Shimizu and Sony Are Testing AI Robots on Construction Sites: Could Site Inspections and Painting Be Automated Next?

Shimizu’s latest trials suggest that construction automation will advance first through bounded tasks: repeatable patrol routes, image capture, progress checks and painting across controlled surfaces. In Sydney and NSW, robots could reduce repetitive labour and improve evidence collection, but they will not replace principal certifiers, trade judgement or site safety duties. The practical issue is whether projects can redesign access, exclusion zones, data handling, hold points and contractor responsibility around robotic work.

Construction robotics has spent years appearing more convincing in demonstrations than on live projects.

Factories provide fixed workstations, consistent lighting, controlled surfaces and repeatable movements. Construction sites provide almost the opposite: changing access routes, open edges, temporary services, unfinished floors, moving workers, variable lighting and work areas that may look different every morning.

That makes the latest program from Shimizu Corporation significant.

The Japanese contractor is conducting field trials involving humanoid robots for site inspections and robotic arms for painting. Its inspection trial involves a robot navigating a designated route with a camera, while captured footage is being considered for analysis through multimodal artificial intelligence systems.

The development is connected to, but should not be confused with, a single Sony-built system performing every task.

Shimizu says its wider program includes technical cooperation from partners such as Sony. Shimizu and Sony had already begun joint construction-site mobility experiments in 2021, testing whether a six-legged wheeled robot could negotiate level changes, openings, narrow passages and other site obstacles while supporting patrol and monitoring work.

The important development is therefore not a science-fiction machine arriving to build an entire tower.

It is the gradual conversion of specific construction activities into work packages that a machine can repeat, record and escalate to a person.

Inspection Is Likely to Automate Before Construction Judgement Site inspection is an unusually broad term. It can describe everything from taking progress photographs to deciding whether structural, waterproofing, fire-safety or finishing work complies with an approved design.

Those activities should not be treated as equivalent.

The inspection functions most suited to early automation are generally repetitive collection tasks:

Travelling along a predefined route at scheduled intervals. Capturing photographs or video from consistent positions. Recording visible progress in rooms, corridors or plant areas. Checking whether an access path has become obstructed. Identifying obvious changes between one patrol and the next. Flagging an area for review when the observed condition differs from a reference image or model. Creating a time-stamped visual record for the project team. A mobile robot may perform these activities more consistently than a busy supervisor who is simultaneously coordinating deliveries, trades, safety issues, requests for information and programme changes.

What the robot cannot automatically establish is whether every visible difference is a defect, an approved variation, incomplete work or a normal stage in the sequence.

That decision may depend on drawings, specifications, approved substitutions, manufacturer requirements, site instructions and conditions that are not apparent in an image.

The near-term opportunity is therefore automated observation, not autonomous acceptance.

Robot Evidence Does Not Replace a Mandatory NSW Inspection This distinction is particularly important in NSW.

The NSW Government’s guidance on principal certifiers states that the principal certifier carries out mandatory inspections during construction and may request additional documents to establish compliance.

A robotic patrol could produce useful evidence before an inspection. It could help the builder identify incomplete areas, assemble progress records or alert the project team to a condition requiring attention.

It does not become the principal certifier, remove required critical-stage inspections or independently authorise occupation.

Project teams will need to separate at least three layers of review:

Automated capture: The machine records the condition and detects a potential difference. Project review: A competent supervisor, consultant or trade assesses the finding against the project requirements. Formal inspection or acceptance: The authorised person conducts any required statutory, contractual or trade-specific sign-off. Blurring these layers would create a new form of project risk.

A detailed digital report may look authoritative while still lacking the legal status, context or professional judgement required for the decision being made.

Painting Is a More Demanding Test of Physical Automation Painting appears repetitive, which makes it an obvious automation candidate.

A robotic arm can be trained to reproduce a movement over a broad surface, maintain a defined distance and potentially deliver a more consistent application rate.

Yet the visible coating pass is only one part of professional painting.

Before paint reaches a wall, the work may require:

Inspection of plasterboard joints, repairs and previous coatings. Filling, sanding and dust removal. Moisture or contamination assessment. Masking around floors, windows, joinery, fittings and adjacent finishes. Selection of the correct primer and coating system. Control of ventilation, spray drift and access. Judgement around edges, corners, reveals and irregular surfaces. Final inspection under realistic lighting. These preparation and interface decisions are where many painting failures begin.

Automation will not make an unsuitable substrate suitable. It may simply apply the coating more efficiently over the wrong condition.

For that reason, the most credible first use cases are likely to involve broad, repetitive surfaces inside a controlled zone.

Large walls, soffits, car-park areas, warehouses and repeated apartment layouts may offer stronger commercial conditions than a detailed renovation filled with occupied rooms, delicate finishes and one-off architectural features.

