Collaborative Robot Implementation 2026: Cost, ISO 10218 and Cobot ROI

In plants across Thailand and Vietnam, two problems keep landing on the same desk: headcount that cannot be recruited or retained, and a product mix that changes too often to justify a dedicated machine. Collaborative robot implementation is the option most production engineering teams now look at first, because it promises a small footprint, fast redeployment and no cage. That last promise is where projects go wrong. When ISO 10218-1:2025 and ISO 10218-2:2025 took effect on 1 April 2025, the unit of safety moved explicitly from the robot to the collaborative application, which means the robot, the task, the gripper and the surrounding environment assessed together. A robot arm sold as collaborative is not, on its own, a safe installation.

This article is written for plant managers, production engineering leads and regional IT teams responsible for sites in more than one ASEAN country. It covers the five layers that make up the real cost, how to select the first process, how to calculate payback without flattering the numbers, and the practical points that differ between a Thai site, a Vietnamese site and whatever your group standard assumes. All statistics are labelled by region, because a figure from Japan, Thailand, Vietnam or a global dataset will lead you to a different decision, and mixing them is the fastest way to lose credibility with your parent company.

Three reasons collaborative robot implementation became a realistic option in ASEAN plants

Interest in cobots looks like a sudden trend. Underneath it are three structural changes in labour cost, labour supply and investment policy. None of them is a single-year phenomenon; all of them play out over five to ten years, which is roughly the service life you should be planning around.

Reason 1: labour is still inexpensive in Thailand, and that is not an argument against automation

Thailand-specific: the statutory minimum wage was still in the range of 337 to 400 baht per day as of July 2026. It is set province by province, so there is no single national figure (source: Bangkok Shuho). Even during a period of rising consumer prices, no large revision was made.

The consequence matters for how you frame the business case. The argument used most often in Japan-specific investment proposals, that wages have risen so automation is now cheaper than people, does not transfer to Thailand unchanged. If you write a capital request for a Thai site using the wage-inflation argument, the ROI section will collapse under scrutiny, and it will collapse in front of your regional HQ rather than in front of you.

In Thailand the defensible axes are different: can you actually recruit and retain the headcount the line needs, can you hold quality variation down, and can you run the line during hours when nobody wants to work. Those three questions survive a review by group finance. A simple wage-differential argument does not.

Reason 2: the labour force itself is shrinking

Thailand-specific: NESDC, the National Economic and Social Development Council, published an assessment as of March 2024 projecting that Thailand’s labour force will decline by more than 3 million people per decade. Against a projected labour demand of 44.71 million in 2037, supply is expected to fall short. The same assessment indicated that automation could lift productivity by around 5 per cent and reduce labour demand by more than 2 million. It also noted that automation adoption in Thai manufacturing stands at only about 5 per cent (source: Bangkok Shuho).

Read those two numbers together and the planning premise becomes clear: you can still hire today, and you will not be able to in ten years. Equipment fixed to a single process, such as a purpose-built machine, is a poor hedge against that. A cobot that can be moved to a different process when the product mix changes is a better one, which is exactly why the technology gets a hearing at plants that were previously indifferent to robotics.

Reason 3: investment policy now points at automation

Thailand-specific: in the first half of 2026, applications to the Thailand Board of Investment (BOI) totalled 1,299 projects worth THB 1.47 trillion (USD 43.6 billion), up 37 per cent year on year. Within that, the Smart and Sustainable Industry category accounted for 132 projects worth approximately THB 17.2 billion (approximately USD 507.6 million) (source: Thailand Business News).

Two cautions before anyone quotes that in a proposal. First, those are application statistics: they describe how much investment was applied for, not what incentive any individual project receives. Second, the incentive structure differs by category, and the automation-related conditions are specific. We return to the actual conditions in the Thailand chapter below, because this is one of the most frequently misquoted points in the whole subject.

Global context: robot demand has doubled in ten years

Global: according to the IFR World Robotics 2025 report, new installations of industrial robots reached 542,000 units in 2024, the fourth consecutive year above 500,000 units. Asia accounted for 74 per cent of installations, and the worldwide operational stock reached 4.664 million units, up 9 per cent year on year. Demand is roughly double the level of ten years earlier.

Against that backdrop, ASEAN plants have moved past the question of whether to automate and into the question of where to start. If you are building a plant-wide roadmap rather than evaluating a single cell, our overview of factory automation planning for Thai plants sets out the sequencing logic that should sit above any individual cobot decision.

Why do more plants start with a collaborative robot instead of a fenced industrial robot?

Conventional industrial robots are fast, rigid and long-lived, but as a rule they require a cell enclosed by a safety fence. When you retrofit one into an existing plant, the space for that fence is the first obstacle you hit.

In our own project experience, floor area is the single most common constraint on retrofit work. The cell dimensions including fencing simply do not fit, the layout change requires a line stoppage, and that alone pushes the investment decision back by six months. A cobot may, depending on the outcome of the risk assessment, allow you to reduce fencing, which lowers that barrier.

The second reason is high-mix, low-volume production. Where a line changes over dozens of times a month, a dedicated machine never reaches the utilisation that justifies it. A robot that changes behaviour with a program switch can deliver better capital efficiency even at a higher unit price.

Note the wording carefully: a cobot may allow you to reduce fencing. That is not the same statement as a cobot needs no fence. The gap between those two sentences is the subject of the next two chapters, and in commercial terms it is where unbudgeted scope appears — material rather than incidental, and always discovered after the quotation has been approved.

What is a collaborative robot? Cobot vs industrial robot and the safety fence myth

What a collaborative robot actually is

A collaborative robot is a type of industrial robot designed to operate in the same space as people. Its joints typically incorporate torque sensing or current monitoring so that unintended contact is detected and motion stops, and so that force and speed can be limited. Payloads generally run from a few kilograms to around 20 kg, and operating speeds are set lower than those of conventional industrial robots.

