A cobot, or collaborative robot, is an industrial robot built to work in shared space alongside people moving slower, sensing contact, and stopping safely instead of hiding behind a cage. It trades the raw speed and payload of a traditional robot for flexibility, quick setup, and the ability to stand next to an operator.
Cobots earned their reputation by making automation approachable for plants that could never justify a fenced robot cell. But "collaborative" is a property of the whole application, not a sticker on the arm, and the safety standards behind that word were substantially rewritten in 2025. This post covers what makes a robot collaborative, how cobots are actually kept safe, how they differ from industrial robots, the jobs they do well, and whether they pay off.
What Makes a Robot "Collaborative"?
A collaborative robot is designed so that a person can be near it, or even in contact with it, without a physical barrier, provided a risk assessment proves the specific application is safe. The arm itself typically has rounded edges, limited speed and force, and sensors that detect resistance and stop the motion. Crucially, safety is never granted by the robot alone. The end-of-arm tool, the workpiece, the speed, and the task all decide whether an application is genuinely collaborative. A cobot holding a blunt gripper at low speed may share space with an operator; the same cobot swinging a sharp blade does not, no matter what the brochure says.
How Are Cobots Kept Safe?
Cobot safety rests on international standards, and the ground shifted recently. For a decade the reference was ISO/TS 15066:2016 a technical specification that set biomechanical limits on force, pressure, and speed for human-robot contact and described four ways to make an application safe. In 2025 that guidance was folded directly into the third edition of the core robot-safety standard, ISO 10218-1:2025 (requirements for robot manufacturers) and ISO 10218-2:2025 (requirements for integrators and robot cells), which now serves as the single authoritative reference for both conventional and collaborative robot applications, adding cybersecurity and software-validation requirements the 2011 editions never anticipated. The revision even drops the phrase "collaborative robot" in favor of "collaborative application," reinforcing that safety is a property of the deployment, not the device.
Whatever the edition on the cover, the four methods of achieving safe interaction are what an integrator actually chooses among:
Two points get lost in cobot marketing. First, power and force limiting, the "it stops when it bumps you" behavior, is only one of the four, and it is the most restrictive on speed and payload. Many real cobot cells use speed-and-separation monitoring so the arm can run fast when nobody is close and slow only when a person approaches. Second, none of these methods removes the need for a risk assessment of the specific application; the standards tell you how to evaluate contact, not that contact is automatically fine.
Cobot vs Industrial Robot: What's the Difference?
The honest split is speed and payload versus flexibility and proximity. A traditional industrial robot is fast, strong, and tireless behind a fence; a cobot is slower and lighter but can stand at a bench next to a person and be redeployed to a new task in an afternoon. Neither is "better", they solve different problems.
| Attribute | Cobot | Industrial robot |
|---|---|---|
| Typical payload | ~3–35 kg | Tens to hundreds of kg |
| Speed | Reduced for safe contact | High, optimized for cycle time |
| Guarding | Often none, if risk-assessed | Fenced cell, interlocks |
| Setup | Hours; hand-guide teaching | Days to weeks; specialist programming |
| Redeployment | Easy, moved between tasks | Fixed installation |
| Best for | Low-volume, high-mix, tending | High-volume, high-speed, heavy |
A useful rule: if the job needs speed, reach, or muscle, a fenced industrial robot usually wins on cost per part. If the job needs flexibility, a small footprint, and life beside a human, a cobot wins, and it wins bigger in plants fighting the labor shortage where the goal is to take dull, repetitive tasks off scarce people rather than to run lights-out.
What Jobs Are Cobots Good At?
Cobots earn their keep on the repetitive, low-payload tasks that wear people down: machine tending (loading and unloading a CNC or press), pick-and-place and packing, palletizing light cases, screw driving and assembly, dispensing and gluing, and, paired with a camera, inspection. That last one connects cobots to computer-vision quality inspection: a cobot can present a part to a vision system or move the camera itself, automating both the handling and the check. Where a cobot struggles is anything demanding heavy force, long reach, split-second speed, or fine judgment on highly variable parts.
Payload and reach set hard limits worth respecting up front. Most cobots handle roughly 3 to 35 kilograms, and the usable payload drops once you subtract the weight of the gripper itself, a 10 kg-rated arm holding a 3 kg gripper has 7 kg left for the part. Reach is similarly modest, usually under about 1.3 meters, so a cobot suits a workbench, not a wide press line. Reading the rated payload as the part weight is the single most common sizing mistake, and it turns a promising cell into an underpowered one.
The best first application is usually the most boring one: a single machine that runs a steady part and currently ties up an operator to feed it. That job is predictable enough to automate safely and painful enough that the payback is obvious.
How Do You Deploy a Cobot Safely?
A cobot is not plug-and-play just because it lacks a fence. The deployment that goes smoothly follows a sequence, and the plants that skip the assessment step are the ones that end up bolting on guarding after the fact.
- Pick a task that fits. Steady, low-payload, repetitive, and currently manual. Resist the urge to automate the hardest job first.
- Do the risk assessment. Evaluate the whole application, arm, tool, workpiece, speed, and the people around it, against ISO 10218-2. This decides which of the four safety methods you need.
- Choose the safety method. Power and force limiting for true shared contact, speed-and-separation monitoring for faster running with zone sensing, monitored stop or hand guiding where the interaction is intermittent.
- Design the end-of-arm tooling. The gripper and workpiece are where most residual hazards live; rounded, low-pinch tooling keeps an application collaborative.
- Validate and train. Test the safety functions, verify contact stays under the limits, and train operators on what the cobot will and will not do near them.
- Monitor and improve. Track the cobot's uptime and cycle like any other asset through machine monitoring and expand only once the first cell is trusted.
Do Cobots Actually Pay Off?
Often, but the payback comes from redeployed labor and consistent output, not from a robot working for free. A cobot that frees an operator from feeding a machine all shift lets that person do work only a human can, which is the real return in a tight labor market. Payback periods are typically counted in months to a couple of years for a well-chosen tending or palletizing job, and they stretch or vanish when a cobot is forced onto a task that needed a faster, stronger robot. Count the full cost too: the arm is often the smaller line item next to tooling, integration, and the risk assessment, and a cell that runs one shift will pay back half as fast as the same cell run across two or three. The economics improve further when the cobot is not an island: connecting it to the plant's IIoT layer means its cycles, stops, and output roll into the same picture as everything else, so its contribution to throughput is measured, not assumed. That connectivity is the difference between a clever gadget and a monitored asset inside a smart factory.
By the Numbers
The safety framework is now consolidated: the 2025 third editions of ISO 10218-1 and ISO 10218-2 absorbed the collaborative-robot guidance formerly held in ISO/TS 15066:2016 so integrators work from one standard covering both conventional and collaborative applications, including the force and pressure limits for human contact. The demand pressure is real too: U.S. Bureau of Labor Statistics JOLTS data has shown on the order of 400,000–550,000 open manufacturing jobs in a typical recent month (BLS JOLTS), which is the labor gap cobots are most often bought to relieve. Where Harmony fits: a cobot is one more asset on the floor, and its value shows up only when its output is connected to everything else. Harmony ties machine and cell data, cobots included, into one real-time operational layer, so automation is measured alongside the rest of the plant (see how Harmony connects machines and systems or a real deployment).