Before the Robot, Redesign the Work

Before making the painter robotic, make the work robot-ready.

A robot spraying a wall makes progress visible. The movement is controlled, the coverage looks consistent, and a demanding physical task is being performed by a machine.

What the demonstration rarely shows is how that wall became available.

Were the patches corrected? Was the surface suitable for coating? Had the electrician finished? Who cleared the floor, protected the openings and brought the equipment upstairs? Where will the machine work next?

Robotic painting is technically achievable. A 2022 field study demonstrated autonomous interior wall painting using a mobile platform, robotic arm, wall reconstruction and coverage planning. The more important question now is whether the construction production system can continuously present a machine with enough suitable, predictable and economically usable work.

That changes where we begin.

That is the central proposition: design the operation around automation, with consequences for which machines we build, where we place intelligence and how we measure value.

In a factory, the product comes to the machine

Conventional industrial robotics rests on considerable engineering outside the robot.

Factories standardise component positions, fixtures, material flow, work sequence, quality tolerances and workstation conditions. They organise repeatability before asking a machine to repeat. The robot benefits from choices already made about the work.

Often, the product comes to the machine.

On a construction site, the machine has to travel through the product.

That product is spread across rooms, floors and buildings. It is incomplete, changing daily and shared with other trades. A wall shown as a clean plane on a drawing becomes an actual surface with openings, imperfect geometry, patches, moisture conditions and access restrictions. Two neighbouring rooms can be at different stages of readiness.

The robot must reach suitable work through that changing environment. Doors, lifts, floor conditions, hoses, stored materials and other teams become part of its production problem. A large wall area on a plan can turn into several small, interrupted work areas on site.

Painting adds another layer of dependency. It involves inspection, preparation, filling, sanding, cleaning, protection, coating and verification. Different substrates and specifications require different sequences. Drying and curing conditions govern when subsequent operations can begin.

A late intervention by another trade can reopen finished work.

The conventional factory has concentrated much of this coordination around a production line. Construction finishing must recreate it across dispersed locations. Off-site production changes the arrangement for some building components; finishing inside completed structures still faces the movement problem.

Construction robotics therefore includes the design of a moving production system: how to prepare work, release it, reach it, execute it and hand it over.

The painter is carrying more than a tool

An experienced painter absorbs a remarkable amount of variability.

The worker notices a bad patch, changes the angle around a socket, adjusts pressure, shifts a ladder, changes tools and works around an obstruction. They judge whether another pass is needed, identify an edge problem, respond to a different substrate and speak to another trade when access is blocked.

Much of this happens without a separately written instruction for each exception. The painter combines physical effort, perception, movement, judgement and coordination.

Replacing that person completely means reproducing enough of those capabilities—or arranging the work so that fewer of them are needed during execution.

This creates an important design question: why begin by building a machine that reproduces the whole painter when a particular operation needs only part of the painter’s capability?

A sanding pass does not necessarily require the same equipment as material handling. Paint application does not require every judgement involved in substrate acceptance. A machine can take over physical effort while a person retains positioning and exception handling.

Human adaptability provides a useful baseline for design. The task is to decide which capabilities to mechanise, which to automate and which conditions to resolve elsewhere in the production system.

Make the work readable before asking it to be executable

“Robot-ready” needs an operational meaning.

A ready surface must satisfy defined entry conditions: appropriate preparation, completed predecessor work, suitable materials and environmental conditions, accessible geometry and a protected work area. The assignment must specify the finish, exclusions, acceptance criteria and response to exceptions.

A dashboard marking a room green does not make the room ready. The physical conditions and their representation have to agree.

This can mean standardising surface handovers, clearing repeatable access routes, grouping compatible areas into longer work runs and setting explicit rules for edges and openings. A robot-ready workfront is a designed operating condition, not merely a room with a surface left to paint.

The architecture can be understood through four capabilities:

Machine-readable. The system can reliably represent the physical environment and current work status: surfaces, openings, obstructions, completed operations and relevant conditions.

Machine-plannable. It can determine what should happen, where, in what sequence and with which resources, using readiness, dependencies and access windows.

Machine-verifiable. It has a defined way to establish whether the required condition or quality has been achieved. Acceptance is designed before delegation; actual results are measured after execution.

Machine-executable. Machines increasingly carry out the defined operation, with feedback and a clear route for situations outside their operating limits.

Together these form a production loop. They are capabilities to develop together, rather than a demand to digitise an entire site before testing a robot.