The critical point is that these are functions, not safety in themselves. Fit a blade to a robot with contact detection and the severity of a contact event rises immediately. What the specification sheet describes is the performance of the robot as a component, in isolation from your task, your gripper and your operators.

Cobot vs industrial robot: the comparison that matters on the shop floor

Ignore the marketing comparisons and look only at the attributes that change a decision.

You will see cobots described as the affordable option for SMEs. Compared arm to arm that is not reliably true. The economics of a cobot for SMEs appear only when you evaluate the total cost including fencing, layout change and teaching effort, and they can disappear again if the application needs a scanner, a light curtain and a rebuilt conveyor feed.

Why is “a collaborative robot needs no safety fence” wrong?

The conclusion first. No standard anywhere states that a fence can be omitted because the robot is a collaborative model. What determines whether guarding is required is not the robot type but a risk assessment of the whole application: what the robot is holding, at what speed, in what location, next to whom, performing what task.

That logic was already implicit in earlier practice. The April 2025 revision put it into the structure of the standards themselves, which is why it is now much harder to argue with a supplier who has quoted on a fence-free assumption. The next chapter sets out exactly what changed.

What changed in ISO 10218:2025 and why “collaborative application” is the operative term

ISO 10218-1:2025 and ISO 10218-2:2025 took effect on 1 April 2025

ISO 10218, the international standard for industrial robot safety, comes in two parts: part 1 covers the robot itself and part 2 covers the robot system and integration. Revised editions of both took effect on 1 April 2025. This was the first full revision since the 2011 editions, so more than a decade of accumulated practice was folded in at once (source: EVS International).

The ISO/TS 15066 collaborative requirements moved into the main standards

Until then, the requirements for collaborative operation sat in a separate technical specification, ISO/TS 15066:2016. That document carried the power and force limiting framework and the approach to assessing contact with different parts of the body, and in day-to-day practice people simply said “TS 15066” when they meant cobot safety.

In the 2025 editions, those collaborative application requirements were incorporated into the bodies of ISO 10218-1 and ISO 10218-2, so TS 15066 is no longer referenced as a standalone document. Calling it withdrawn would overstate the position. What is accurate, and what matters for your purchase specification, is that design and procurement documents should now cite conformity with ISO 10218-1:2025 and ISO 10218-2:2025.

From “collaborative robot” to “collaborative application”

The change with the largest practical effect is terminological. The subject of the requirements moved from the collaborative robot to the collaborative application.

This is not a drafting preference. It carries three consequences that hit your budget directly.

1. The word “collaborative” in a catalogue is not a safety guarantee. A robot is not collaborative by itself. Only the combination of robot, task, tool and environment can be called a collaborative application. A risk assessment for the specific installation is therefore mandatory, and no supplier declaration about the arm can substitute for it.
2. Change the gripper and you have a different collaborative application. Swap a vacuum pad for an electric gripper on the same arm and both the contact geometry and the way force is transmitted change. Re-assessment is required. This is the clause most often missing from maintenance budgets.
3. Quoting on a fence-free assumption pushes cost into a later variation order. If the risk assessment concludes that you need a safety laser scanner, safety mats, measured force and pressure verification, or a layout change, those items were usually not in the original quotation. This connects straight into the cost structure in the next chapter.

The four modes of collaborative operation are essentially unchanged

The modes themselves did not change. The same four remain, in a new location within the documents. The A3 and Robotiq technical paper “ISO/TS 15066 Explained” remains a clear engineering explanation of how each mode behaves.

In practice you rarely pick one mode. A single station is usually designed with different modes applying to different phases of the cycle, for example power and force limiting during the pick and a monitored stop during manual replenishment.

What is new in the 2025 editions

• Cybersecurity requirements appear for the first time. The scope is deliberately limited to what can affect the robot’s safety functions. These clauses do not regulate plant network security in general, and your group IT team should not read them as an OT security framework.
• Functional safety requirements are stated more clearly, particularly how performance requirements for safety-related control systems are expressed.
• New robot classifications and test methods were added.
• Design requirements and operating mode requirements were expanded.

One point deserves emphasis because it is regularly misunderstood in vendor presentations. Force and pressure thresholds are not handed to you as a single universal number. They are determined by the risk assessment for the specific application. Do not let a design proceed on the premise that staying under one published figure equals safety.

Note also that national adoptions such as ANSI/RIA R15.06 and CSA Z434 align with the 2025 editions on their own timelines, which differ by country. If your project touches export specifications or a group engineering standard written elsewhere, confirm the position for each country involved.

Which standard applies in Thailand and Vietnam?

Here we deliberately avoid a categorical answer. Whether a given robot safety standard applies as mandatory law in Thailand or Vietnam can depend on industry, equipment classification and location, including the rules of the specific industrial estate. Confirm the position in writing with the competent authority and with local specialists.

That said, one approach has worked consistently in our project work: put the requirement in the purchase specification.

> As a deliverable of this contract, the supplier shall submit a risk assessment report prepared in accordance with ISO 10218-1:2025 and ISO 10218-2:2025, covering the target application, the anticipated contact locations on the body, the collaborative operation modes adopted, and the residual risks with corresponding measures.

Write that clause and integrators price the safety engineering into the quotation. Omit it, tell the integrator “it is a cobot so no fence is needed”, and the safety engineering falls out of the quotation and returns later as a variation order. The clause costs you nothing at tender stage and is the single highest-leverage sentence in the document.

Writing one specification for several countries

This is where multi-site groups lose time, and it is worth handling deliberately rather than discovering it at the second site.

If you operate plants in more than one ASEAN country, resist the urge to write a separate specification per site. Instead, split the document into two parts. The technical and safety core, meaning the ISO 10218-1/-2:2025 risk assessment deliverable, the collaborative operation modes, the gripper and workpiece scope, the teaching handover and the spare parts terms, is written once and applies to every site. A short country annex then carries only what genuinely varies: local statutory and industrial estate requirements, import and customs handling, the language of operator documentation, and the acceptance witnesses required by the local organisation.

Two practical benefits follow. Quotations from different countries become comparable, because every supplier is pricing the same technical core. And when the group later standardises on a model, you already have a document that a second site can issue without re-litigating the safety scope.

A related friction is worth naming. Regional HQ or the parent company frequently mandates a group safety document template, and those templates are often built around an older edition of the standard or around a national adoption such as ANSI/RIA R15.06. Suppliers then face a conflict between the template and the 2025 editions. The workable resolution is to require the ISO 10218-1/-2:2025 assessment as the technical deliverable and treat the group template as a reporting format that the assessment output is mapped into. Agree that mapping before the tender closes, not during acceptance, and put the mapping effort in the supplier’s scope so it appears in the quoted price.

Collaborative robot implementation cost: a five-layer breakdown

Break the cost into five layers and far less falls through. Be careful how this maps onto a capital request: the investment figure is the sum of layers 1 to 4, and layer 5 belongs in a separate line as annual running cost, not in the investment figure. Fold layer 5 into the investment total and you will subtract it again as annual running cost in the ROI calculation below, double-counting it. What the investment figure must never be is the arm price alone.

Layer 1: the robot unit

The arm, the controller and the teaching device, whether pendant or tablet. This is the easiest layer to compare on a catalogue basis, and it is also only a fraction of the total. Most disputes about cobot cost start with someone comparing layer 1 to a total.

Layer 2: end effector and peripherals

• Grippers: mechanical fingers, vacuum, custom jaws, screwdriving spindles
• Tool changers, where one arm handles several part numbers
• Vision: positioning, presence and absence checking, pose recognition
• Workpiece feed and discharge: chutes, stockers, conveyors, tray magazines
• Pedestals, frames and fixtures
• Safety devices: safety laser scanners, safety mats, light curtains, force and pressure measurement

In our experience the highest technical difficulty is never the arm. It is the gripper and the workpiece feed. Projects that defer these two items almost without exception slip their schedule.

Layer 3: system integration

Design, fabrication, installation, commissioning, robot teaching, risk assessment, and documentation covering operating manuals, work instructions and safety records.

Japan-specific: published estimates indicate that integration cost frequently lands between 50 and 150 per cent of the robot unit cost (source: Physical AI Hojokin Navi). On a unit costing JPY 3 million, that implies integration of JPY 1.5 million to JPY 4.5 million. This is the layer that breaks budgets built from a unit price quotation alone. The percentage will differ in Thailand and Vietnam, as discussed below, but the structural point that integration is comparable in magnitude to the arm holds everywhere.

Layer 4: site and utilities work

Electrical work covering voltage, capacity and earthing, compressed air piping, floor reinforcement and anchoring, layout changes, definition of safety zones, and resolving interference with existing equipment. On retrofits into an existing building this layer expands more often than any other, and it is the layer your local contractor, not your robot supplier, will quote.

Layer 5: hidden running cost

This is the layer most often absent from a capital request, and the one that determines whether the installation is still running in year three.

• Re-teaching effort every time the setup changes
• Rewriting work instructions, which recurs with every new part number
• Operator training, which in ASEAN plants means multiple languages
• Re-assessment of risk when the gripper changes, which as set out above is a new collaborative application
• Spare parts, maintenance contracts and consumables such as gripper jaws and vacuum pads
• Maintaining the manual fallback procedure, which is the cost of keeping the ability to revert to people when the robot is down

Layer 5 is not capital expenditure; it is an annual operating cost. If you do not subtract it in the ROI calculation, your payback period will look shorter than it is.

Japan-specific reference ranges

The figures below are Japan-specific estimates published by the cited source. They cannot be applied directly to a Thai or Vietnamese quotation, and they are shown only to convey the relative weight of each layer. We keep them in JPY because the source provides no exchange rate basis for conversion (source: Physical AI Hojokin Navi).

Total-project examples from the same Japan-specific source:

These three examples cover different models, different unit counts and different target processes, so they cannot be read as a cost-per-unit comparison. Divide them out and the cost per unit actually rises, from approximately JPY 5.5 million to approximately JPY 7.75 million to approximately JPY 9.4 million, which reflects the different processes and gripper configurations rather than any effect of unit count. The reason a roll-out gets cheaper is not the number of units in itself but whether layer 3 integration cost can be compressed by reusing design work, which is covered later in this article. It is the single most useful thing to explain to a parent company that has judged cobots on a first-unit business case.

Where the cost structure differs when you buy in Thailand or Vietnam

Apply the Japan-specific proportions unchanged in ASEAN and you will be wrong in four places.

1. The labour cost ratio is different. Thailand-specific: with the minimum wage held in the range of 337 to 400 baht per day, the absolute value of the labour you can remove is smaller than in Japan. Identical equipment removing identical hours produces a longer payback in Thailand. Vietnam-specific: wage growth of 8 to 10 per cent per year has been reported, which changes the picture again, as covered in the Vietnam chapter.

2. Import duty and the BOI machine list. Most arms and peripherals are imported. The handling of duty and VAT, and whether the item appears on an approved machine list for a BOI-promoted company, change the effective acquisition cost. Import timing and procedural lead time then drive the project schedule rather than following it. We covered the machine list and import mechanics in the context of material handling equipment in our article on AGV and AMR implementation; if you are evaluating robots and transport in the same budget cycle, read the two together.

3. Local integration rates. Integrator day rates in Thailand and Vietnam are generally below Japanese levels, but the number of firms that can carry a collaborative application through to a documented risk assessment is limited. Screen candidates on whether safety engineering is inside their scope, not on the day rate.

