terrAIn

Robots That Move Dirt

Take a 4×8 ft bed of soil. Add a fleet of RC construction vehicles — an excavator, a dump truck, a bulldozer, a loader. Wire them with ESP32 microcontrollers and connect them to a Jetson running ROS2. Now teach them to work together.

That's terrAIn. It's a robotics platform where the terrain is real, the physics is real, and the machines actually dig. An overhead camera watches the bed, maps the surface, and feeds the AI planner. The robots execute. Sometimes they get stuck. Sometimes the soil doesn't cooperate. That's the point.

The whole thing fits in a greenhouse, a shed, or a backyard. Same footprint as a raised garden bed — because it is one.

Why Dirt Matters

Most AI today is disembodied. It reasons in text but has never pushed against anything. Intelligence didn't evolve in a chat window — it evolved through friction with the physical world.

terrAIn is our answer to that gap. When a student writes code and a machine digs a trench that collapses, they learn something no simulator can teach: the world pushes back. Soil has weight. Friction is real. Plans fail and you adapt.

For kids, that's where the learning lives — in the gap between what they planned and what actually happened. For researchers, it's a cheap, repeatable testbed for multi-agent coordination on deformable terrain. For everyone else, it's genuinely fun to watch tiny excavators move real earth.

What's in the Stack

The terrain bed: a robotics sandbox at garden-bed scale.

• The bed — 4×8 ft, filled with soil, sand, or gravel. Optional irrigation for agricultural experiments.
• The fleet — Huina 1:14 scale vehicles (excavator, dump truck, bulldozer, loader), each retrofitted with an ESP32-S3 and micro-ROS.
• The eye — overhead gantry with RGB and depth cameras for terrain mapping and vehicle tracking.
• The brain — Jetson Orin Nano running ROS2, handling perception, planning, and multi-robot coordination.
• The sim — NVIDIA Isaac Sim for training policies before they touch real dirt. Sim-to-real transfer.

Who It's For

terrAIn started as a research project and turned into something families, schools, and maker spaces actually want to use.

• Homeschool families — flexible schedule, multi-age, one kit serves the whole household. Real engineering, not worksheets.
• STEM programs and co-ops — project-based units in robotics, coding, and physics. Curriculum-aligned and open source.
• Makers and hobbyists — serious hardware without needing a lab. ESP32, ROS2, Jetson — the full stack.
• Researchers — a low-cost, repeatable testbed for embodied AI, multi-agent coordination, and terrain manipulation.

The Fleet

Every vehicle starts as a stock Huina RC model — solid metal construction, good mechanics, affordable. We retrofit each one with an ESP32-S3 for wireless control and micro-ROS, so they talk to the Jetson base station as autonomous agents.

Excavator

The workhorse. Huina 1550 or 1580 — tracked chassis, full arm articulation, bucket. Digs trenches, loads trucks, shapes terrain. This is probably the first vehicle you buy.

$55 – $85

Dump Truck

Pairs with the excavator. Hauls soil and gravel from point A to point B, articulated dump bed for unloading. The coordination between these two is where multi-agent planning gets interesting.

$55 – $75

Bulldozer

Pushes material and levels surfaces. The blade makes it the best tool for grading — flattening terrain to a target elevation. Tracked, so it handles loose soil well.

$60 – $80

Loader

Front-loading bucket on a wheeled chassis. Good for scooping and placing material. Faster than the excavator for bulk moves, less precise for detail work.

$55 – $75

Jetson Base Station

NVIDIA Jetson Orin Nano running ROS2. Handles perception from the overhead cameras, builds terrain maps, coordinates the fleet, and runs the AI planner. Optional overhead gantry kit included in the Research tier.

$499 – $599

Controller: Atlas DS

terrAIn uses the Atlas DS as its controller — a dual-screen handheld that gives you both the bird's eye and the driver's seat at the same time.

Bottom screen shows the overhead terrain map. Top screen shows the camera feed from whichever vehicle you're controlling. You can plan on the map view, then switch to manual driving, or just watch the autonomy run. One device handles teleop, monitoring, and field configuration for the entire fleet.

Students see both perspectives — the strategic overhead view and the robot's-eye view — which makes the gap between planning and execution visceral instead of abstract.

Pick Your Scale

Two tiers. Start with Home if you're a family or a small group. Go Research if you want the full fleet, simulation, and multi-robot projects.

Support the Build

terrAIn is open source. Everything — firmware, ROS2 packages, curriculum docs, Isaac Sim scenes — lives on GitHub. We're building this in the open because the best robotics education shouldn't be locked behind a vendor.

Your support funds hardware kits for schools and homeschool co-ops that can't afford the upfront cost, plus documentation and certification paths so students can show what they've built.

OpenCollective GitHub Sponsors STEM kits