A robot or device acts on a command only if it is one someone approved.
Robot fleets and connected devices usually check who is connected and which channel they may use. They do not check what the command says, or whether a person approved it. A stolen credential, a compromised fleet manager or a planning agent acting on a poisoned instruction can send a well‑formed command that nobody approved.
The gap
- The channel is secured, the command is not checked. Permissions typically say who may publish to which topic or queue, not what the message contains.
- An order is trusted because of where it came from. Nothing ties a robot order, or a device command, to the person who approved it.
- Planning agents widen the exposure. A robot driven by an automated planner can be steered by a manipulated instruction, and the human origin of the order is lost along the way.
The design: command authorization at the edge
- An accountable person, or an agent with delegated authority, approves one order or command: this robot or device, this action, this time window.
- The approval becomes a short‑lived token that traces back to that person and can only narrow as it is delegated, for example from a fleet manager to a single vehicle.
- A small gate next to the robot or device, on its controller or an edge gateway, compares the real order or command with the token before it is acted on. A different order, a different target or an expired approval is refused, and every decision leaves a signed record.
- Verification is offline, against a locally held signed trust bundle. Nothing calls home.
Streamed or continuous control, such as live teleoperation, is a harder case and is not covered by this first design.
Safety comes first
The gate is never a safety function and never sits in a safety loop. Emergency stops, protective stops, bumpers, scanners and every other protection stay exactly as they are and never depend on it. If an approval expires while a robot is moving, the gate stops accepting new orders and records it; what the robot does next is decided by the operator’s own safety assessment, not by SPT‑Txn.
What it would not stop
Physical tampering, sensor spoofing and a compromised robot controller are outside what a command check can fix. Neither does it judge whether an approved order was wise; it records who approved it.
If this is useful to you
We are looking for robot and device builders, fleet operators and researchers to review the design and help build a first simulated demonstration.
The same check for plants and machines →
The same check for drones (design) →
How the token chain works →
rudi@violetskysecurity.com