4. Commissioning travel from overseas. If a Japanese or European manufacturer or integrator performs commissioning, travel, accommodation and engineer rates are added. These items are frequently absent from a first quotation, so state in the specification how many engineers, for how many days, and at whose cost.

Collaborative robot price and supplier comparison: four types, not thirty model numbers

Japan-specific unit price ranges

The figures below are Japan-specific estimates from the cited source, kept in JPY for the same reason as above (source: Physical AI Hojokin Navi). Street prices in Thailand and Vietnam differ, and the total including local service terms differs more. Use this table only to understand relative positioning between families.

Choose the supplier type your constraints require

A list of model numbers does not help you decide. Sort suppliers into four types and let your own constraints select the type first. Selection converges much faster that way.

Type 1: global majors, chosen for service network and installed base. Appropriate when you run plants in several countries and want one standard model across the group. Distributors and service points exist across ASEAN and reference installations are easy to find. Unit prices are higher; you are buying continuity of parts supply and support.

Type 2: Japanese suppliers, chosen for documentation and an existing integrator network. Appropriate when you must align with a parent company equipment standard, when you want the same brand as your existing industrial robots, or when technical documentation and shop-floor training need to run in Japanese. Where a Japanese engineering department reviews the design, this choice produces the least friction.

Type 3: Taiwanese and Korean suppliers, chosen for integrated vision and price balance. Several families integrate vision into the robot platform, which can simplify a station that includes inspection or vision-guided positioning. A reasonable shortlist when you are balancing capability against price.

Type 4: Chinese suppliers, chosen for lowest initial cost. Genuinely attractive for a trial where minimising capital outlay is the objective. The risks are local service presence and continuity of parts supply. Before placing one in a critical process, get the service terms and parts availability confirmed in writing.

Four items that bite later if you buy on price alone

These are the four that surface after commissioning. Add them as columns in your quotation comparison sheet so that they are visible before the decision, not after.

1. Distance to the nearest service point. Response capability differs between greater Bangkok, the Eastern Seaboard around Chonburi and Rayong, the north around Lamphun, and northern versus southern Vietnam. Do not accept “we have local service”; ask how many hours from your gate.
2. Parts lead time. When the robot stops, how many days until the replacement part arrives, and is it held in country or ordered from the home market. For a critical process this determines availability directly.
3. The language of robot teaching. Which languages the teaching pendant interface supports. Placing a model with no Thai or Vietnamese interface into a plant where local engineers are the primary users raises training cost and concentrates knowledge in one person.
4. How many integrators can work on the model. If only one firm in country supports your chosen platform, you have lost both commercial leverage and your maintenance alternatives.

Which process should you automate first? Where collaborative robots fit and where they do not

Process selection sits alongside cost structure as the determinant of success. Our working rule is blunt: where the process fits, a somewhat rough design still produces a working station, and where the process does not fit, no amount of design refinement will rescue it.

Processes that fit

• Screwdriving automation, where torque control and missing-screw prevention are delivered together
• Machine tending cobot applications, loading and unloading machine tools
• Inspection support, presenting the part to a camera or reorienting it
• Cartoning and small-format palletising of light items
• Label application and sealant or adhesive dispensing
• Part of an assembly sequence, at the stations where positional accuracy can be held
• Loading and unloading test equipment
• Unattended running overnight, converting hours when nobody is present into productive hours

Processes that do not fit

• Short cycle times. A conventional industrial robot inside a fence is both cheaper in total and faster.
• Heavy parts. Exceeding the payload is obviously out, but running near the limit is also poor practice given speed limits and service life.
• High-severity tools that could genuinely contact a person. Blades, lasers and hot surfaces do not become less severe because contact is detected. Guarding or separation becomes the premise.
• Randomly oriented parts in bins. The cost of 3D vision plus gripper can exceed the arm itself, and the business case breaks.

Five decision axes

A ten-point readiness checklist

Use this to score a candidate process. Starting with a process that scores eight or more yes answers is the realistic approach.

1. Can you state the labour hours for the process from measured data over the last three months?
2. Can you convert the defect rate and rework hours for the process into money?
3. Do you record the number of changeovers and the time each one takes?
4. Can you express the variation in workpiece shape and dimensions numerically?
5. Is there a credible design for feeding and discharging parts without a person doing it?
6. Have you evaluated candidate gripping methods against physical samples of the part?
7. Can you show, on a drawing, the routes where people walk and work around the process?
8. Is the process free of high-severity tooling such as blades, heat sources or lasers?
9. Have you verified the power, air and floor conditions at the intended location?
10. Can you commit at least two internal people to be trained on robot teaching?

If the measured data behind items 1 to 3 does not exist, you cannot make an investment decision at all, only a guess. The mechanics of collecting labour hours, defects and cycle time as routine data are covered in our article on choosing a production management system for a Thai plant. If the process you are considering is an inspection station, the combination of robot handling with automated inspection is discussed in how to decide on AI visual inspection.

Practical issues when implementing a collaborative robot in a Thai plant

From here the content is Thailand-specific. The focus is on the points where a process imported unchanged from a parent company playbook breaks.

The wage figure is a daily rate, and treating it as hourly wrecks the business case

Thailand-specific: the minimum wage is stated as 337 to 400 baht per day, set by province, and it remained at that level as of July 2026. Read it as an hourly figure and your ROI calculation is wrong by an order of magnitude, in the direction that makes the investment look attractive.

To derive an hourly labour rate, divide the daily rate by the contracted daily hours. For example, a daily rate of 400 baht against 8 contracted hours gives approximately 50 baht per hour. Your actual fully loaded hourly cost is that figure plus social security contributions, bonus provision, meal allowance and transport, all of which vary by company. Use your own historical figures for the loading factor rather than a rule of thumb.

BOI corporate income tax exemption is 50 per cent as the base case, and 100 per cent is conditional

Thailand-specific: for the Smart and Sustainable Industry category, the corporate income tax exemption is 50 per cent as the base case. The 100 per cent level applies only where the company introduces automation or robotics into the production line and sources at least 30 per cent of the value of the upgraded machinery from Thailand’s domestic automation industry.

In other words, installing a cobot does not by itself produce a full tax exemption. The domestic sourcing condition is the operative constraint, and because it is a condition on sourcing, it can influence supplier selection. Decide early whether you intend to design for it, because retrofitting the sourcing ratio after you have placed orders is not possible.

Two qualifications must travel with those figures. First, the first-half 2026 application statistics cited earlier, 1,299 projects worth THB 1.47 trillion of which 132 projects worth approximately THB 17.2 billion sat in Smart and Sustainable Industry, describe application volume and are a separate matter from what incentive a project receives. Second, BOI incentive structures differ by category, so the conditions applicable to your category must be checked individually. Have the applicability of any incentive confirmed in writing by a tax adviser and a legal adviser before it appears in a capital request.

Machine list approval and import lead time

Where a BOI-promoted company imports robots and peripherals, machine list approval is a precondition. The point most often missed is that the scope is not only the arm: grippers, pedestals, safety devices and vision components can all fall within it.

That creates a chain of dependencies. The configuration must be frozen before the application can proceed; the application must proceed before the equipment can be imported; and if import slips, you miss the installation window. In our own project planning we set the configuration freeze date by working backwards from the installation window, and for robot projects we add margin beyond that because gripper prototyping sits on the critical path.

Design the teaching interface language and the work instruction language separately

In Thai plants, Thai, English and Japanese are often joined on the same line by operators from Myanmar, Cambodia and Laos. The practical response is to separate the language of the teaching pendant interface from the language of the work instructions.

• Teaching pendant interface: used by maintenance and production engineering. Standardise on English or Thai and train at least two people, ideally three.
• Work instructions and abnormality response procedures: used by every operator on the line. Build them around photographs and diagrams and provide the language versions you actually need. The less text there is, the more stable the operation.

Trying to force both onto a single language guarantees that one of them fails in use.

Power interruptions and voltage dips in the rainy season

Thailand-specific plants experience outages and momentary voltage dips during the rainy season. If power is lost mid-cycle, you need standardised procedures for homing the axes, for dealing with a part still held in the gripper, and for restoring synchronisation with peripheral equipment.

In our experience the recovery procedure must also specify who performs it. Without that, a stoppage at night or over a weekend simply waits until the next morning. Decide the scope of UPS coverage at design stage as well: controller only, or controller plus peripherals, because the answer changes the electrical design.

The installation window comes round only a few times a year

Thailand-specific: the periods when a line can be stopped are effectively limited to Songkran in April, the year-end and new year break, and the days around Lunar New Year. Miss the window and you wait months for the next one.

Investment approval, ordering, import, fabrication and installation therefore have to be scheduled backwards from the window. A plan that says work will begin once the budget is approved will, more often than not, miss it. Where your parent company runs a fixed capital approval calendar, raise this explicitly: an approval that arrives in May has lost the April window and is effectively an approval for the following year.

Vietnam plants: what changes in a market where installations are growing

Vietnam-specific: analysis based on IFR data indicates that new installations of industrial robots in Vietnam grew by approximately 27 per cent in 2025 (source: RMIT University Vietnam, 24 April 2026). The same article reports wage growth in the range of 8 to 10 per cent per year and identifies textiles, electronics and logistics as the sectors driving automation.

That is the central difference from Thailand. Thailand-specific minimum wages have been held flat, whereas Vietnam-specific wage growth can serve as a direct motive for automation. In a Vietnamese business case there is room to build future labour cost increases into the model, provided the assumed rate comes from your own payroll history and local HR information rather than from a published national figure applied uncritically.

Vietnam-specific, in logistics: the same source describes Viettel Post’s Hanoi warehouse using autonomous robots to handle 4 million parcels per day, equivalent to around 50 per cent of Vietnam’s e-commerce shipment volume. It also notes the accumulated precision manufacturing base built by companies such as Samsung, Intel and Amkor, while cautioning about reliance on an FDI-led model and raising the question of how much value beyond assembly is retained domestically in the robotics value chain.

For a foreign-invested plant, two disciplines are worth adopting.

1. Keep institutional questions in writing. Investment incentives, import procedures and any certification requirements relating to safety in Vietnam should be confirmed in writing with the competent authorities and local specialists. We deliberately do not cite decree numbers or preferential tax rates in this article. Requirements change, and their interpretation varies case by case; a written confirmation for your specific case is the only basis you should build a schedule on.

2. Confirm service and spares coverage before selecting a model. The distribution of manufacturer and distributor service points differs between the north around Hanoi and the south around Ho Chi Minh City. Verify at model selection stage whether on-site response is available from your intended location and whether parts are held in country.

There is also a scheduling point that mirrors Songkran and is easy to underestimate if you plan from a Thai calendar. Vietnam-specific: Tet compresses the practical installation window, and the effect extends beyond the public holiday itself, because local fabrication, transport and contractor availability tighten before it and staffing takes time to normalise afterwards. If a group plans a two-country roll-out around a single window, Thailand and Vietnam do not align, and treating them as one milestone is how the second site ends up commissioned by whoever is available rather than by the team that learned on the first one.

For groups operating in both countries, standardising on one model to share maintenance know-how and spare parts is often the right call. It holds only where both countries have viable service coverage for that model, so check coverage before you write the standard, not after.

How to run a collaborative robot implementation: six steps

Step 0: measure the current state

Measure labour hours, defect rate, cycle time and changeover frequency for each candidate process. Not perceived values, not standard times: actual data from the last several months. If this is vague, every subsequent ROI figure is an estimate dressed as a calculation.

The minimum measurement set is five items: operator working time, waiting time, changeover time and frequency, defect and rework hours, and process downtime with its causes.

Step 1: select one process

Using the decision axes and the checklist above, narrow the first target to a single process. Attempt two or three at once and you will face gripper design problems and part feed problems simultaneously, and commissioning will not converge.

Position the first unit as an investment in learning, with the explicit objective of leaving risk assessment, robot teaching and maintenance capability inside your organisation. That framing also gives you something honest to say when the first-unit payback is longer than the group hurdle rate.

Step 2: risk assessment, including the gripper

Conduct the risk assessment within the framework of ISO 10218-2:2025. The critical instruction is to assess the system with the end effector and the workpiece in place. As set out above, the subject of the standard is the collaborative application, not the arm.

Elements to include: the robot’s motion envelope, gripper geometry and gripping force, workpiece weight and shape including any sharp features, the movement routes of people nearby, the anticipated contact locations on the body, and behaviour under fault conditions.

This step fixes the collaborative operation modes you will use, whether safety-rated monitored stop, hand guiding, speed and separation monitoring or power and force limiting, and the safety devices you need. This is the point at which cost becomes real. Conversely, any quotation obtained before step 2 is an order-of-magnitude estimate, and should be labelled as such in your capital request.

Step 3: proof of concept with your own parts and gripper

A PoC without your workpiece and your gripper tells you nothing. Watching a standard demonstration part move in a vendor showroom does not establish whether your process works.

Four things must come out of a PoC as numbers: whether your actual part is gripped reliably, whether positioning holds across the full variation range, whether the cycle time target is met, and how many minutes the program change takes at changeover. A PoC that cannot produce those four numbers is faster to redo than to argue about.

Step 4: installation, commissioning and training

Install and commission within the agreed window. The deliverables at this stage are the risk assessment report, the operating manual, work instructions, abnormality response procedures, robot teaching procedures and the record of operator training.

In our experience, planning to deliver training in a single block on the final day of commissioning fails almost every time. Spread it across the commissioning period and train while the line is actually running; retention is far better and the questions that matter surface while the integrator is still on site.

Step 5: roll out, where integration cost falls

From the second unit onward, the risk assessment framework, pedestal design, gripper specification and teaching procedures can be reused, so layer 3 integration cost drops. The Japan-specific total examples above cover different models, configurations and processes, so they cannot be used as evidence of a falling cost per unit. What falls on a roll-out is layer 3, and only on the assumption of the same model, the same configuration and a similar process. Depart from that and the second unit carries the same design cost as the first.

If roll-out is the plan, standardise the same gripper, the same pedestal and the same program structure while you are still building the first unit. Engineer the first unit as a bespoke optimum for one process and nothing is reusable at the second, which removes the cost reduction entirely and, with it, the argument you were going to make to your parent company.

Load planning and changeover planning across a line with several robots also need to be handled at the planning layer rather than station by station. That perspective is covered in our comparison of production schedulers.

Seven clauses to write into every purchase specification

Include these seven items to make quotation comparison meaningful. Without them, each supplier prices a different assumption set and the quotations are not comparable.

1. The risk assessment report as a contractual deliverable, prepared to ISO 10218-1/-2:2025, with the scope explicitly including the gripper and the workpiece
2. The scope of the robot teaching handover, meaning what your team will be able to do unaided, with the number of people trained and the training hours stated
3. Service hours and SLA, covering the hours cover is available, the on-site response time and whether remote support is included
4. Parts lead time, item by item for major components, stating whether stock is held in country or ordered from the home market
5. The unit rate for re-teaching, expressed as cost and duration per additional part number
6. The cost of re-assessment when the gripper changes, quoted in advance rather than negotiated later
7. The language of training, specifying which languages and which documents are provided in each

Items 5 and 6 exist to make layer 5 visible at quotation stage. Sign a contract with those two left blank and unplanned costs arrive after handover, usually at the moment you add the second part number.

Cobot ROI: how to calculate payback and four traps

The basic formulae

Annual saving = (labour hours removed x fully loaded hourly rate) + (defect cost reduction) + (reduction in overtime and night shift premiums) − (annual running cost)

Payback period = total initial investment, being the sum of layers 1 to 4, divided by the annual saving

Annual running cost must include maintenance, electricity, consumables and re-teaching effort, which is layer 5.

Four traps

Trap 1: treating the daily wage as an hourly rate. Thailand-specific minimum wages are stated per day. Mistake 400 baht per day for an hourly figure and the labour rate is eight times too high, making the payback look like one eighth of reality. Always derive the rate as daily rate divided by contracted daily hours.

Trap 2: counting the entire labour content of the process as removed. Even after the robot is installed, people remain for setup, material replenishment, abnormality response and quality checks. Book the full process labour as a saving and the result will diverge sharply from actuals. In our experience, planning on 60 to 80 per cent of the nominal labour content is realistic, and the exact ratio should come from your PoC measurements because it varies by process.

Trap 3: omitting layer 5 from annual running cost. Re-teaching, training and re-assessment recur every year. Leave them out and the payback shortens on paper only.

Trap 4: judging the whole group on the first unit’s integration cost. The first unit carries all the design cost, all the commissioning cost and all the learning cost, so it is the least favourable case you will ever build. Conclude from a first-unit payback that cobots do not suit your operation and you have thrown away the roll-out effect. Present both cases side by side in the capital request: first unit, and steady state after roll-out.

A worked example, which is arithmetic and not a benchmark

Every figure below is an assumption used for illustration. None of it is a market rate or a benchmark. Substitute your own numbers.

Assumptions

Annual saving, itemised

Case 1: first unit, two shifts only

2,500,000 divided by 217,400 is approximately 11.5 years.

This is the number worth sitting with. At Thailand-specific wage levels, replacing minimum-wage work in a single process does not pay back on labour savings alone. That does not mean collaborative robot implementation fails in Thailand. It means that a business case resting only on labour reduction fails in Thailand, and that you need a different justification.

Case 2: adding unattended overnight running

Assume the same equipment runs unattended for a further three hours at night, allowing work previously subcontracted to be brought in house, and assume that saves 180,000 baht per year in subcontract charges.

Net saving = 217,400 + 180,000 = 397,400 baht per year

Payback = 2,500,000 divided by 397,400, approximately 6.3 years

Case 3: second unit, roll-out

Assume the second unit reuses the risk assessment framework, pedestal design, gripper specification and teaching procedures, holding initial investment to 1,800,000 baht, with savings equal to case 2.

1,800,000 divided by 397,400 is approximately 4.5 years.

Comparing the three cases shows what actually governs cobot ROI: whether you can add running hours, and whether you can roll out. Where the hourly labour rate is low, those two variables dominate the payback period and labour reduction alone does not.

One last point on inputs. The labour hours, defect and changeover data behind any of these figures are meaningless unless measured. Whether that data exists as routine output determines how quickly you can make the decision at all, which is usually a bigger constraint than the arithmetic.

Five failure patterns we keep seeing

All five are avoidable at design stage, and all five cost money after handover.

Failure 1: quoting on “it is a cobot so no fence is needed”, then adding safety devices later

The most common pattern by a wide margin. The budget is built fence-free, the risk assessment then calls for a safety laser scanner, safety mats, measured force and pressure verification and a layout change, and the additional cost arrives as a variation.

The countermeasure is specific. Make the risk assessment a premise of the quotation and a named deliverable in the specification. Making the collaborative application the subject of ISO 10218-1/-2:2025 was intended precisely to close this gap; use it.

Failure 2: leaving the end effector until later

Time goes into selecting the arm, and the gripper is left to the integrator to sort out. In reality the technical risk lives in the gripper and the part feed, not the arm.

Surface finish, rigidity, dimensional variation, orientation, contamination and oil all change whether gripping succeeds. Defer the gripper question to the PoC and that is where you discover it does not work, which sends you back to model selection with the schedule already spent.

Failure 3: people continue to feed and unload the parts

The robot performs the operation, but one person still places parts in front of it and another still removes finished items. No headcount is released. This is where projects stall at semi-automation and stay there.

Decide where the parts come from and where they go at process design stage. Leave feed and discharge equipment out of the quotation and this failure follows automatically.

Failure 4: only one person can perform robot teaching

The moment that individual takes leave, resigns or is moved to support another line, changeovers stop. The flexibility advantage of a cobot is real, but it is delivered by people, not by the robot.

Train a minimum of two people, preferably three. That is the reason for stating the number of trainees in the purchase specification rather than leaving it to the integrator’s discretion.

Failure 5: designing an ASEAN site with an imported cost structure and an imported safety culture

Apply Japan-specific or European cost proportions unchanged and you will overstate the labour saving. Import a safety philosophy without adapting how it is explained and it will not match local working practice, at which point it gets worked around on the night shift.

What has worked for us is not stopping at installing the safety device. It is explaining why that stop exists, in the language of the people who work next to it. A protective device whose reason has not been communicated will be defeated sooner or later, and no amount of documentation filed with regional HQ changes that.

Frequently asked questions about collaborative robot implementation

What is a collaborative robot, and how does a cobot differ from an industrial robot?

A collaborative robot is a type of industrial robot designed to work in the same space as people, with functions for limiting force and speed and for detecting contact. The main differences are that safety fencing may be reduced depending on the risk assessment outcome, and that the footprint is generally smaller. In exchange, operating speeds are lower, so high-throughput processes remain better served by a fenced industrial robot.

How much does collaborative robot implementation cost?

Evaluate the sum of five layers: the arm, peripherals including the end effector, system integration, site and utilities work, and the hidden running cost after handover. Japan-specific estimates put the arm at 30 to 50 per cent of the total, integration at 20 to 40 per cent and peripherals at 10 to 30 per cent, with integration frequently landing between 50 and 150 per cent of the arm cost (source: Physical AI Hojokin Navi). In Thailand and Vietnam the proportions shift because of import duty handling, BOI machine list treatment, local integration rates and overseas commissioning travel.

Does a collaborative robot really need no safety fence?

No standard states that fencing can be omitted because the robot is a collaborative model. In ISO 10218-1:2025 and ISO 10218-2:2025, effective 1 April 2025, the collaborative requirements previously in ISO/TS 15066:2016 were incorporated into the main standards and the operative term moved from collaborative robot to collaborative application. Whether guarding is needed is determined by a risk assessment of the application as a whole, including the robot, the task, the gripper and the environment. Force and pressure thresholds are likewise not given as a single universal figure.

How do you calculate cobot ROI?

The basic formulae are: annual saving = (labour hours removed x fully loaded hourly rate) + defect cost reduction + reduction in overtime and night premiums − annual running cost, and payback = total initial investment divided by annual saving. In Thailand, derive the hourly rate as daily rate divided by contracted daily hours; for example a daily rate of 400 baht over 8 contracted hours is approximately 50 baht per hour before on-costs. Three further disciplines matter: do not count the whole labour content of the process as removed, do subtract layer 5 running costs, and do evaluate the first unit and the post-roll-out steady state separately.

How should you choose a robot system integrator in Thailand?

Integration capability for robots exists in Thailand, but the number of firms that can carry a collaborative application through to a documented risk assessment is limited. Screen on capability rather than price: whether they can deliver a risk assessment report to ISO 10218-1/-2:2025 as a contractual deliverable, whether they have installed the model you have shortlisted, whether they have gripper design experience, what their on-site response time and parts lead times are, and which languages they can train in. Asking for a redacted sample risk assessment report from a previous project is a fast and revealing screening question.

Can an SME or a single process justify a collaborative robot?

