Physical AI has mostly learned indoors.
Today's robot foundation models are trained largely on kitchens, warehouses, and carefully controlled factory benches.
The outdoors is a different world. Muddy ground shifts under a robot's feet. Crops change week to week. Dust, glare, and rain blind cameras. Animals, people, and machines move. The machines nearby weigh tons.
A robot that has never seen those conditions will not be optimized for work in them, and the data that would teach it has never been collected at scale. That is the corpus we are building.
- Terrain
- Slopes, ruts, soft and saturated soil, gravel, stubble, rubble
- Materials
- Granular media, mud, deformable plants, fruit, livestock, rebar and aggregate
- Conditions
- Weather, low sun and night, dust, seasonal and regional variation, noise and interference
- Machines
- Tractors, on and off-road vehicles, implements, loaders, telehandlers, sprayers, and their control buses
- Scale
- Work measured in lives, acres, rows, bushels, pallets, yield, bags, gallons, and shifts, not tabletop reaches or controlled environments
w