Elyment’s Sydney painting and surface-preparation work illustrates why the application method cannot be separated from preparation, protection, access and final-finish expectations.

The Site Must Be Designed Around the Robot Construction technology is often presented as though a robot can be delivered to an existing project and immediately begin replacing labour.

In practice, the project may need to be reorganised around the machine.

A robotic inspection or painting package may require:

A mapped and maintained travel route. Minimum doorway, corridor and turning clearances. Protected openings and defined no-go zones. Stable communications and positioning signals. Battery-charging and secure storage facilities. Controlled interaction with workers, mobile plant and deliveries. A recovery process if the machine stops in a work area. An emergency-stop and isolation procedure. Consistent naming of rooms, levels and inspection points. A person responsible for reviewing exceptions. These requirements are easier to establish on a large new-build project with repeated floors and centralised site management.

They are harder in an occupied Sydney strata building where the access path may involve a loading bay, booked lift, protected common corridor and private apartment.

A machine that operates successfully across an open commercial floor may not be commercially useful when it must be transported, set up and supervised for a small residential work area.

Where Automation Could Fit Into a Sydney Work Programme Progress patrol

Potentially automated component: Scheduled route, image capture and change detection. Human hold point: Interpretation of delays, variations and incomplete work. Operational control required: Consistent route, time stamps, room identification and exception review. Safety observation

Potentially automated component: Detection of visible obstructions or entry into nominated zones. Human hold point: Hazard assessment and direction of corrective action. Operational control required: Current site rules, escalation protocol and protection against false confidence. Painting

Potentially automated component: Repeatable application over broad prepared surfaces. Human hold point: Substrate acceptance, masking review, coating selection and finish approval. Operational control required: Exclusion zone, ventilation, overspray control and tool isolation. Floor-condition capture

Potentially automated component: Mapping visible level changes or photographing exposed areas. Human hold point: Assessment of bond, contamination, moisture, movement and repair scope. Operational control required: Calibrated equipment, known datum and trade-specific testing. Handover records

Potentially automated component: Organising repeat photographs and identifying missing viewpoints. Human hold point: Acceptance of quality, compliance and completion. Operational control required: Data ownership, retention rules and authorised sign-off. Human-Robot Work Zones Create Their Own Safety Risks A machine intended to improve safety also introduces plant, movement, electrical, tooling and control-system risks.

SafeWork NSW’s collaborative-robot guidance treats the robot, end tool, task, operator and surrounding workplace as one connected system.

Its supporting research identifies issues including collisions, unexpected movement, unsuitable tooling, insufficient testing, bypassed safeguards and cybersecurity failures that could cause physical harm.

On a construction site, this means the risk assessment cannot stop at the robot’s factory specifications.

It must consider how the machine will interact with:

Workers who may not be part of the robotic operation. Temporary stairs, ramps and openings. Changes in lighting and floor condition. Mobile scaffolds, trolleys and material deliveries. Wet paint, dust, water and temporary power. The selected end-of-arm tool or camera payload. Manual intervention during a fault or obstruction. An autonomous route established on Monday may become unsuitable on Tuesday after materials are delivered or another trade establishes a work zone.

Route control therefore becomes a live site-management responsibility rather than a one-time programming activity.

A Camera-Carrying Robot Also Creates a Data Question Continuous visual capture can improve project records, but it may also record employees, subcontractors, residents, confidential drawings, security arrangements or activities unrelated to the inspection purpose.

NSW projects should consider the Workplace Surveillance Act 2005 where an employer uses camera, computer or tracking surveillance involving employees at work.

The deployment plan may need to address notification, visibility, signage, authorised use and the purpose for which footage is retained.

The project should also define:

Who owns the captured images and derived reports. Where the information is processed and stored. Who can access footage containing workers or private areas. How long raw footage is retained. Whether the data may be used for productivity monitoring. How subcontractors and visitors are informed. What happens when the AI system makes an incorrect allegation or classification. These are not secondary technology settings.

They affect industrial relations, trust, contractual rights and the willingness of workers to operate alongside the system.

Robotic Work Will Change Tender and Subcontract Scope Before robotic activity becomes routine, construction contracts and work orders will need clearer boundaries.

A usable scope should identify:

The intended function: Whether the robot is collecting evidence, applying material, transporting equipment or supporting another task. The operating envelope: Where, when and under what environmental conditions it may operate. The preparation responsibility: Which party must clear routes, prepare surfaces, establish power and protect adjacent work. The acceptance criteria: What output is measured and who decides whether it is satisfactory. The safety controls: Exclusion zones, emergency procedures, supervision, inductions and isolation requirements. The data terms: Ownership, access, storage, permitted analysis and deletion. The failure pathway: What happens to the programme and cost when the machine cannot complete its task. Without these terms, robotic work could create the same disputes already seen between human trades: unclear preparation, incomplete access, unsuitable preceding work and disagreement over who owns the delay.

