FIELD GUIDE · ROBOTICS EVIDENCE

Demo, pilot or deployment?

A video can show that a robot moved. It cannot, by itself, tell you how often it works, who helped, or whether the same result survives outside the lab.

Use this checklist whenever a robotics announcement sounds bigger than the evidence. A controlled demonstration can be valuable; the point is to name what has actually been shown.

01 · START WITH THE CLAIM

What job, in what environment?

Ask for the task definition before asking whether a robot is “general”. Picking totes in one warehouse, folding laundry in a lab and navigating an unknown building are different jobs. Record the object, workspace, lighting, floor, safety boundaries and allowed resets.

02 · FIND THE HUMAN

Autonomy, teleoperation or assistance?

“AI-powered” does not mean “unattended”. Look for a human operator, remote intervention, demonstrations supplied during the run, safety staff, manual resets and hidden post-processing. A system may be useful while still remaining supervised.

03 · SEPARATE THE STAGES

Demo, pilot and deployment are not synonyms

Demo: a selected example showing that a capability can work under stated conditions.

Controlled lab result: a measured experiment with a defined setup, baseline and limitations. It may advance the science without proving a product.

Pilot: a limited trial in an operational setting. Ask whether the customer, duration, intervention rate and success criteria are public.

Deployment: repeated work in a real operation. Ask for the job, site, uptime, throughput, failure handling and boundary of the claim.

In our first Robotics Brief, GXO reported more than 100,000 totes moved by Digit at an existing Flowery Branch site. The same release described Statesville and Lebanon as planned future sites. That is a live-site milestone plus two planned expansions—not evidence that the future sites are already operating.

Read the GXO and Agility primary release → · See how Brief 001 labels it →

04 · ASK FOR THE NUMBERS

Throughput needs a baseline

A count is not a rate. For a performance claim, look for attempts, success definition, cycle time, downtime, human interventions, failure recovery and comparison with a person or previous process. If details are missing, keep the claim narrow: “the company reports X in this setting.”

05 · READ THE EVIDENCE HIERARCHY

Match the source to the claim

A company announcement is the primary source for what the company says it installed or plans to do. A paper is the primary source for a controlled experiment and its methods. Neither automatically supplies the other’s missing evidence.

The moth-inspired robot paper in Brief 001 describes a controlled materials proof of concept: human priming, UV-triggered chemistry and a local wing-flapping loop. It is interesting hardware research, not evidence of a robot freely evolving abilities in the field.

Read the primary paper → · Read the full evidence-labeled brief →

THE ONE-LINE TEST

Can you state the boundary?

Try: “The source shows this task, at this site or setup, for this duration, with this human involvement. It does not yet establish that broader claim.” If the sentence cannot be written without assumptions, the claim needs more evidence.

Want the next evidence-led read? Subscribe to the NXTDLY Robotics Brief →