Technically yes, and starting with a single process is what we recommend. Be clear, though, that the first unit carries the design and learning cost in full, so its payback will be the longest you ever see. Frame the first unit as an investment in retaining risk assessment, robot teaching and maintenance capability in house, and plan to earn the return from the second unit onward. A cobot for SMEs works on that sequencing; it rarely works as a one-off purchase judged on its own payback.

Can our own team learn to do the robot teaching?

Yes, but only if the scope of the teaching handover is written into the specification at order stage. With nothing written, the contract ends when the integrator has completed commissioning, and every new part number then becomes a chargeable visit. Specify the number of people to be trained, which should be at least two, the training hours, the language of the materials, and the unit rate for re-teaching per additional part number, and require all of it in the quotation rather than after handover.

Conclusion: collaborative robot implementation is decided by process selection and risk assessment

The key points, in the order they affect your decision.

1. Labour reduction alone is not a sufficient decision axis. Thailand-specific minimum wages have been held in the range of 337 to 400 baht per day, so removing labour from a single process produces a long payback. What determines cobot ROI is whether you can add running hours and whether you can roll out. Vietnam-specific conditions differ, with wage growth reported at 8 to 10 per cent per year, so the same business case template does not transfer between the two countries.

2. Since April 2025, the unit of safety is the collaborative application. The word collaborative in a catalogue is not a safety guarantee. Change the gripper and you need to re-assess. A quotation built on a fence-free assumption generates safety device costs later.

3. Look at cost in five layers: the arm, peripherals, system integration, site and utilities work, and the hidden running cost after handover. Japan-specific estimates put integration at 50 to 150 per cent of the arm cost, so a budget built from a unit price is structurally short.

4. Process selection decides the outcome. Use the five axes of cycle time margin, workpiece variation, changeover frequency, floor area and safety risk to narrow to one process, then verify it in a PoC using your own part and your own gripper.

5. Clear the country-specific issues first: how the hourly labour rate is derived, the BOI conditions where 50 per cent is the base case and 100 per cent requires the domestic sourcing condition, machine list approval, multilingual teaching and documentation, rainy season power quality, and installation windows that do not align between Thailand and Vietnam. Have every institutional question confirmed in writing by the competent authority and local specialists.

Installed in the right process with a proper risk assessment behind it, a collaborative robot becomes equipment you can keep using through years of product mix change. Start from model selection and leave the process and the gripper until later, and both cost and schedule expand. This is a field where sequence determines the result.

Before you choose a model, the most valuable work is establishing which process, what the risks are, and what the total comes to across all five layers. TOMAS TECH works with foreign-invested manufacturers in Thailand and Vietnam on exactly that stage: reading the measured process data, selecting the first process, and drafting purchase specifications that include the risk assessment as a deliverable. An exploratory conversation is entirely welcome, whether the question is whether a particular process is a sensible candidate or whether a quotation you have received looks reasonable to a third party. Tell us about the process as it runs today, via our contact form.

References

1. IFR (International Federation of Robotics), “World Robotics 2025” press release — global: 542,000 new industrial robot installations in 2024, Asia 74 per cent, operational stock 4.664 million units

https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years

1. RMIT University Vietnam, “Opportunities for Vietnam in the robotics value chain” (24 April 2026) — Vietnam-specific: approximately 27 per cent growth in new installations in 2025, wage growth of 8 to 10 per cent per year, Viettel Post handling 4 million parcels per day

https://www.rmit.edu.vn/news/all-news/2026/apr/opportunities-for-vietnam-in-the-robotics-value-chain

1. EVS International, “Collaborative Robot Safety Standards 2026: ISO 10218:2025 and TS 15066” — effective 1 April 2025, and the shift in terminology to collaborative application

https://www.evsint.com/collaborative-robot-safety-standards-2026-iso-10218-2025-ts-15066/

1. A3 / Robotiq, “ISO/TS 15066 Explained” — technical explanation of the four modes of collaborative operation

https://www.automate.org/robotics/tech-papers/iso-ts-15066-explained

1. Thailand Business News, “Thailand FDI Surges 37% to $43.6B in H1 2026” — Thailand-specific: BOI applications in H1 2026 of 1,299 projects worth USD 43.6 billion, including 132 Smart and Sustainable Industry projects worth approximately USD 507.6 million

https://www.thailand-business-news.com/investment/321188-thailand-fdi-surges-37-to-43-6b-in-h1-2026-led-by-ai-data-centers

1. Bangkok Shuho, “Thailand’s minimum wage held at 337 to 400 baht per day despite rising prices” (in Japanese) — Thailand-specific

https://bangkokshuho.com/thainews-948/

1. Bangkok Shuho, “Government think tank warns of serious labour force shortage in Thailand by 2037” (in Japanese; NESDC, 27 March 2024) — Thailand-specific

https://bangkokshuho.com/thaieconomy-87/

1. Physical AI Hojokin Navi, “Collaborative robot quotation comparison and cost breakdown” (in Japanese) — Japan-specific market estimates

https://physical-ai-hojokin.jp/articles/kyoudou-robot-mitsumori-hikaku/

1. Mordor Intelligence, “Collaborative Robots Market” — global market growth

https://www.mordorintelligence.com/industry-reports/collaborative-robot-market

1. Kanto Bureau of Economy, Trade and Industry, “Robot introduction policy package, May 2026 edition” (in Japanese) — Japan-specific support schemes

https://www.kanto.meti.go.jp/seisaku/iot_robot/robot/data/robot_package.pdf

Note: amounts and ratios in this article are published or estimated values from the cited sources, labelled as Japan-specific, Thailand-specific, Vietnam-specific or global. Japanese domestic figures are kept in JPY because the sources provide no exchange rate basis for conversion. Worked calculations are arithmetic based on stated assumptions, not benchmarks. The applicability of any regulation, incentive or standard should be confirmed in writing with the competent authority and with local specialists.