Inspiration

Numerous injuries happen to toddlers and elderly people each year (5M+ accidents, some). I wanted to build a tool which would allow identifying the hazards to mitigate them, eg sharp corners etc.

This should help for places where vulnerable people come.

What it does

The user uploads photos, then a world model (we use Marble) turns them into a 3D space with a collision mesh. Then in simulation "todlers" and "grannys" walk in the environment identifying the hazards. The user is notified with what is dangerous..

How we built it

Using Devin, LLMs, and world models. The site is hosted on cloudflare (via chatgpt sites plugin)

Challenges we ran into

Mesh takes a long time to generate, so we used sample meshes initially. Simulation works, but more precision is needed.

Accomplishments that we're proud of

We identified real world application of world models where they can make a difference beyond games. We built the whole system end to end.

What we learned

World models are extremely powerful, well beyond image or video generation.

What's next for HazardDetection

I am planning to publicly release.

Built With

• devin
• reactor
• typescript
• world-model