The order nevertheless matters. Asking a machine to execute while the work is poorly represented, poorly sequenced or ambiguously accepted pushes those unresolved problems into the robot.

A geometric scan can locate walls, openings and surface deviations. It cannot, by itself, establish moisture, adhesion, cleanliness or every aspect of coating suitability. Readiness combines appropriate measurements, inspection and process records; a model of intended geometry is only one input.

The central engineering proposition is this: smarter execution systems can allow simpler machines to do valuable work.

Structuring the work shifts part of the intelligence into scanning, digital representation, readiness rules, planning and quality verification. The execution machine receives a better-defined assignment. Whether this lowers total cost must be measured, but it changes the architecture we should compare.

Some site variability demands better robotics. Some demands better organisation. Treating both as problems for the robot alone is an expensive starting assumption.

A four-times-faster task is not a four-times-faster project

The useful unit of painting output is a square metre of the specified finish, accepted at the required quality, with the necessary preparation and protection included.

A square metre sprayed measures an activity. It does not necessarily measure completed production.

NIST’s construction-productivity research distinguishes task, project and industry measures; gains at one level cannot simply be transferred to another.

Consider a deliberately simplified example. A sequence takes five working days, of which one is paint application. Make application four times faster and the sequence falls to four and a quarter days, assuming the other activities stay unchanged.

The task is four times faster. The sequence is 15 per cent shorter.

This is arithmetic, not a painting benchmark. On actual projects, parallel operations and the critical path determine how much time reaches the delivery date. The practical question is where the saving goes.

Does the improvement release another activity sooner? Can the crew move directly to ready work? Does it reduce overtime or rework? Or does it produce a faster machine waiting for its next assignment?

Stanford researchers’ robot-evaluation framework includes the product, organisation and process alongside safety, quality, schedule and cost. That is the right scope: measure the operation changed by the machine, including supporting workers and neighbouring activities.

Remove effort progressively

There is a useful progression between largely manual work and autonomous execution:

Manual → Better workflow → Skill + tools → Mechanisation → Machine-assisted work → Selective automation

In manual work, the person supplies energy, movement and judgement. Better workflow removes waiting, unnecessary movement and sequencing losses. Training, specialisation and better tools increase capability and consistency.

Mechanisation transfers physical energy to equipment: airless spray, power sanding, pumps, mixers and extraction. The worker retains control of placement, movement and judgement. In machine-assisted work, equipment performs more of the repetitive operation while a person positions it, supervises, handles exceptions and verifies results.

Automation extends responsibility into sensing, planning, execution and feedback. Research is already combining these functions: a 2024 Puttybot study used force feedback and vision-based defect identification in an autonomous plastering trial.

These are not compulsory stages for every operation. A sufficiently bounded, repetitive task can justify automation early. The progression explains an industry-wide path through which human effort can fall without every investment requiring full autonomy.

The objective throughout is more accepted output from each human hour. Where technically and economically justified, it becomes fewer human hours for the same specified physical output.

The first important machine might not paint

Spraying attracts attention because it makes automation visible. The operation limiting throughput can sit elsewhere.

Sanding is a compelling candidate. It is repetitive, tiring, dusty and quality-sensitive. A worker paired with a specialised sanding machine can transfer substantial physical effort to equipment while retaining control over contact, positioning and local corrections. Full autonomous navigation need not be solved to make that arrangement useful.

Dust extraction also has demonstrated value. NIOSH’s drywall-sanding research showed reductions in exposure from vacuum controls. Mechanisation still requires appropriate safety design: powered spraying, for example, introduces hazards addressed in HSE guidance.

Other constraints point to different machines: pumps and mixers for material supply, handling equipment for movement, specialised platforms for access, or scanners and inspection tools for understanding surfaces.

A calibrated measurement system can support assessment of geometry, waviness, flatness, openings and defects. Reliable information about completed work and readiness can improve assignment, sequencing and correction across several crews.

This creates another proposition: a machine that understands the work can create more initial system value than one that executes a single task. Its advantage would come from improving several downstream decisions. The proposition is strongest where uncertainty, premature release or late defect discovery is the actual constraint.

Specialisation follows the same logic. Manufacturing offers a useful precedent in production lines combining fixtures, conveyors and task-specific equipment. Automation is built around operations, without requiring each machine to reproduce a complete worker.

