# SimGate

- **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:** [Angelina Degay](https://cerebralvalley.ai/u/angelinadeg)
- **GitHub:** https://github.com/angelinadegay/meta-ai-simgate.git
- **Demo video:** https://www.youtube.com/watch?v=52OaklSf2fo&t=4s
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3310

SimGate is a plain-language gateway to physical simulation. You ask a question like “Can this robot climb out of a crater on Mars?” and SimGate builds the scene on real NASA terrain, runs real physics, checks its own results, and explains what happened and why. No robotics or simulation background is needed.

The idea came from my undergraduate research on tensegrity robots at Rutgers. The hardest part wasn’t the robot; it was building realistic worlds to test it in. Every terrain took hours, most tools assume flat ground, and almost nothing real is flat.

The problem SimGate solves: AI is confident about physics, and often wrong. In the physical world, a confident guess breaks things. SimGate puts a layer of evidence between AI and the physical world:

Real worlds: NASA and USGS elevation data from Mars, the Moon and Earth, generated terrain on six worlds with real gravity, and an indoor apartment with doorways, stairs and a ramp.
Robots and structures: rovers, walkers and a humanoid in MuJoCo, buildings in earthquakes in OpenSees, and races with up to six robots. Proof, not guesses: it stress tests a robot under 24 conditions at once, finds where it breaks with uncertainty attached, and proves why it failed by testing one change at a time.
Where can it go: it maps where a robot can safely travel from its measured limits, spot-checks the map with simulation, corrects it when the evidence disagrees, and plans a safe route.
Design a robot for a place: Meta’s Muse Wonder designs robots that must fit a lander, they race on real Martian terrain, each failure shapes the next generation, and a winner must succeed on at least 4 of 5 new runs before SimGate trusts it.

SimGate is simulator-agnostic: the same plain-language layer drives MuJoCo for robots and OpenSees for structures, and could sit on top of other simulators. It runs on a normal laptop with no GPU.

## More from Meta Global AI Developer Hackathon

- [Abubakar Jibrin](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3307)
- [refuzz](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3308)
- [Sinux](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3309)
- [Gonaj](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3311)
- [Syed's Enterprises](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3312)
- [RoboLN AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3313)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3310. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt