Spatial AI: Large datasets for robot foundation models
Collecting large-scale real-world data to teach robots how to navigate the world and complete tasks.
Spatial AI (https://spatial-ai.com/) is building large datasets for robot foundation models.
Open Source: https://huggingface.co/datasets/spatial-ai/sea-small
The Problem:
The biggest obstacle to bringing robots into everyday life is the software. We need far more capable models to control general-purpose robots effectively and enable them to carry out tasks in the physical world.
The Solution:
To train capable models, we need vast amounts of high-quality data. Our first dataset SEA (Spatial Everyday Activities) is the largest well-curated egocentric dataset of people performing tasks.
Story:
I’ve been building humanoids and dynamic robots for nearly a decade, starting in my garage when I was a kid.
https://alexandros-petkos.com/humanoid-work
https://alexandros-petkos.com/quadruped-work
Alex