The Week in Robotics

We looked hard for Chinese rounds this week and found plenty of aggregate activity but no single deal we could pin to an exact company, amount and date inside the window. Rather than print a number we can't stand behind, we're telling you it's a gap. We also dropped a $22M Singapore round because no primary source for it exists publicly.

One widely-shared story didn't make it in: a "Unitree H1 Pro European launch." Two of our researchers went looking independently and neither could find a Unitree announcement or a single tier-1 outlet behind it — one aggregator citing another. If it's real, we'll run it when there's a source.

Three named customers put humanoids to work this week — and not one of them signed a production order. That distinction is the whole story, so we're leading with it rather than burying it.

Illustrative: UBTECH Walker S1 · the pilot itself runs AeiROBOT's ALICE M1

2 unitsnamed customerproof-of-concept

A Korean circuit-board plant puts two humanoids on the floor

AeiROBOT placed two ALICE M1 humanoids into Daeduck Electronics, a South Korean PCB manufacturer, to pick up, move and organize boards. The demonstration phase runs to year-end with results expected around six months out; AeiROBOT says it aims to place roughly 20 robots across pilot sites this year and start commercial shipments in 2027.

Robolist context Only 342 of the 3,056 robots we rank — about 11% — have a single confirmed customer on file. A named two-unit pilot sounds small, but on the evidence base that actually exists in this industry it is a genuine data point, and it is exactly how the record starts.

Source: The Elec · Jul 28

3 named makersChina

Robots reach the sewing floor at three Chinese garment makers

Aitu ran production trials at Sunrise Group, Eifini and Chenfeng Group, with robots loading fabric onto template sewing machines and feeding pocket setters and fusing presses — one of the hardest manipulation problems in manufacturing, because cloth doesn't hold its shape.

Whose numbers these are A 97% fabric-separation rate, 98% of output meeting quality standard and an 18-month payback all come from the trial operators themselves. No unit count or trial length was given. Treat them as claims until somebody independent measures them.

Source: Xinhua · Jul 24

$100Mzero customers named

Teradyne books $100M of robot revenue and names nobody

The parent of Universal Robots and MiR reported $100 million of Robotics segment revenue for Q2, inside a record quarter for the company as a whole ($1,329M total). For the robots specifically: no customer names, no unit counts, no segment backlog.

Why it's in this section anyway It is the weakest kind of deployment evidence — audited money with nothing attached to it. We include it because it's the one hard, filed number about robot demand published this week, and because the silence around it is the point: the biggest cobot vendor on earth tells you what it earned, never who bought.

A number we removed Secondary coverage put the segment up 33% year on year. Teradyne's own filing doesn't state that, so we cut it rather than pass it along.

Source: Teradyne Q2 2026 results (SEC 8-K) · Jul 28

We also chased a widely-repeated claim that Tesla had 500 Optimus units working at Fremont. Tesla's own Q2 update says Optimus production there has not started. The claim is false and we're naming it, because it will keep circulating.

ROBOTERA L7 — down 48 places. We did that, not the market.

We re-cut our own scoring math this week. So there is no movers table.

Normally this section shows the robots whose Robo Index score jumped on real signal. This week it would be dishonest to run it. On July 29 and July 30 we shipped two changes to the scoring formula itself. Comparing our last run on the old math to the first run on the new one, across the same 3,729 robots: the average score fell 4.2 points (56.9 → 52.7), 1,485 robots moved by 5 points or more, and 3,425 of them — 92% — scored lower than they had three days earlier.

Not one of those moves was caused by anything a robot maker did. Every single one was caused by us. Publishing that as "this week's biggest risers and fallers" is exactly the trick we built this newsletter to avoid.

Why we changed it Too many robots were scoring near-identically at the top, so trivia decided medals rather than evidence. The new math leans on what can be proven — confirmed customers, documented specs, real commercial status — and it is deliberately harsher. The catalog got a worse report card because we made the exam harder.

What actually changed on the board Our top-ranked humanoid is now Agibot's 远征A3 at 69.7, up from #29 under the old formula. AGIBOT G2, the previous #1, is now #4. ROBOTERA's L7 fell from #8 to #56 and Robbyant R1 from #2 to #28. Read those as a change of ruler, not a change of robot.

Movers return next week, measured like-for-like on the new formula — so the first numbers you see will mean what they say. Most trackers can't separate a genuine move from their own housekeeping. We can, and when it's housekeeping we'll tell you.

And one number that went the other way: the robot catalog got smaller this week. We added 1 robot and retired 13 — duplicates, merged entries, and five records that turned out not to be robots at all. We'd rather shrink honestly than grow on padding.