Motionwell Automation builds both kinds of robot cell in Singapore, and the cobot vs industrial robot question usually arrives before a layout has been drawn. We integrate ABB industrial robots and JAKA collaborative arms, with seven JAKA units purchased to date, alongside Yamaha SCARA arms for lighter handling. The honest headline: the decision is made by rate, payload, reach, duty cycle and how often a person is genuinely inside the space, not by which technology sounds more modern, and the arm is rarely the largest line on the bill either way. Machines are designed, assembled and tested at our Woodlands Link facility, and the company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3.
Where we stand, said plainly before you read further. We do not manufacture robot arms, cameras or drives. We buy them, and what we design is everything between the arm and your product: the tooling, the frame, the feeding and presentation, the safeguarding, the control system and the recovery logic. So we have no commercial reason to push you towards a collaborative arm, and a section below is devoted entirely to when a cobot is the wrong purchase. We are not a notified body and do not issue CE certificates.
This page covers what actually differs between the two classes, the arithmetic of throughput per square metre, whether a fence costs more than the speed you give up, when each wins, the middle path where an industrial arm is made safe rather than replaced, the hybrid cell that sits between the two pure options, and a decision checklist. The end-of-line version of the same argument is on our automated palletizing systems page, the reasoning behind a collaborative application is in our guide to collaborative robot safety standards, and the guarding work that follows either choice is on the machine safety and CE marking page. If you have a part weight, a reach and a rate target, skip ahead and talk to an engineer.
What Actually Differs Between a Cobot and an Industrial Arm?
Less than the marketing suggests in the arm, and more than expected in everything around it. Set out as collaborative vs industrial robot, both are articulated machines with servo joints, an encoder per axis and a controller. The difference is that a collaborative arm carries safety-rated functions intended for working close to people as standard, while an industrial arm is specified for a guarded cell unless safety-rated speed, standstill and axis limiting are added on its controller.
| Selection factor | Collaborative arm | Industrial arm |
|---|---|---|
| Rated speed | Capped in collaborative operation so contact stays inside the limits | Full programmed speed, limited by the mechanism rather than by contact |
| Payload class | 3 to 25 kg across published vendor datasheets | 3 to 300 kg and above across published vendor datasheets |
| Reach class | 500 to 1300 mm across published vendor datasheets | 500 to 3000 mm and above across published vendor datasheets |
| Datasheet repeatability | ±0.03 to ±0.10 mm | ±0.02 to ±0.05 mm |
| Duty cycle | One or two shifts with natural gaps | Designed for continuous running |
| Guarding | Possibly none, decided by the risk assessment | Fence, interlocked doors, scanners or light curtains |
| Footprint | Fits beside an existing line without a cell around it | Cell area plus safe operator and forklift access |
| Redeployment | Move to another line and re-teach in a shift | Fixed installation |
| Relative cost at equal payload | High | Highest |
Two cautions on that table. The payload, reach and repeatability figures are ranges published across vendor datasheets for each class, not measurements from any one machine, so they show relative strengths rather than specifying a purchase. And repeatability is not placement accuracy: it says nothing about gripper slip, part tolerance, fixture wear or thermal drift, which usually dominate what you measure on the finished assembly.
The last row is easy to misread. On the class comparison published on our SCARA robot selection guide, a six-axis industrial arm is the highest relative cost at equal payload and a collaborative arm sits one class below it. That is an ordinal ranking and not a price list: it tells you which class is dearer at equal payload, not by how much, and it says nothing at all about the cell built around the arm. The cost list in that same guide is the more useful part of it — end-of-arm tooling, part feeding, vision, fixturing and safeguarding, with the robot rarely the largest line on the bill. Compare the cells before you compare the arms.
One more difference no datasheet states. Rated payload is measured at the tool flange and assumes the load sits close to it, so the tooling comes out of the budget left for the part, and at reach the moment rather than the mass is what limits the arm. Read the payload-versus-centre-of-gravity chart before choosing a model.
How Much Throughput Does a Force Limit Actually Cost You?
This is the arithmetic that decides most of these projects, and it is usually done too late. A collaborative arm run as a power-and-force-limited application has its speed capped so any contact stays within the contact limits. That cap is the technology working as intended, and it is also the rate ceiling. You do not get around it by buying a faster collaborative arm.
