Back to blog

Chinese vs European cobots: what a simulation shows before you buy (2026)

Most cobot comparisons online are spec tables and price lists. This post does it the other way round: one task, four arms, one simulator, every assumption written down. Then a pre-purchase checklist.

Why the datasheet comparison misleads

Look at maximum speed on the official pages and the four arms seem far apart. The UR5e datasheet lists a maximum TCP speed of 4 m/s (TCP, the tool centre point, is the working tip of the arm). Dobot's CR5A page also says 4 m/s. The Fairino FR5 and the xArm 7 list 1 m/s.

A cobot that works next to people rarely runs at its top speed. How fast it may move is set by the risk assessment of the cell, not by the datasheet. Universal Robots' founder wrote: "A robot in itself is never safe", and every application needs its own assessment. Limits for contact with a person come from pain-threshold research per body region, first published in ISO/TS 15066; according to SICK's summary of the 2025 revision, that technical specification is now folded into ISO 10218, with the allowed forces and pressures as a central element.

You will also see 250 mm/s quoted as "the cobot speed limit". That number is the limit for manual or teach mode, when a person is guiding or jogging the robot. The Doosan manual states it directly: "In Manual Mode, the TCP movement speed is limited to less than 250 mm/s." It is not the speed a collaborative cell runs at in production.

For our example we capped motion at about 1 m/s in transfer and 0.5 m/s on approach. That is a modelling choice for a cell shared with people, not a number from a standard. Your integrator's risk assessment sets the real value.

The example: same task, four cobots

The scene was identical for every robot:

• Boxes arrive on a conveyor and go onto a 0.48 × 0.40 m pallet, 2 × 2 pattern, two layers: 8 boxes per pallet.
• Box 180 × 140 × 120 mm, 3 kg. Suction tool about 0.5 kg. Vacuum on and off take 0.15 s each.
• The same Cartesian path and the same smooth (minimum-jerk) speed profile for all four arms, peaking at about 1 m/s.
• For each robot we checked every move against that robot's joint speed limits from its datasheet. For the Dobot, where the datasheet is unclear for joints 3 to 6, we assumed 180°/s on all joints (a conservative choice). Had a joint gone over its limit, the move would have been slowed down. No move needed it.

The robot models came from two places. Dobot CR5 and Fairino FR5 are the manufacturers' own published URDF files, converted for MuJoCo. The Dobot file is the CR5 from Dobot's CRA-series package; its reach and speed below come from Dobot's CR5A page, which that same file corresponds to. xArm 7 and UR5e come from MuJoCo Menagerie, a model library curated by Google DeepMind. MuJoCo itself is open source under Apache 2.0. The simulation is kinematic: it checks motion and timing, not forces, and the grip is simulated (the box simply follows the tool).

What the table says

1. Same cycle, because the speed cap decides

All four arms ran a mean cycle of 5.05 s (4.41 s to 5.59 s depending on where the box goes on the pallet). A full pallet of 8 boxes took about 39 s on each. The busiest joint reached 73% of its speed limit on the Dobot, about 75% on the Fairino and xArm, and 84% on the UR5e. None of them was the bottleneck. The 4 m/s on two of the datasheets bought nothing here, because the task never asked for more than about 1 m/s.

That does not mean the arms are equal in every job. In a fenced cell where the robot may run at full speed, the arms with higher top speeds could pull ahead. We have not simulated that case, so we make no claim about it.

2. Payload margin matters more than it looks

The box weighs 3 kg and the suction tool about 0.5 kg, so the arm carries 3.5 kg. The xArm 7 is rated for exactly 3.5 kg. On paper it passes. In practice you have zero margin: a slightly heavier tool, a heavier product variant or a second suction cup puts you over. The three 5 kg arms carry the load at 70% of their rating.

A kinematic simulation does not model what a heavy load does to acceleration; a real arm near its rating will likely move somewhat slower. We list that under "check on the real machine".

3. Reach and joint limits show up only on your layout

Every arm reached the conveyor and the whole pallet without moving anything. The UR5e's farthest target was 644 mm from its shoulder, well inside its 850 mm reach. The xArm 7's 700 mm reach was enough for this small pallet but leaves the least room for a bigger one.

The simulation also caught something no spec table would. On the Fairino FR5, the obvious arm posture (elbow up, facing the work) is fine. The alternative posture, reaching "from behind", cannot reach the pallet at all, because the base joint hits its ±175° limit. On a real cell you find that during commissioning.

4. The real difference is price

With throughput equal, the arm price is what moves payback. The Fairino FR5 at a German dealer is listed at €6,544 including VAT (€5,499 net). UFactory's own store lists the xArm 7 at US$9,999 to US$10,194 depending on configuration, without VAT. Our Dobot CR5 figure (€14,000) is a net price from one German dealer listing. For the UR5e (€30,000) we took the low end of the US$29,500 to US$48,000 we found in US price guides, also without tax, which favours UR. So the Fairino figure includes VAT and the others do not; on its net price the Fairino payback would be about 7.6 months instead of 7.9. Ask your dealer for a real quote either way.

Every dated shop listing we found for Universal Robots, Dobot and Fairino arms, what each price includes and what a full cell costs is in cobot prices in 2026: Universal Robots vs Dobot vs Fairino.

Payback: the formula, not just the number

Payback pages online give ranges and hide their inputs. Here is ours in one line:

Payback (months) = (arm price + €25,000) ÷ (operators replaced × shifts × monthly employer cost)

With one operator, two shifts and €2,000 a month, the divisor is €4,000. Fairino: €31,500 ÷ €4,000 = 7.9 months. UR5e: €55,000 ÷ €4,000 = 13.8 months.

Two things change this number more than the brand does (and running one shift instead of two doubles it):

• How much labour the cell really replaces. If the operator still has to refill the conveyor for half the shift, the divisor halves and payback doubles.
• Integration cost. Our €25,000 is a placeholder. In MIT interviews with manufacturers and robot makers (Sanneman, Fourie and Shah), the authors report that, according to one large robot maker and integrator, robots are now inexpensive but integration is not, and integration costs 4 to 5 times the robot itself. That is one interviewee talking mostly about industrial robots, not a statistic, but it explains why a cheaper arm shortens payback less than its price tag suggests. In our example the arm is 21% of the cell with the Fairino and 55% with the UR5e.

For comparison, Qviro quotes Universal Robots as saying its cobots "can achieve ROI in just 195 days", about 6.4 months. That is a vendor claim. UR does not publish the inputs behind that figure; ours are written out above. Neither figure is your figure.

Where the Chinese and European arms differ outside the simulator

A simulation compares motion, not after-sales life. Ask about these directly, for any brand:

• Local support. Who in your country answers the phone when the controller shows an error? Are spare parts stocked locally?
• Certificates and documents. Ask for the declaration of incorporation, the safety certificates for the safety functions, and the EMC test reports, as files, before ordering.
• Software and programming. Which teach pendant or programming tool comes with it, what languages and protocols it supports, and whether your integrator has used it before.
• Model range. If you may need a bigger arm later, check that the same brand offers it. Dobot, for example, sells the same CRA series up to the CR10A, CR16A and CR20A (10, 16 and 20 kg).

If you buy directly from China, someone has to handle sourcing, the factory acceptance test on your parts and shipping. We do that as part of our China robotics consulting. The integration, risk assessment and CE marking of the cell stay with an integrator in your country.

Pre-purchase checklist

What to measure before you ask anyone for a quote

• The part: size, weight, and the heaviest variant you will ever run.
• The gripper or suction tool you expect to use, and its weight. Add it to the part weight.
• The station: a sketch with dimensions, heights, and where people stand and walk.
• The target: parts per hour, and how many minutes per shift the cell may be stopped for pallet changes or refills.
• The farthest point the robot must reach, measured from where its base can stand.

What to ask the seller

• The rated payload at full reach, and how speed drops near the rated payload.
• The joint speed limits and the TCP speed you can expect in a collaborative setting, not just the maximum.
• Whether a URDF or other kinematic model of the exact model and version is published. Without one, nobody can simulate it.
• What a factory acceptance test on your own parts would cost, and whether you can film it.
• The full price: arm, controller, teach pendant, cables, software options, shipping and the local warranty.

What a simulation can answer

• Does each candidate arm reach every point on your layout, in a posture its joints allow?
• What is the cycle time of the motion at the speed you agree on, and does it meet your parts per hour?
• Is the load within each arm's rated payload, and how much margin is left?
• With your cost inputs, how do the candidates compare on payback?
• With the physics option: does the suction cup or gripper hold the part, and what happens near the payload limit? (In our physics demo we had to assume joint torque limits, because Dobot publishes payload but not torques. Treat such results as indicative.)

What a simulation cannot answer

• Whether the cell is safe. That is the risk assessment, done by the integrator, and it ends with CE marking of the cell. A simulation is not a safety assessment.
• The exact cycle on the real controller, which has its own acceleration and blending behaviour. We do not generate controller programs.
• Your PLC, conveyor logic, part presentation or operator interruptions, unless we model them separately.

Free vendor tools vs a neutral comparison

You can simulate some of this yourself. Universal Robots offers UR Studio, a browser tool "provided without charge" for building a 1:1 model of a cell with UR robots and checking reach and cycle time. RoboDK Professional costs €3,995 for a permanent licence (€1,500 a year for maintenance after the first year) and is the right tool if you want to write the actual robot program offline for many brands.

A vendor tool shows one brand, and a programming tool assumes you have already chosen the robot. A buyer usually needs the same task on a Chinese and a European arm, side by side, with the assumptions written next to the numbers, from someone paid by you, not by any robot maker.

The same MIT study found manufacturers using simulation for "informing which robotic or other tools to purchase", and one of them said it was "too risky to automate in many cases without first simulating". If you want the technical detail of how we build the cell, read how we simulate a robot cell before you buy.

FAQ

More detail on the service is on the robot simulation page. Specs and prices for these and other arms are in our robot registry.