Releases, events and notes from projects we are working on.
A quadruped in a compressor house or tank farm is not there to patrol. It carries a gas detector, a thermal camera or an acoustic sensor to a spot too dangerous to send a person, and relays what it finds. The IP rating on the chassis is not the certification that matters here, and the catalogue is honest about that gap.
Read more
Hour by hour: where a robot earns its booking, where it should stand still, and the four things it does not do. Written for one evening with a crowd, not for a permanent reception desk.
Read more
Two European dates are easy to merge into one deadline, and only one of them moved. Regulation (EU) 2023/1230 still applies from 20 January 2027. The AI Act rules for AI embedded in regulated products were extended to 2 August 2028.
Read more
How heavy, how tall, through which door and up which lift. The machines rented for events run from 16 kg to 185 kg, and almost everything that sinks a booking in its last week is answered by a tape measure.
Read more
The EU-27 runs 231 industrial robots per 10,000 manufacturing employees against a global average of 132. The figure has two moving parts and only one of them is robots, which is why it answers a different question from the one it gets asked.
Read more
The A3 won its category in the IFA Innovation Awards and the X2 was named an honoree in the same one. Which award that is, how it differs from the picks collected at CES, and what neither kind tells you about your own venue.
Read more
The G2 publishes 2 LiDAR units and 3 RGB-D cameras as counted rows; the X2 Ultra names the same sensor types with no count; the X2 Basic states zero. Reading the difference is worth more than the word "autonomous" on its own.
Read more
AimDK_X2, in Python and C++, is listed identically for the X2 Basic and the X2 Ultra. AGIBOT's own comparison table draws the real line elsewhere: customer-side development, not supported on the Basic and supported on the Ultra.
Read more
Two 96-line units against one sounds like a clear win for the D1 Max. What each configuration covers, and what the catalogue does not publish, is a different question.
Read more
The D1 Ultra-W weighs 4 kg more than the standard D1 Ultra, 20 kg against 16 kg with battery fitted. That weight does not go into the frame for its own sake: effective payload doubles, from 5 kg to 10 kg. Here is what that trade actually buys a buyer, and what the catalogue still will not tell you.
Read more
A rated 50 kg, and no higher figure published. That number puts the D1 MaxPro in a different job than the rest of the quadruped line, and it changes what the machine should be asked to do.
Read more
The X2 Ultra publishes numbers a fixed-base cell never needed: speed, runtime, per-arm payload. Read against each other, they tell a buyer what mobility actually costs.
Read more
D1 Ultra, D1 Max and D1 MaxPro publish three different continuous step heights. Here is what that number actually predicts about a route, and what it leaves for you to check yourself.
Read more
A purchase order runs 12 to 16 weeks of lead time. A rental answers your question inside 48 hours. What you can rent, at what weekly rate, and what still has to be taught on site.
Read more
A humanoid bought for a university robotics lab has to be programmed, not just watched. What the X2's open SDK, degrees of freedom and tier split actually decide for a lab.
Read more
A mall, hospital or office tower is not one floor repeated. Lifts, secured doors and cleared corridors are a coordination task with facilities before they are a spec line, and that is the part a demo never shows.
Read more
The C5 runs 3 hours scrubbing or 8 hours dry-mopping on the same battery, and that split, not a single headline number, is what decides whether one machine covers a shift.
Read more
The C5 docks, empties its sewage tank and refills clean water on its own. It does not restock detergent or empty a mop head. Here is where the line sits and what it means for staffing an unattended shift.
Read more
The C5's headline is 1,920 m²/h. Before that number tells you anything, measure it against your own floor plan, your path width and your obstacle density.
Read more
D1 EDU and D1 MaxPro compared on the two figures that actually decide a round, and why top speed is the one to read past.
Read more
The X2 Basic and X2 Ultra differ by five joints on paper. What actually separates them is what each one can see.
Read more
The A2 Ultra, A3, X2 Basic and D1 Ultra rank one way by height and another way entirely by torque.
Read more
The A2 Ultra runs about 2 hours before a swap; the A3 runs up to 10. Which one fits a shift depends on the layout, not the spec sheet alone.
Read more
The A2 Ultra's teleoperation kit runs at 50 ms wired latency and a 105-degree field of view, and every session it records is training data, not just a demo.
Read more
Swift Picker adjusts from 10 to 95 N; Titan Picker from 40 to 140 N. The right choice is rarely about force alone.
Read more
The D1 Ultra's own numbers explain what recovering from a stumble actually requires, and why it isn't a heavier platform's problem in miniature.
Read more
A 558 mm arm reach and 120 Nm of joint torque decide a pick-and-place cell layout more precisely than a general spec comparison ever will.
Read more
Swift Picker is rated IP40 and cycles in 0.25 seconds; Titan Picker is rated IP67 and takes 0.75. The right choice depends on the cell, not the catalogue page.
Read more
The three ratings across the catalogue answer three different weather questions, not one sliding scale of more outdoor.
Read more
Permissive, non-commercial, or no licence file at all: three different answers across one manufacturer's own repositories, and the difference decides what you can ship.
