Cobot vs industrial robot: what you give up, and what you get.

That's the whole difference, and it's smaller than the marketing suggests. Both are six-axis arms with servo motors and gearboxes, both are programmed from a pendant or a PC, and both fall under the same safety standard. The 2025 editions of ISO 10218-2 go further than the old ones did. They talk about collaborative applications and collaborative tasks, not collaborative robots, because the same arm can be perfectly safe placing a foam part and dangerous swinging a sheet-metal blank.

So when people ask about the difference between a cobot and a robot, the honest answer has two layers. At the arm level, a cobot trades speed, stiffness and some precision for built-in sensing and safety functions. At the cell level, what matters is whether the whole application, gripper and part and fixture and all, can meet contact limits or keep people far enough away. A cobot makes the first option possible. It doesn't make it automatic.

You'll also hear cobots described as "easy to program." That's mostly true and mostly unrelated to safety. Hand-guided teaching and graphical pendants came with the cobot wave, and several industrial arm makers now offer the same tools. If programming is your real worry, read our robot programming software guide before you let it pick the arm.

Reading cobot speed specs honestly

Cobot maximum speeds look closer to industrial numbers than they really are in use. UR lists 4 m/s maximum TCP speed for the UR10e. That's a mechanical limit with the safety configuration opened up. In a power and force limited application, you set speed limits so that a collision with a person stays under the body-region limits in the standard, and with a real gripper and part on the flange that speed is usually a fraction of the catalog number.

FANUC is unusually clear about this. The CRX-10iA/L lists 1,000 mm/s in collaborative mode and 2,000 mm/s in high-speed mode, which is only usable when the application is safeguarded another way. It's the trade-off every cobot has, written on the spec sheet.

Payload, reach and repeatability

Payload and reach mostly overlap. Cobots now run from a few kilos to about 30 kg, and our cobot payload and reach guide covers how to size one with the gripper's weight and inertia included. Industrial arms run from a few kilos to well over 1,000 kg, so heavy parts and long reach with high payload are industrial territory by default.

Repeatability is closer than people expect. ABB lists 0.02 mm for the GoFa 10, the same as FANUC's M-10iD/12. But repeatability isn't stiffness. A light arm deflects more under process forces such as drilling, deburring or press-fitting, and that shows up as path error long before it shows up in a repeatability test. For force-heavy work, the industrial arm usually wins on accuracy under load.

Here's the part buyers often miss. Only one of the four methods, power and force limiting, actually depends on the robot being a cobot. Monitored standstill and speed and separation monitoring work with any arm whose controller has the right certified safety functions, and FANUC, ABB, KUKA and Yaskawa all sell those options for their industrial lines. Hand guiding needs a guiding device and a safety function, not a light arm.

So an industrial arm with safety-rated speed and zone monitoring plus a pair of laser scanners can be a perfectly good collaborative application. It slows when you walk up and stops before you get close. What it can't do is touch you safely. If your task needs a person and the robot working on the same part at the same moment, that's where the cobot earns its place.

Our cobot safety guide covers each method in detail, with the separation-distance math and the body-region force concept behind power and force limiting.

The phrase "no guarding needed" comes from cobot brochures, and it's the most expensive misunderstanding in this category. In the US, ANSI/A3 R15.06-2025 adopts ISO 10218-1 and -2:2025, and both put the application under a risk assessment. A collaborative application is one the assessment says can share space with people using the four methods above. It's not a robot category that skips the assessment.

What tends to put guarding back? Sharp or hot parts, sheet-metal edges, knives and screwdriver bits on the tool. Payloads heavy enough that a slow bump still hurts. Pinch points between the arm and a fixed surface, like a machine door frame or a column. And moves at head or neck height, where the allowed contact forces are lowest. Our qualification and safety work usually finds one or two of these on any cell that was sold as fence-free.

What an industrial cell gives back

A fenced industrial cell has a simpler safety story. People are outside, the fence and interlocks keep them out, and the robot runs at full speed. You lose floor space and the ability to walk up to the work, and you gain a cycle time and safety case most plant safety managers already know.

For a cobot cell, you'll measure collision forces at the worst-case points with a force and pressure measurement device, document the speed limits that made them pass, and lock those settings. It's more engineering per cell. It's also the reason the cobot can stand next to an operator who's loading the next part. The 2025 standard changes are covered in our post on what changed in ISO 10218 and R15.06.

A move takes time to accelerate, cruise and decelerate. With a trapezoidal profile, a move of length d at top speed v and acceleration a takes d/v + v/a seconds when the arm has room to reach v. If the move is too short, it never reaches v, and time is set by acceleration alone. That's why a 100 mm pick feels the same on both arms and a 600 mm transfer doesn't.

The example uses three speeds. The first, 250 mm/s, stands in for a slow, contact-safe speed near a person's upper body. The second is the CRX's 1,000 mm/s collaborative-mode figure. The third is 2,000 mm/s for a guarded cell. The accelerations are illustrative too. Your real numbers come out of the risk assessment and the robot's motion planner, so treat this as the shape of the answer.

The lesson we take from it is that you shouldn't force a cobot into a cycle-bound job. If the machine you're tending has a 90 second cycle, the arm's speed barely matters and a cobot is a good fit. If you're feeding a press every six seconds, it matters a lot. Our machine tending cells are sized this way, machine cycle first.

Arm prices in the same payload class overlap, and some cobot vendors publish theirs. Our cobot price guide tracks published list prices, and the industrial robot cost guide covers the fenced side. But in a working cell the arm is usually a minority of the spend. The gripper, fixturing, part presentation, safety devices, integration, programming and validation add up faster than people expect.

The cobot cell saves in a few predictable places. There's less fence and fewer interlocks, a smaller footprint, and often a cart-mounted base you can roll to another machine. It spends more on safety engineering per cell, because force measurement and speed-limit tuning are real hours. And if it runs noticeably slower, you might need a second cell, or a second shift on the first one, to hit the same output.