Painting Automation Depends on the Trades Before It A painting robot may be accurate but still remain dependent on multiple preceding activities.

Plastering must be complete. Repairs must be cured. Floors and joinery may need protection. Dust-producing work must be finished or contained. Lighting must be sufficient for both operation and quality review.

In a renovation, this connects painting automation directly to demolition and floor preparation.

Sequencing removal, grinding, levelling and installation remains important because concrete dust, adhesive removal and floor grinding can damage or contaminate completed paintwork if the programme is reversed.

The robotic system does not remove this dependency.

It makes the dependency more important because the machine is less able than an experienced tradesperson to improvise around an incomplete handover.

The same principle applies to flooring removal, concrete grinding, levelling and installation .

Automated visual mapping may assist with records, but bond testing, contamination assessment, moisture investigation and decisions about weak or moving substrates still require competent site judgement.

The Economic Case Will Depend on Repetition The commercial calculation is not simply the hourly wage of a worker compared with the operating cost of a robot.

Projects must account for:

Purchase, lease or service fees. Integration and site-mapping costs. Transport, charging and secure storage. Operator and worker training. Work-zone establishment and supervision. Software, connectivity and data-processing costs. Cleaning, maintenance and tool changes. Downtime when the environment falls outside the operating limits. Human review of every exception the system generates. A short apartment renovation may not contain enough repeated work to recover these costs.

A multi-storey project with similar floor plates, repeated wall areas and nightly inspection routes may present a stronger case.

Automation is most valuable where the task is frequent, measurable, physically demanding or difficult to staff, and where the surrounding process can be standardised without reducing quality.

A Practical Pilot Pathway for Sydney Projects A sensible trial should begin with one narrow activity rather than a promise to automate an entire trade.

Select a bounded use case. Choose a repeatable patrol route or one broad painting zone with clear output criteria. Map the complete operating environment. Review thresholds, openings, temporary services, lighting, other plant, public access and daily site changes. Consult the affected workforce. Explain the purpose, operating hours, safety controls, recorded data and escalation process. Define the human hold points. State which decisions the machine may support and which decisions remain with a supervisor, trade, consultant or certifier. Operate in shadow mode. Compare the robotic result with the existing human process before relying on its output. Measure project value. Assess time saved, coverage, false alerts, setup effort, downtime, defects, worker feedback and programme impact. Scale only after the work package is stable. A successful demonstration does not establish that the system will perform across different sites, surfaces or project phases. Could Site Inspections and Painting Really Be Automated Next? Parts of them can.

Construction patrols, repeat photography, progress capture and first-pass anomaly identification are credible automation targets because they can be structured around routes, time intervals and measurable visual outputs.

Painting robots may become practical across broad, repetitive and properly prepared surfaces, particularly where access can be controlled and the same movement is repeated many times.

The harder functions will remain human-led: understanding why a condition exists, accepting preparation, resolving interfaces, modifying the sequence, communicating with other trades and taking responsibility for the final decision.

For Sydney property owners, builders and project managers, the immediate lesson is not to budget for a workerless construction site.

It is to expect better machines to enter carefully defined portions of the programme, creating new requirements around planning, safety, evidence and accountability.

SITE AUTOMATION AND PROJECT DELIVERY REVIEW

Review the Work Package Before Introducing the Machine Assess operating zones, human hold points, preparation responsibilities, site safety, data handling, inspection status and trade sequencing before robotic inspection or automated application becomes part of the programme.

Request a Project and Automation Review

The Operational Conclusion Shimizu’s trials indicate that physical AI is moving from controlled research into real construction conditions.

The progress is meaningful, but the machines are not eliminating the site-management system around them.

They are testing whether that system can be made more observable, repeatable and productive.

Inspection robots will need routes, evidence rules and authorised reviewers.

Painting robots will need prepared surfaces, protected work zones and human quality acceptance.

Both will need clear responsibility when conditions change or the system fails.

The next stage of construction automation will therefore be decided less by how human the robot looks and more by how professionally the project is organised around it.

Sources and References Shimizu Corporation: Construction robotics field trials Sony: Joint construction-site mobility experiments with Shimizu NSW Government: What principal certifiers do SafeWork NSW: Guidelines for safe collaborative robot design and implementation NSW Legislation: Workplace Surveillance Act 2005 Elyment: Sydney painting and surface-preparation work Elyment: Sequencing removal, grinding, levelling and installation Elyment: Flooring removal, concrete grinding, levelling and installation Elyment: Request a Project and Automation Review