark plans the pick-and-place from the scene image, and a rule-based guardrail enforces the safety policy as a minimum.

A web app ties it together. You scan a real object with a phone, see its risk card, and send it to a digital twin, where the arm handles it on the Moon and on Earth side by side.

Results:
- Simulation benchmark (10 scanned objects × 5 environments): 100% policy compliance, against 0% for a robot that ignores material on the Moon.
- Held-out look-alike photos: 0/18 handled unsafely, against 5/18 with the Perception Encoder alone.
- Guardrail test: all 112 deliberately bad plans were corrected.

All ESD risk is simulated and labelled as such. The approach carries over to any physical AI that must handle unknown materials safely.

## More from Meta Global AI Developer Hackathon

- [Adarsha Chakma](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1637)
- [Jean Kpoti](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1638)
- [LifeSOS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1639)
- [TekTık](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1641)
- [Plainproof](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1642)
- [VeeCut](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1643)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1640. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt