Motionwell Automation designs, builds and integrates collaborative robot cells in Singapore, and the useful starting point for anyone asking about cobot integration in Singapore is that “collaborative” describes the application, not the arm. Seven JAKA collaborative arms have been purchased to date. JAKA and HitBot cobots are deployed across medical device, electronics and general manufacturing sites here, we have integration experience on customer-supplied Universal Robots arms across three projects, and grippers come from Schunk and DH Robotics. The delivered references behind this page are a collaborative arm riding an autonomous mobile robot in a QA laboratory serving multiple test instruments, a cobot unloading station for a consumer goods distribution centre, and JAKA arms mounted overhead for pick-and-place and capping where floor area was the constraint. Cells 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. A collaborative robot is a purchase. A collaborative application is an engineering result, produced by a risk assessment covering the arm, the tooling, the workpiece, the speeds, the fixture and the people who share the space. Fit a sharp tool or a heavy part and the application stops being collaborative while the arm on the purchase order stays the same. We build cells both ways, and where a fenced industrial arm is the better and cheaper answer we say so at concept review rather than quote around it.
Below: what the four collaborative modes permit, how the two production modes differ in cost and cycle time, how force limits get verified and why tooling and workpiece usually set them, which applications a cobot earns its place in, what changes on a mobile base, and where a guarded industrial robot is the right machine. The wider safety workflow sits on our machine safety and CE marking page and in our note on collaborative robot safety standards. If you already have a task, a part and a cycle time, talk to an engineer.
What Makes an Application Collaborative, Rather Than the Robot?
An arm sold as collaborative has safety-rated functions built in: monitored speed, monitored position, monitored force or torque, and a stop architecture that can be relied on rather than merely programmed. Those are real and worth paying for, and they describe the arm on its own, with nothing on the flange and nothing in the gripper. Your cell is the arm plus a tool, a part, a fixture, a bench edge and a person, and any of those can carry the hazard.
Four situations turn a collaborative robot into a non-collaborative application, and all four are common:
- The tool is the hazard. A blade, a hot tip, a screwdriver bit, a probe, a thin nozzle. Small contact areas concentrate pressure, and some geometries cannot pass a contact assessment at any usable speed.
- The workpiece is the hazard. A sharp-edged pressing, a glass container, an energised assembly. On the battery line below, the pack is live and the arm is the least dangerous object in the cell.
- The payload is the hazard. Force limiting governs what the arm does to a person. It says nothing about a part dropped on a foot.
- The geometry is the hazard. A hand trapped between the tool and a fixed surface cannot move away, and the standards assess that case more strictly than a free impact. The fixture layout creates it, not the robot.
Sorting this out uses the ordinary machinery method. ISO 12100 fixes the order of risk reduction: inherently safe design first, safeguarding second, information for use last. A cobot cell is still a robot system, so ISO 10218-1 covers the robot and ISO 10218-2 the integrated system, which is the part a machine builder owns. Most real cells therefore end up hybrid: collaborative behaviour at the one station where a person needs to be, physical safeguarding where the risk actually sits.
What Do the Four Collaborative Operation Modes Actually Permit?
Four modes are recognised, with no ranking among them. Most cells use more than one in sequence, because the robot does different things at different moments in the cycle.
| Mode | What it permits | What the cell needs | What it costs |
|---|---|---|---|
| Safety-rated monitored stop | Holds position while a person is in the shared space, resumes when it is clear. Suits occasional intervention: reloading a tray, clearing a jam | Safety-rated presence sensing, monitored standstill, restart logic | All throughput while a person is present. Cheap if intervention is rare |
| Hand guiding | Motion commanded through a guiding device with an enabling switch and a safety-rated speed limit. Suits teaching, setup, lifting assistance | Enabling device, reduced-speed mode, emergency stop in reach | An operator present for every motion, so it does not scale to production volume |
| Speed and separation monitoring | Runs fast when nobody is near, slows and stops as a person approaches. Suits shared aisles with predictable traffic | Safety scanners, zoned speed profiles, measured stopping performance, floor space | Floor area and sensing. It fails safe by stopping, so a busy aisle stalls the cell |
| Power and force limiting | Contact permitted but bounded, so it stays within body-region limits. Suits close interaction through most of the cycle | Verified force and pressure measurement with the real tool and part, rounded tooling | Cycle time, permanently. The speed cap is the mechanism, not a tuning parameter |
Mode selection is a question about people, not about robots. Ask how often someone is truly inside the space and what they do there. Reloading a tray twice a shift makes a monitored stop simpler, cheaper and faster than tuning force limits. An operator working alongside the arm all shift makes power and force limiting the only mode that will not destroy the cycle time. And if the hazard is the gripper or the part, no mode fixes it: that hazard needs safeguarding of its own.