Put numbers on the other end of the range. On the palletizing work we have delivered, a four-axis column platform runs 6 to 10 cartons per minute, and where a line needs more than 12 per minute we specify the ABB IRB 460: a four-axis palletizing robot with 110 kg payload and 2.4 m reach, rated above 2,190 cycles per hour, used for standard carton palletizing up to 25 kg per case. A cobot palletizer is the right choice when your case rate sits in the low single digits per minute. Those are not adjacent numbers.
Now the part that gets missed: floor area belongs inside the comparison, not beside it. A fenced cell needs the cell area plus safe operator and forklift access, which is genuinely more square metres than an arm standing beside a conveyor. But throughput per square metre is a ratio, and the fenced cell can be several times larger in the numerator. Work it out on your own layout: sustained parts per hour on top, and underneath the whole floor area the option occupies including access and aisle. A cell that takes more floor and produces far more per hour can still win in a plant paying Singapore rent, and it cannot win at all in a plant with no aisle to spare, because it does not fit.
One caution on the numerator. A datasheet cycle time is measured gate to gate with a dummy load and nothing else happening. Your real cycle adds vision acquisition and processing, settling before the vacuum releases, gripper actuation, the PLC handshake and any wait on a conveyor. Those additions, rather than the arm, usually decide whether a station makes rate, which is why we time them on the actual parts before quoting either class.
Is the Fence Cheaper Than the Speed You Give Up?
Often, and it is worth seeing why the two costs are not the same shape.
Guarding is a one-time bill of materials plus engineering: fence panels and posts, an interlocked access door, safety laser scanners or light curtains, the safety circuit, the stopping distance measurement and the validation evidence. It is bought once. Lost throughput is recurring. It lands on every shift for the life of the machine, and if the line is capacity-constrained it does not appear as a cost at all, it appears as revenue you never booked.
We do not publish prices, because the same nominal cell moves a long way on decisions taken before hardware is ordered. What we can be specific about is where the two bills separate.
| Line item | Collaborative cell | Fenced industrial cell |
|---|---|---|
| The arm | High relative cost at equal payload | Highest relative cost at equal payload |
| Perimeter guarding | Possibly none, decided by the risk assessment | Fence, posts, interlocked doors, scanners or light curtains |
| Safety engineering | Contact assessment against the real tooling and workpiece, measured force and pressure readings, validation records | Risk assessment, performance level architecture, measured stopping distance, validation records |
| End-of-arm tooling | Rounded edges, limited protrusions, controlled pinch points, compliant features where suitable | Sized for payload and cycle, without a contact constraint on its geometry |
| Floor area | Working envelope plus the space a person occupies beside it | Cell area plus safe operator and forklift access |
| Feeding, presentation and controls | The same problem either way, and a bowl feeder tuned to an awkward part can approach the price of the arm | As on the left |
Read that table and the conclusion is uncomfortable for both camps. Only the bottom row is genuinely identical: feeding, presentation and controls cost what they cost, and none of it gets cheaper because the arm has a force limit. Two rows differ in kind rather than in amount, since both cells need an arm and both need safety engineering, but neither the work nor the price is the same work and price. The three rows that actually decide the comparison are perimeter guarding, tooling geometry and floor area. That is why we quote after a layout review rather than from a catalogue.
Where the fence stops being the cheaper answer is equally clear. If demand is genuinely below what a force-limited arm delivers, the extra speed buys nothing, and paying for guarding to get throughput you will never sell is the same error in the other direction.
When Is a Cobot Genuinely the Right Answer?
Six situations where a collaborative arm earns its place, four of them carrying delivered work behind them and two of them selection rules we apply before a layout exists.
Low volume with high mix. When changeover cost dominates cycle time, a reprogrammable loading device beats a faster one. On our cleanroom automated test equipment build, a dedicated gantry would have been faster per cycle, but with mixed connector types and modest volume per variant the loading device had to be reprogrammable rather than quick. The collaborative arm holds ±0.05 mm repeatability at the test fixture interface, calibrated against a reference pin at the start of every batch. What was given up, stated plainly in that project’s own record, is peak throughput on any single high-volume part number, and the build is in the cleanroom ATE case study.
Shared space that cannot be divided. In the QA lab automation programme, re-ordered in four consecutive years by a global medical device manufacturer, technicians work in the same room while a compound AMR and collaborative arm move samples between racks and instruments. Fencing the automated stations would have been simpler to justify in the risk assessment, and would have cut the lab in two. Station-level safeguarding was used instead.
Frequent human intervention. If a person enters twice a shift to reload a tray, a safety-rated monitored stop is simpler, cheaper and faster than tuning force limits. If the operator works next to the robot all shift, power and force limiting is the only mode that will not destroy the cycle time. Which of those applies is a question about your process, not about the robot.
Floor space that cannot take a fence. Where floor area is the binding constraint we mount JAKA collaborative arms overhead for pick-and-place and capping, which frees the entire floor for conveyors and operator access while keeping full reach. It costs easy maintenance access, so it is a trade rather than a free upgrade.
Short redeployment cycles. A collaborative arm moves to another line and is re-taught in a shift; a fenced cell is a fixed installation. If your plan genuinely moves work between lines, that flexibility is worth real money, and it appears nowhere on a datasheet.
Modest payloads with awkward variety. The cobot unloading station built for a consumer goods distribution centre handles mixed SKUs across beverages, snacks and detergent with a five-magnet electromagnet quick-change end-of-arm tool that switches between gripper and vacuum modes on its own, and X and Y pitch adjustment on stepper motors so one tool covers several pack sizes. The variety, not the speed, is what that cell was designed against. More on the FMCG and co-packing automation page.
When Should You Not Choose a Cobot?
This is the section that decides whether the rest of the page is worth trusting. Cobots are oversold, and these are the situations where a guarded industrial arm is the better engineering answer.
The hazard is the payload, not the arm. A force limit is irrelevant if the part is a sharpened blade, a hot moulding or a live battery module. On our EV battery dismantlement line the workpiece is energised and cannot be switched off: modules weigh 30 to 80 kg, the pack carries 400 V and above, and two ABB IRB 6700-200/2.60 robots at 200 kg payload and 2.6 m reach do the fastener removal, busbar disconnection and module extraction inside fenced stations. Collaborative rules were deliberately not used to justify an open cell there, and the reasoning is in the battery dismantling case study.
The tooling cannot be made safe. Long, thin or pointed end-effectors concentrate force, and some geometries cannot pass a transient contact assessment at all. That constraint usually arrives after the arm is bought, which is why the assessment belongs in concept review: redesigning a gripper after commissioning costs several times more than addressing it during concept design.
Cycle time dominates. If your takt time needs full robot speed, you will fence the cell anyway and pay the collaborative premium for nothing. The rule of thumb we apply internally: if the line runs three shifts, or the rate is above what one arm can hold on a single-case pick, we will tell you to buy the industrial robot.
Nobody is ever actually present. If the cell runs unattended, collaboration buys nothing and costs throughput. Safety functions protecting a person who is not in the room are a waste, and the same is true where one high-volume part number runs all day: the flexibility a collaborative arm buys is flexibility you are not using.
The payload or the reach is out of band. A part plus tooling that only just fits inside the rated payload does not fit, because part weights drift and someone adds a heavier variant next year. Reach fails the same way: an arm at floor level cannot place the top layer of a full pallet, and the usual fix is a lift column or a linear seventh axis, which is added cost and added footprint. Once you are paying for a column, a pedestal and a fence anyway, compare that total against a purpose-built industrial palletizer first.
The duty cycle is continuous. A collaborative arm suits one or two shifts with natural gaps; an industrial arm is designed for continuous running. This difference shows up long after the acceptance test rather than during it, which is exactly why it has to be settled before purchase.
The environment is harsh. Washdown, dust, chemical splash and temperature have to be checked against the arm’s own ingress rating, and there is a second-order trap: a protective jacket or cover changes the contact geometry the collaborative assessment was based on. A cover is not a neutral accessory on a force-limited arm. It is a change to the safety case.
Two exclusions while we are being direct. We do not build production welding cells, and we do not tell you a platform is our standard when it is not on our shop floor. Where a proven standard machine covers your application at a lower price than anything we would build, the useful answer is to say so.
Can an Industrial Arm Be Made Safe Instead of Replaced?
Yes, and it is the option often missing from a comparison framed as two choices. Safety functions on the robot controller can limit an industrial arm’s speed, standstill and permitted axis range in a safety-rated way, which changes what safeguarding the cell needs without changing the arm.