Read more
Buy from a distributor already inside the single market and importer obligations never reach you. Import directly and Regulation (EU) 2019/1020 puts them on you specifically.
Read more
Cycle time end to end and a success rate with failure modes separated out are the two figures that turn a demo into a specification, and the two easiest to skip.
Read more
A short PoC answers whether a task is possible. It does not answer how the machine ages, and MTBF is not on the list of things it settles.
Read more
Starting risk assessment after the cell is built is the most common reason a working deployment cannot be switched on. The A2 Ultra's own 169 cm, 69 kg envelope shows why the assessment has to be site-specific from day one.
Read more
AGIBOT publishes a hot-swap claim for three humanoids and one quadruped, and a different picture for the rest of its quadrupeds. Which pack, how many, and whether the swap is stated as tool-free decides who can actually do it on the floor.
Read more
Every AGIBOT supply comes with commissioning, an operator briefing and a year of EU service handling. Where that ends and billed engineering support starts is a term worth reading before signing, not after a machine stops.
Read more
"Autonomous" describes software sitting on a specific, checkable parts list, and that list is shorter than the word implies on more than one AGIBOT platform. Read it by SKU, not by product line.
Read more
A published obstacle figure is a single number; a route is several of them together. AGIBOT's D1 Max publishes obstacle height, crossing height under payload, crawling height and stair height as four separate figures, and no minimum width at all, so a route needs all four plus a tape measure.
Read more
The G2 trades legs for a wheeled base and gets a height range no bipedal platform in this catalogue matches, 1225 to 1795 mm, on a machine running two separate compute chips for two different jobs.
Read more
AGIBOT's humanoids each publish a mains voltage range and a charger current draw; its D1-series quadrupeds publish a charge time and little else. Both are decisions a site survey should settle before power is run to a layout, not after.
Read more
Store list prices for AGIBOT humanoids, what they leave out for an EU buyer, our weekly rental rates from 1,500 to 7,500 EUR, and what a delivered quote includes.
Read more
The choice isn't about which platform looks more capable. It comes down to the floor, and to which kind of payload figure is actually being compared, because a per-arm number and a total-payload number aren't the same question.
Read more
Not a single job. This catalogue alone has two tiers with different sealing, payload and terrain figures, so "what can a robot dog do" has a different answer depending which one is actually being asked about.
Read more
The honest answer sits in payload, joint and sensor figures already published for this catalogue, not in what a demo video implies, and a two-week trial is how to find the edge of it rather than guessing from a spec sheet.
Read more
Not every AGIBOT platform ships with the same radios, and one of them does not publish a wireless spec at all despite supporting OTA updates and a mobile app.
Read more
Operating temperature range is a published figure that decides this before payload or sensors do, and the platforms in this catalogue split into genuinely different tiers.
Read more
A single day is enough to settle whether a task is worth pursuing, not enough to prove it works in production, and knowing the difference is what makes the day worth booking.
Read more
A short trial can prove a robot tells a good part from a bad one against a reference drawing. It cannot prove that holds up over months of production, and treating the two as the same claim is where this kind of pilot goes wrong.
Read more
5,168 humanoid robots shipped in 2025, per the analyst who tracks the category, is a different question from whether a specific supplier answers a support ticket. Both matter, and they are not the same question.
Read more
A robot walking a route and flagging what looks unsafe sounds like a safety system. It isn't one, and that distinction is the most important thing about how this offering actually works.
Read more
It sounds like fleet management. It's a browser interface served from one robot's own onboard computer, one machine at a time, and that distinction decides whether it answers the question that brought you here.
Read more
Bulgaria and Romania, and Greece and Cyprus, are two different legal answers to the same question, and importing directly from outside the Union is a third, more expensive one.
Read more
No advertising, no profiling, no automated decision-making, and no third-party embeds, stated as specifically as the platforms this catalogue actually sells. A privacy policy is worth reading with the same attention as a spec sheet.
Read more
From 1,500 EUR a week for an X2 Basic to 7,500 EUR for a G2: which robot suits which event, how long it runs, and what the venue and crowd need.
Read more
Mapping, perception and who owns what it says: the four items that decide whether a reception robot works, and the one that gets bought first by mistake.
Read more
Range and IP rating decide the round before speed does, and exactly where a wheeled guard robot would already have stopped.
Read more
A dancing robot at an event runs on preset movement, not choreography written for your show. Here is what AGIBOT actually publishes, and what still has to be engineered around it.
Read more
The platforms we represent were recognised from four different sources at CES 2026. Here is which ones they were, and why the machines running all day on the stand mattered more to us than the trophies.
Read more
A humanoid platform on show at the World Economic Forum, and what the reaction told us.
Read more
Quadruped platforms are finding a serious role in visitor-facing entertainment environments.
Read more
The AGIBOT A2 set a Guinness World Record for the longest journey walked by a humanoid robot: 106.286 km over four days.
Read more
Our engineering team spent a week in Shanghai working on customisation options for European customers.
Read more
