What do you want to build today?
Each machine here is taught part by part: what every part does, what every step asks of it, and why each one is hard. Then it follows the machine out to a fleet.
Pick the work above, or roll for a machine. Each one is drawn part by part, then followed out to its fleet.
Follow one piece of work.
Each page takes one piece of work through the parts of one machine, then out to its fleet: the data coming back, the next version trained or written, and that version going out again. The fleet steps name the Device to Cloud Flywheel solution that does the work, and the pages are grouped here by that solution.
The fleet drawings show 1,000 machines to show the problem at that size; Device to Cloud Flywheel has been run on fleets of single digits to low tens of devices. Each solution's name and “you have” text come from its solution page, with punctuation or wording adjusted in places.
Laptops, desktops, and Raspberry Pis
You have: A handful of ordinary computers you look after, a home server, a few Pis, the machines under someone's desk. Today you keep a VPN up so you can SSH in and fix things by hand.
- Follow one reading
Connected sensor node
A failing bearing shows as short, fast knocks, and the node has to catch them from inside a sealed box on one battery that lasts for years.
A ROS 2 robot
You have: Robots already running ROS 2. You have a stack you are not going to replace, and you want fleet operations around it rather than instead of it.
- Follow one weed
Laser weeder
A young weed can look almost the same as the crop, and the machine has to find it and hit a growing point a few millimetres across while it keeps moving.
- Follow one tote
Warehouse AMR
It has to know where it is in aisles that all look alike, and it has to stop for a person whatever its navigation software is doing.
A PX4 drone fleet
You have: Drones running PX4. Flights produce logs you want off the aircraft, and firmware and parameters you need to roll out without bricking a fleet.
- Follow one flight
Inspection drone
Beside a live line, the flight controller has to correct for every gust, several hundred times a second and on time. It still has to bring the aircraft home with charge to spare.
Cameras and vision devices
You have: Cameras in the field, fixed, on a machine, or a browser on a laptop. You want to see them live and pull frames back for training.
- Follow one weedLaser weederfollow one weed, shown above
- Follow one frame
Smart camera
It can't send every frame, so the camera itself has to decide which few seconds are worth keeping. The light changes all day, and the site may be one its model has never seen.
A robot arm running a learned policy
You have: An arm, a LeRobot SO-101 or similar, and a policy you retrain as you collect more demonstrations.
- Follow one shirt
Laundry folding robot
Cloth has no fixed shape, so every grasp and every fold has to be chosen from what the cameras see of this one garment.
PLC and Modbus machines
You have: Industrial machines speaking Modbus behind a small computer. The meaning of a register is knowledge that currently lives in someone's head.
- Follow one readingConnected sensor nodefollow one reading, shown above
- Follow one carton
PLC packaging line
The PLC runs the line and has to stay in charge of it, so everything read or changed from outside passes through a few numbered registers whose meaning is written down, if at all, in someone's notes.
Heavy equipment, simulated then real
You have: Machines too expensive to experiment on. You want to train against a simulator and deploy the result to the real thing through the same path.
- Follow one bucket
Autonomous excavator
The machine can't know what the soil is like until the bucket is in it. A person far away also has to be able to take over, and the machine must never move when the link is lost or late.
Photoreal simulation with Isaac
You have: An Isaac Sim scene and a need for training imagery that looks like what the machine will actually see.
- Follow one bucketAutonomous excavatorfollow one bucket, shown above
Training a policy and shipping it to the fleet
You have: Field data coming back from machines, and a model you want to improve and redeploy without a bespoke pipeline each time. It runs across the other solutions rather than describing one kind of device.
- Follow one weedLaser weederfollow one weed, shown above
- Follow one shirtLaundry folding robotfollow one shirt, shown above
- Follow one toteWarehouse AMRfollow one tote, shown above
- Follow one frameSmart camerafollow one frame, shown above