On the battery dismantlement line, ABB SafeMove2 safety functions running on the ABB OmniCore controller provide safe speed monitoring, safe standstill and safe axis range limiting, with an ABB Pluto B46 safety PLC monitoring the safety circuits independently of the robot controller. Each function is watched by two processing channels that cross-check each other continuously, and if either sees a parameter exceed its programmed limit the system triggers a controlled stop within the category 1 stop time, typically under 500 ms for those payloads. That dual-channel architecture is a fundamental requirement for reaching ISO 13849 Performance Level d on those functions.
| Safeguarding route | What limits the robot | What you have to prove | Where it fits |
|---|---|---|---|
| Power and force limiting | Contact energy, so speed is capped by the limits | Measured force and pressure at representative contact points, against the real tooling and workpiece | Close, frequent interaction all shift |
| Speed and separation monitoring | Measured distance to a person, dynamically | Scanner zone verification, and a separation distance calculated from approach speed, robot speed, measured stopping distance and sensor uncertainty | Shared area with predictable traffic |
| Safety-rated monitored stop | The robot stops when someone enters | Stop function performance level and reliable restart logic | Occasional intervention, machine tending |
| Safe speed and axis limiting on an industrial arm | Safety-rated speed, standstill and axis range in the controller | Dual-channel monitoring, stop category and time, performance level for each function | An industrial arm that must slow or restrict itself near people or fixtures |
| Fixed guarding at full speed | Nothing, the arm runs at programmed speed | Guard integrity, interlock performance level, access and recovery procedure | Sustained rate with no routine human entry |
The fourth row is the one worth adding to your shortlist. It lets an arm sized for the payload and the rate run at full speed when the cell is closed, then drop to a safety-rated speed or a restricted envelope when a person needs access. It is not a licence to remove the fence. It changes how much guarding the assessment demands, and where the access points can be.
Is the Hybrid Cell the Honest Answer?
On a line running mixed volumes it often is, and it belongs on the shortlist beside the two pure options rather than being treated as a compromise between them.
The shape is a fenced industrial arm doing the work that needs rate, payload or reach, with collaborative loading stations at the boundary where people hand parts in and take finished work out. The person never enters the fast envelope, and the robot never runs slowly to protect somebody who is not standing next to it. The interface is a buffer, a rack or an indexing table, and the safety case is drawn around that interface rather than around the whole cell.
The battery line is that shape without the label: fenced robot stations do the disassembly, and the risk assessment deliberately separated them from the manual loading area at the head of the line. The palletizing cell follows the same logic differently, guarded to ISO 13849 Performance Level d with SICK M4000 safety light curtains at the pallet removal openings, so the forklift driver interacts with the pallet position and never with the arm.
Two conditions have to hold for a hybrid to be worth its extra interface. The manual and automatic work must be genuinely separable, which is a process question rather than a layout question. And somebody has to own the handshake between the zones: part present, zone clear, station ready, fault. Every interface is a place where two suppliers each assumed the other owned it.
Which Standards Decide the Guarding Scope?
The assessment decides the design, not the other way round. Start from ISO 12100 risk assessment, which identifies the hazards and sets what has to be reduced. The robot and the integrated system are then covered by ISO 10218-1:2025 and ISO 10218-2, with each safety function rated for performance level under ISO 13849-1:2023.
Two editions matter this year. ISO 10218-1 was published in February 2025 as the third edition and the first substantive revision since 2011, adding robot classifications with matching functional safety requirements, safety-related cybersecurity requirements and end-effector guidance. Most of what ISO/TS 15066:2016 said about collaborative operation has moved into Part 2 rather than sitting in a separate technical specification, so if you are writing a specification or a safety file now, check which document your contract names. ISO 13849-1:2023 is the edition ISO 10218-1:2025 references for robot control system safety functions, so a design documented against the 2015 edition will need its performance level calculations restated when the machine is re-assessed.
Singapore does not require CE marking, but the Workplace Safety and Health Act requires dangerous parts of machinery to be securely fenced and places duties on the occupier and on anyone supplying machinery for use at work. Most multinational manufacturers here specify CE conformity as an internal standard anyway. Motionwell delivers the physical scope that follows either choice: guard fencing and interlocked access doors, safety laser scanners, safety-rated stop circuits, LVD and CE testing, and Ministry of Manpower lifting certification where the machine includes lifting equipment.
Which One Should You Specify for Your Application?
Run these seven questions in order. The first that gives a hard answer usually settles it, and if two disagree the honest answer is the hybrid cell above.
- What is your sustained rate, not your peak? The number the line has to hold across a shift. If it needs full robot speed, the guarding decision is already made.