Should You Specify Speed and Separation Monitoring or Power and Force Limiting?
These two carry most production cobot cells, chosen for opposite reasons. One buys speed at the cost of floor area and sensing. The other buys proximity at the cost of speed.
| Speed and separation monitoring | Power and force limiting | |
|---|---|---|
| Underlying idea | Never touch. Hold a separation distance that shrinks as the robot slows | Touching is allowed, but bounded by measured limits |
| What is engineered | Sensing coverage, zone geometry, speed per zone, stopping performance | Tool and part geometry, contact area, speed cap, verified force and pressure |
| Speed when nobody is near | Full application speed | Still capped, because limits apply whenever a person can be present |
| What defeats it | Cluttered floors, forklifts, a separation distance the room cannot give you | A tool or workpiece geometry that cannot pass a contact assessment |
| Floor area | Significant, and it is floor you cannot fill with pallets | Minimal, often the real reason it gets specified |
| Validation evidence | Stopping distance measurements, scanner zone verification, safety-distance calculations | Force and pressure records at each credible contact point, by body region |
| Best fit | Machine tending and transport, where a person passes through occasionally | Assembly, inspection loading and lab tending, with a person beside the arm |
The detail worth checking in any quotation is that the separation distance is not a number to be looked up. It comes from the approach speed of a person, the robot’s current speed, the measured stopping distance of the whole system and the uncertainty of the sensing, so zones, speed profile and stopping performance are one coupled problem. Sensor positioning on our machines follows ISO 13855, which sets the approach speed constant at 2000 mm/s up to 500 mm and 1600 mm/s beyond. Guess it and you build a cell that looks right and stops too late.
One argument for power and force limiting is easy to miss on a datasheet: its safety case is built around the tool, the part and the surfaces near the arm rather than around a floor plan. Separation monitoring draws its zones against a layout, so a pallet stand left standing in a scanner zone trips the scanner and stalls the cell, and the safety case is lost at the moment someone reshapes the zone around the obstruction instead of moving it.
How Are Force Limits Verified, and What Sets Them?
By measurement on the built cell, with the actual end effector and the actual workpiece. The arm’s published figures cannot stand in for that measurement, because they are established on the bare arm, before your tool and your part are on it.
Force and pressure are two different results. Pressure is force divided by contact area, so a gripper finger with a small radius, a part corner, or the tip of a screwdriver bit can fail a pressure limit at a force that passes with room to spare. Rounding an edge to increase the contact patch is often the cheapest fix available, and it costs nothing at concept stage.
Contact type changes the limit. The standards draw a distinction here: transient contact, where a person can move away, is assessed separately from quasi-static contact where the body is held between the moving tool and a fixed surface, and the quasi-static case is the more restrictive of the two. The layout creates that case: a bench, a fixture wall, a conveyor rail. Half the design work in a force-limited cell is removing the surfaces a hand could be pinched against.
Limits are body-region specific. Contact acceptable against a forearm is not necessarily acceptable against a hand or a face, so measurement is taken where contact is credible given the arm’s reach and the operator’s posture.
So the tooling and the workpiece set the limit, not the robot. Across our medical device and electronics assembly work, redesigning a gripper after commissioning costs several times more than addressing it at concept review, and a compliant fingertip or a rounded bracket edge can move a contact scenario from unacceptable to acceptable without touching the robot program.
So design the tool against the contact assessment at the same time as against the part, and treat the evidence as a test plan written at concept stage rather than a report written afterwards. On regulated lines it is referenced from the equipment qualification protocols rather than filed separately, the discipline described on our computer system validation page. Each safety function is rated for performance level under ISO 13849-1, as on any other machine.
Which Applications Does a Cobot Genuinely Earn Its Place In?
The honest test is whether a person is really in the space, doing something a machine cannot cheaply take over, for a meaningful part of the cycle. If not, the collaborative premium buys nothing.
| Application | What the arm does | What usually sets the limit |
|---|---|---|
| Machine tending | Loads and unloads a machine that would otherwise wait for an operator between cycles | Part edges and swarf. For high-duty tending across many machines we have used linear gantry systems instead, because reach and duty beat proximity |
| Laboratory instrument tending | Loads specimens into instruments, retrieves them, returns them to a rack | The instrument’s control interface, not the arm. An instrument with no accessible interface is the project risk |
| Inspection and vision-station loading | Presents a part to a camera at a repeatable position and orientation, then sorts on the result | Presentation repeatability and lighting access. The arm is now part of the optical setup |
| Dispensing | Carries a dispense valve along a path, or presents the part to a fixed valve | Path speed stability, and the material. A heated or reactive material is a hazard the arm cannot limit |
| Packaging, unloading and decanting | Moves units between cartons, totes and trays where the pattern changes per SKU | Tooling versatility and pitch adjustment. Cycle rate is usually the ceiling |
| Screwdriving and fastening | Drives fasteners to a recorded torque on a part a person also works on | The bit. A protruding driver tip concentrates the whole contact force onto a very small area, which is precisely what the assessment measures |
Palletizing beside an operator belongs there too, limited by rate and reach, and our palletizing systems page works it through. The delivered work behind several of those rows is worth being specific about.
In a QA laboratory for a medical device manufacturer, a collaborative arm on an autonomous mobile robot retrieves samples from storage, loads them into universal testing machines and returns them, under a warehouse management layer running natively in an Allen-Bradley PLC that manages 70 positions across 7 racks. Technicians work in the same room throughout, which is why the cell is collaborative rather than fenced. The build is in the QA lab automation case study, the architecture around it on our laboratory automation capability page.
At a consumer goods distribution centre, a cobot unloading station decants mixed SKUs from cartons into tote bins with a 5-magnet electromagnet quick-change 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 product sizes without a mechanical changeover. It reads the carton ID at the infeed and reconfigures from a tool change rack, as set out on the FMCG and co-packing automation page. Where floor area is the constraint we mount JAKA arms overhead for pick-and-place and capping, freeing the floor for conveyors and operator access at the cost of harder maintenance access. The capping version uses a custom end-of-arm tool with pneumatic torque control and cap presence verification, described on the capping and sealing systems page.
One boundary on screwdriving. Fastening is well within scope: on our SCARA sensor panel assembly line, electric screwdrivers work to a target torque set per fastener in the recipe from 0.1 to 2.0 Nm, measured by a rotary torque transducer in the spindle, with final torque and rotation angle logged against the panel serial number. That instrumentation transfers to a collaborative arm unchanged. The safety case does not, because a driver bit is the small-radius protruding geometry contact assessments struggle with, so that station may need guarding even in an otherwise collaborative cell.
What Changes When the Cobot Rides on a Mobile Base?
Everything about the safety case, because the collaborative space stops being a place on the floor plan and becomes something the vehicle carries around.
| Fixed-base cobot cell | Cobot on a mobile base | |
|---|---|---|
| Where the collaborative space is | One location, drawn on the layout, assessed once | Wherever the vehicle is, plus everywhere it travels between stations |
| What the assessment covers | The cell, its tooling, its workpieces and its access points | The vehicle in transit, plus every docking station as its own application |
| Who protects the person | The cell’s own sensing and guarding | On-board scanning while driving, the station assessment while working: two cases that must not overlap |
| What breaks it later | A new fixture or a changed part | A moved bench or rack, a pallet in the aisle. The floor is not under the robot’s control |
Two design rules keep this tractable. Sequence the base and the arm rather than driving and reaching at once, so the two safety cases do not have to be superimposed. And treat each docking station as its own application rather than approving “the robot” once, because the station determines who stands nearby and what the arm can reach. The QA laboratory system is layered accordingly: on-board scanning while the vehicle drives, defined traffic zones on the lab floor, and a cell-level assessment at each station covering the gripper and the sample.
The positioning stack there is worth borrowing whatever base you use. SLAM navigation gets the vehicle to within about plus or minus 50 mm, nowhere near a fixture handoff, and the arm’s joint positioning repeats to within plus or minus 0.1 mm at the tool centre point but only relative to wherever the base parked. A wrist camera closes the gap: it reads fiducials on the tray or fixture and corrects the X, Y and theta offset before every pick and place, bringing final placement to within plus or minus 0.5 mm. Skip that correction and the cell works in commissioning, then starts dropping parts in month three when floor markings wear and rack legs get nudged. Our AMR versus AGV selection guide covers the vehicle choice itself.
Where Is a Fenced Industrial Arm Simply the Better Answer?
Cobots are oversold, and saying so costs us nothing, because we build both.
The hazard is the workpiece. Our EV battery dismantling line is the clearest example we have. The disassembly stations use two ABB IRB 6700-200/2.60 industrial robots at 200 kg payload and 2.6 m reach, fenced, with every safety function engineered to ISO 13849 Performance Level d. The packs are live at 400 V and above, the end effectors are insulated, and collaborative rules were deliberately not used to justify an open cell. Force limiting has nothing to say about high voltage. The EV battery dismantling case study sets out the reasoning.
Cycle time dominates. A power-and-force-limited application has its speed capped so contact stays inside the limits. That cap is the technology, not a setting, and it is where the rate ceiling comes from. You do not get around it with a faster arm; you get around it by adding guarding and speed, at which point you have a small industrial cell with an expensive arm in it. Our rule of thumb on end-of-line work: three shifts, or a rate above what one arm holds on a single pick, and we will tell you to buy the industrial robot.
The tooling cannot be made safe. Long, thin or pointed end effectors concentrate pressure, and some geometries cannot pass a contact assessment at any useful speed. Redesign the tool if you can; if the process needs that geometry, guard it.
Nobody is actually present. A cell running lights-out gains nothing from collaboration and loses throughput. Check who is genuinely inside the space before the arm is ordered, because a cobot bought for an unattended cell pays the collaborative speed cap for nothing.
Reach or payload is the constraint. Rated payload assumes the load sits close to the flange, and a tool holding a part out from the wrist creates a moment that limits the arm long before mass does. Read the payload-versus-centre-of-gravity chart at the real tool centre of gravity and the real reach. If the heaviest part only just fits, it does not fit.
What Do We Supply on a Collaborative Cell, and What Do We Buy In?
We do not manufacture robot arms, cameras or drives, and we are not a cobot distributor. We buy arms and integrate them: JAKA collaborative arms, seven units purchased to date; HitBot for light assembly and tray handling in tight footprints; ABB and Yamaha SCARA where the work is not collaborative at all; and Youibot AMRs where pallets, totes or samples move between cells. We have integration experience on customer-supplied Universal Robots arms across three projects, where the customer had already standardised on that arm.
What we design and build is everything between the arm and your product: the end-of-arm tooling, the fixtures and part presentation, the station frame, the safeguarding and safety circuit, the PLC and HMI, the recipes, and the data the cell produces. Grippers come from Schunk and DH Robotics where a catalogue unit fits, with application tooling designed in-house where it does not. Vision is Keyence and Cognex, lit and calibrated per application, as covered on our machine vision inspection page.
Three honest exclusions. We do not issue CE certificates and we are not a notified body, so third-party certification is not ours to give; we implement and document, and arrange accredited LVD and EMC testing with an external laboratory. We do not sell a catalogue cobot cell off a shelf, and where a standard machine or a proven fenced cell covers your task more cheaply, that is what we say at concept review. And we do not build production welding cells, which we decline outright.
Which Standards Does a Collaborative Cell Have to Answer To?
One change is worth knowing before you write a specification this year. ISO 10218 was revised in 2025, and most of what ISO/TS 15066:2016 said about collaborative operation now sits inside ISO 10218-2 rather than in a separate technical specification. The engineering does not change; which document your contract and safety file name does. Where the cell is partly guarded, the ordinary standards still decide the physical design: ISO 13857 for reach distances, ISO 13855 for scanner and light curtain positioning, ISO 14119 for interlocking devices and how hard each is to defeat.
For machines installed in Singapore, CE marking is not a legal requirement, 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 regardless. Where the machine ships into the EU, the European Commission’s machinery framework is what the technical file answers to, with the workflow on our machine safety compliance page.
What Drives the Cost and Lead Time of a Cobot Cell in Singapore?
We do not publish prices for a collaborative robot cell in Singapore, because two cells that look identical on a layout can differ by a wide margin on decisions taken before hardware is ordered. Where two such cells differ in price, the difference sits in those decisions far more often than in the arm.
| Cost driver | Why it moves the number |
|---|---|
| Collaborative mode selected | Separation monitoring buys sensing and floor area; force limiting buys measurement, tooling redesign and cycle time |
| End-of-arm tooling scope | One part geometry is a component-selection job. Several formats on one tool, or a quick-change rack, is a design project |
| Part presentation | A part arriving in a fixture is cheap. A part in a bin or an arbitrary pose is a vision and tooling problem first |
| Contact assessment difficulty | Sharp, hot, heavy or awkward workpieces push the cell toward guarding, which changes the whole architecture |
| Machine or instrument interfaces | Digital I/O or an Ethernet command set is cheap; a device with only an operator GUI costs real engineering |
| Mobile base | A moving collaborative space means per-station assessment, docking correction, and recovery logic that works with the vehicle anywhere on the floor |
| Safeguarding and validation scope | Scanner zones, stopping distance measurement, force and pressure records, and the documentation holding them |
Lead time runs 16 to 24 weeks from concept approval to factory acceptance testing on a standard build, and 24 to 32 weeks where cleanroom compatibility or full validation documentation applies. Design, assembly and testing happen at Woodlands Link, so a Singapore buyer attends the acceptance test rather than flying to it. That matters more on a collaborative cell than on most machines, because the acceptance test is where your operators find out whether the recovery behaviour is something they will live with or defeat by the third shift. How a build of this shape is scoped is covered in our guide to special purpose machine design.
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
Does buying a collaborative robot mean the cell needs no guarding?
No, and the robot datasheet cannot answer the question. Collaborative describes the application, which is the arm plus the tooling, the workpiece, the speeds, the fixture and the people, assessed together. Many production cells we build are hybrid: collaborative behaviour at the operator-facing station, and physical safeguarding where the hazard is the gripper, the part, or a trap point against a fixed surface. Fit a sharp tool, a hot part, or a payload heavy enough that any contact is unacceptable, and the application stops being collaborative even though the arm has not changed. The risk assessment decides the guarding, not the purchase order.
How are cobot force and pressure limits actually verified?
By measurement on the built cell, with the real tooling and the real workpiece, rather than from the arm's published limits. Force and pressure are two separate results, because pressure is force divided by contact area, so a small radius on a gripper finger or a part edge can fail a pressure limit at a force that passes comfortably. Contact type matters as well: the standards separate transient contact, where the body can move away, from quasi-static contact against a fixture or a wall, and the quasi-static case is the more restrictive of the two. Those records, plus scanner zone verification and stopping distance measurements, are the validation evidence for the cell.
Can a collaborative robot be mounted on a mobile robot base?
Yes, and we have delivered it: a QA laboratory system where an autonomous mobile robot carries a collaborative arm between sample racks and several test instruments, re-ordered in four consecutive years. What changes is the safety case, because the collaborative space moves with the vehicle. The assessment stops being one fixed floor area and becomes the vehicle in transit plus every station it docks at, each with its own surroundings and its own bystanders. Positioning changes too. Navigation lands the base within about plus or minus 50 mm, so the arm needs a vision correction step before it commits to a pick.