The FCC Didn't Ban Cameras. It Made the Sensor Choice a Data Question.

On 28 July 2026 the FCC added foreign-produced "advanced robotic devices" to its Covered List. The practical consequence is simple: new models from those manufacturers can no longer obtain the equipment authorisation needed to be imported and sold in the United States. The definition is narrow but unforgiving — a device that moves over ground, operates away from an operator on instructions or sensor data, weighs more than 2 kg including its dock, carries environmental sensors and a two-way connection of at least 200 kbps, and runs software controlling navigation, perception or data collection. Robotic mowers and floor-care robots sit inside it. Underwater pool robots were explicitly excluded.

There is an exit. A manufacturer can ask the Department of War for review, and a device that receives Conditional Approval is exempted from the list. As of October, only three robotics companies had cleared that route.

On 9 September a European mower manufacturer became the first to receive Conditional Approval, covering four platforms scheduled for the US market in 2027.

On 2 October a California floor-care company's robot and charging dock were exempted under Public Notice DA 26-1063 — a machine built around five cameras and on-device AI, with no cloud dependency at all.

Put the two side by side and the popular reading falls apart. If cameras were the problem, the second approval could not exist. If radar were the answer, the first one would be reproducible by anyone willing to swap a sensor.

Sensor choice changes how difficult the review is. It does not decide the outcome.

Non-imaging. Satellite RTK for position, radar or UWB for obstacle detection, no image capture at all. The cleanest review posture and the weakest semantics: excellent at "where am I, where is the boundary", mediocre at "what is that". For mowing, where the boundary is the hard problem, that trade is often acceptable.

Vision with edge-only processing. Still the best semantics per dollar — plants, pets, toys, boundaries. It survives review only if four things hold at once: inference runs on the device, captured imagery is never used to train models in the cloud, telemetry is opt-in, and the robot functions with no internet connection. Break any one of them and the advantage becomes the liability.

Fusion. LiDAR plus vision plus RTK remains the performance benchmark for Europe and Asia-Pacific. One caution for anyone treating it as a compliance shortcut: a LiDAR point cloud is geospatial data too. "No camera" is not the same as "no sensitive data".

Two sensor configurations is not a marketing decision, it is a platform decision. Which platform carries the non-imaging stack, which keeps vision, and whether one front-end can accommodate both without new tooling — those questions are cheap in the concept phase and expensive after the industrial design is frozen.

And the uncomfortable part: a camera-free mower built offshore, with no approved path, still cannot be authorised. Sensor architecture reduces friction at the design layer; origin and onshoring are decided at the corporate layer, and no amount of engineering substitutes for them.

Budget accordingly. The two workstreams are separate.

We develop robotic mower platforms on an ODM basis across vision, RTK and LiDAR architectures, and we plan sensor variants at platform level rather than retrofitting them into a finished product. For US programmes that means parallel configuration work, edge-compute integration, and coordination of FCC and EMC testing with your certification body.

We do not sell approvals, and we do not represent that any configuration is pre-cleared. What we can do is make sure the hardware is ready for whichever route your regulatory counsel recommends.