- What is the true payload? Heaviest part plus tooling and cabling, checked against the payload curve at your real centre of gravity and reach. If it only just fits, it does not fit.
- What reach does the layout demand? Furthest pick to furthest place, measured on the drawing including the top of a full stack.
- How many shifts, and how many gaps? Continuous three-shift running points at an industrial arm regardless of everything else here.
- How often is a person inside the space, and what for? Twice a shift to reload is a monitored stop. All shift, elbow to elbow, is power and force limiting. Never is a fence.
- Where does the hazard sit? If it is the workpiece or the tooling rather than the arm, no collaboration mode fixes it and physical safeguarding follows.
- How often does the work move? Frequent redeployment between lines is the one advantage a collaborative arm holds that no fenced cell matches.
| Your situation | Start from | Why |
|---|---|---|
| Low volume, many variants, changeover dominates | Collaborative arm | Reprogrammable beats quick when the mix changes more often than the parts |
| Sustained double-digit rate on one part number | Fenced industrial arm | The force limit is the rate ceiling, and you are not using the flexibility |
| Operator works beside the machine all shift | Collaborative arm, power and force limiting | The only mode that survives continuous human presence |
| Heavy parts, long reach, full pallet height | Industrial arm | Payload and reach are out of the collaborative band before safety is discussed |
| Energised, sharp, hot or otherwise hazardous workpiece | Fenced industrial arm | The hazard is the part, so limiting the arm changes nothing |
| Three shifts, unattended running | Industrial arm | Duty cycle, and no person to protect |
| No aisle to spare on an existing line | Collaborative arm, possibly overhead mounted | The fence costs floor area the plant does not have |
| Manual and automatic work genuinely separable | Fenced industrial arm with collaborative loading stations | Rate where it is needed, access where it is needed |
Where the answer comes out mixed, that is not a failure of the checklist. It usually means the cell has two zones in it, and the design question moves to where the boundary sits and who owns the handshake across it.
Which standard editions apply right now?
The editions below are the ones we design and document against on current projects. We check them on the date shown rather than assuming last year's edition still holds.
| Standard | Current edition | What it means for your machine |
|---|---|---|
| ISO 10218-1 — Robotics, safety requirements, Part 1: industrial robots | ISO 10218-1:2025 | Published February 2025, the third edition and the first substantive revision since 2011. It adds robot classifications with matching functional safety requirements, safety-related cybersecurity requirements, and end-effector guidance. Most of ISO/TS 15066:2016 on collaborative operation moved into Part 2. |
| ISO 13849-1 — Safety of machinery, safety-related parts of control systems | ISO 13849-1:2023 | The 2023 edition is the version referenced by ISO 10218-1:2025 for robot control system safety functions. Designs still documented against the 2015 edition will need their PL calculations restated when the machine is re-assessed. |
Frequently Asked Questions
Is a cobot safer than an industrial robot?
Not by itself. Safety is a property of the application rather than of the arm: the workcell, tooling, workpiece and speeds together decide the risk level. A force limit on the arm does nothing about a sharpened blade in the gripper, a hot moulding or a live battery module. Many production cobot cells still end up with safety scanners, light curtains or restricted zones because the hazard sits in the payload rather than in the robot. The risk assessment decides the safeguarding, and it has to be run either way.
How much throughput does a cobot give up against an industrial arm?
Enough to change the machine choice, and the number has to be yours rather than ours. A collaborative arm run as a power-and-force-limited application has its speed capped so that any contact stays within the contact limits, which is the whole point of the technology and also where the rate ceiling comes from. As a published anchor at the other end of the range, the ABB IRB 460 is a four-axis palletizing robot with 110 kg payload and 2.4 m reach, rated above 2,190 cycles per hour. Time your own cycle rather than trusting a datasheet: vision acquisition, gripper actuation, settling and the PLC handshake usually decide whether a station makes rate.
Can I fence a cobot later and run it faster?
Partly, and it is usually the expensive way round. You get around a force limit by adding guarding and speed, at which point you have built a small industrial cell with an expensive arm in it, and the arm's own rated speed is still the ceiling. Check that rated figure against your target rate before assuming a fence solves the problem. Removing power and force limiting also reopens the safety case: the risk assessment, the safety functions, their performance levels and the validation evidence all have to be redone. Decide the guarding strategy at concept stage instead.