Construction can similarly combine sanding, application, inspection and handling systems. The economic hypothesis is that bounded tasks and coordinated specialised machines will often be a better starting point than a machine carrying the whole painter’s capability.

Several machines also bring support, changeover, transport and maintenance costs. Compare complete operating arrangements, rather than judging simplicity by the appearance of one device.

The economics live in the inactive minutes too

Instantaneous speed describes what happens while a machine is executing. Deployment economics include the rest of its time.

Mobilisation, positioning, calibration, masking, refilling, cleaning and movement between rooms and floors all consume capacity. So do incomplete surfaces, blocked access, other trades and fragmented assignments.

A perfectly functioning robot can stand idle because its production system has nothing suitable to give it.

The economics of construction robotics are shaped by the minutes when the robot is not executing as well as the minutes when it is. Speed matters greatly when continuous robot-ready work exists. Utilisation becomes decisive when that work is fragmented.

Ready work is therefore a productive asset. Its value includes its area, condition, location and time window. A room inaccessible until tomorrow cannot fill today’s gap in a machine’s workload.

These considerations are longstanding. Robotic interior-painting research published in 1999 included system preparation, travel, manoeuvring, replenishment, cleaning and floor transport in its analysis of the operation.

Necessary inspection and maintenance belong in the calculation too. The objective is accepted output at a sustainable total cost, not keeping equipment visibly busy.

Count labour across the deployed team, including support and rework. Include equipment, materials, mobilisation and quality losses. If the machine improves consistency, credit the avoided correction. If several people are needed to keep it supplied and positioned, count their hours.

A study of ten construction robots across twelve projects found mixed cost outcomes: six cases reduced cost and four increased it. The heterogeneous sample is not an Indian painting benchmark. It illustrates the distinction between a machine performing successfully and its deployment producing an economic gain.

The robot becomes one part of the production system

The opportunity is to distribute intelligence across the operation: understand the site, release suitable work, allocate resources, execute, verify and use the results to improve the next assignment.

A robot becomes an execution node in that system. Its capabilities and the work around it develop together.

At Zobocon, we are building a system-led execution company, initially in painting. Our mental model is a distributed factory: the building stays in place while people, machines, process, quality and intelligence are digitally orchestrated across its workfronts.

This is a direction for development, not a claim of completed autonomy.

Technology will expand what can be delegated. Vision, sensing, navigation, learning and force control continue to advance. Machine costs can fall as designs and manufacturing mature. In 2026, Shimizu reported trials of robotic arms learning painting movements through imitation. Trial results and equipment costs still need commercial evaluation, but the operating possibilities are changing.

Robots will almost certainly play a growing role in construction painting. The more interesting question is what has to change in painting for them to create their maximum economic value.

The near-term transition is Painter × Machine × Process × Data: capability, physical power, organised work and reliable information reinforcing one another. As more operations become measurable and repeatable, Machine × Intelligence × Automation takes responsibility for more of the execution.

The production-system work is itself part of building automation. It creates the conditions in which autonomous machines can be useful.

The painting robot begins before the robot arrives—with work that can be understood, planned, verified and continuously supplied.

Before making the painter robotic, make the work robot-ready.

That principle leads to a more practical question: if you are actually evaluating a painting robot, what should you establish before committing capital?

I have developed a companion checklist: [Before You Buy the Painting Robot: 20 Questions Every Decision-Maker Should Ask]. >>>> Link to the 20 Questions

Research Links

[1] Huang, Chapman and Butry; NIST SP 1101 (2009). Metrics and Tools for Measuring Construction Productivity: Technical and Empirical Considerations.

[2] Brosque and Fischer; Construction Robotics (2022). A robot evaluation framework comparing on-site robots with traditional construction methods.

[3] Zhou and colleagues; Journal of Field Robotics (2022). Author's publication page; publisher record, DOI 10.1002/rob.22103.

[4] NIOSH and UK Health and Safety Executive. Control of Drywall Sanding Dust Exposures; Construction solvents: Spraying.

[5] Kahane and Rosenfeld; 16th ISARC (1999). Improved Tools for Robotic Interior Painting.

[6] Brosque and Fischer; Construction Robotics (2022). Safety, quality, schedule, and cost impacts of ten construction robots.

[7] Shimizu Corporation, 8 July 2026. Shimizu Corporation Launches Full-Scale Development of AI Robots.

[8] Liu and colleagues; Journal of Field Robotics (2024). Puttybot: A sensorized robot for autonomous putty plastering, DOI 10.1002/rob.22351.