One stack, four layers
Most robotics companies are one of these layers presenting itself as all four. A hardware vendor calls its fleet dashboard a brain. A model lab calls a policy a product. We separate them because the boundaries between them are where the engineering decisions actually get made.
- CrewOne governed memory across the siteHuman gate
tasks, traffic, shared memory
RobotBrainPerception, policy, local plannerBodyActuators, hands, fingertip tactile sensingRobotBrainPerception, policy, local plannerBodyActuators, hands, fingertip tactile sensingcorrections become training data
DataCorrections, capture, evaluation↑ Human gate → Crew
- On the robot
- Site server
- Human gate
Body
The machine that has to survive a hospital corridor or a wharf. We buy anything two suppliers can make and the market is making cheaper every year, and we build what fails in the field and nobody is fixing: hands, fingertip tactile sensing, the thermal and ingress engineering of bought joints, firmware and communications. Read the Body page.
Brain
One agentic brain with two modes. Solo is full local execution on one robot with no network dependency. Crew is every robot on the site network sharing one governed memory. Solo is a first-class state, not a fallback, and the control loop, the safety functions and the manipulation policy never leave the robot in either mode. One qualification, added on 11 September 2026: local autonomy on its own is no longer what separates us. Diligent's Moxi 2.0 shipped on 17 August 2026 with safety and autonomy moved onboard, so a hospital task completes there without Wi-Fi too. What is still ours is the shared half — Moxi is single-vendor and shares memory with nothing, and Solo carries a read replica of the site's core memory and a published re-sync contract. Read the Brain page.
Data
Teleoperation is how robot data gets made, and the industry does not price it. We put it in the pilot budget, capture every correction from the first day of a deployment, and run a fixed evaluation harness before any training release reaches a fleet. Results, including failures, are published. Read the Data page.
Crew
Task allocation and traffic management are shipped and mature across the forty-eight profiles we surveyed, which name fifty-three companies between them. Site-level shared memory is not. Crew is the layer we are building into that gap, and it coordinates other vendors' robots as well as our own. Read the Crew page.
Where the layers meet
Two boundaries do most of the work in this section. The first is the line between what runs on the robot and what runs on the site server: anything that has to be fast or safe stays on the machine, and the site server only sees task boundaries. The second is the human gate: robots propose changes to the site's memory, and a named person approves them before any of it reaches another robot. Every page below returns to those two.