Quick answer: Flying a drone, running a scanner and producing a registered point cloud or a scan-to-BIM model with commercial tools ordinarily has a reasonably predictable outcome, so it generally sits with service delivery rather than R&D — though a novel capture method could itself be experimental. What may be a core R&D activity under s 355-25 of the ITAA 1997 is the narrower part where the achievable detection or reconstruction performance was genuinely unknown: classifying elements from noisy, partial point clouds, resolving deviations below the tolerance the instruments and workflow comfortably report, or detecting defects where labelled examples are scarce and failure modes are rare. Evidence does not make an activity eligible, but without contemporaneous evidence the company may be unable to substantiate why the outcome was not determinable in advance. You self-assess.
7 August 2026 — this article describes the current rules. The 2026–27 Federal Budget announced proposed R&DTI reforms for income years starting on or after 1 July 2028. Until any amendments take effect, the R&DTI continues to be administered under the current legislation.
In reality capture the hard part was never the capture; it is what can reliably be concluded from the data. Companies register "digital engineering" or "reality capture programme" as an activity, attach the scanning cost, the software licences and the modelling hours, and are then asked a question the registration cannot answer: what was the outcome that could not be known in advance? This article is about interpreting captured data — live augmented-reality overlay and real-time model synchronisation are dealt with separately in our Insights.
Capture Is Delivery; Interpretation Is Where the Unknown Might Live
A core R&D activity is an experimental activity whose outcome could not be known or determined in advance on the basis of current knowledge, information or experience, and could only be determined by applying a systematic progression of work that is based on principles of established science and proceeds from hypothesis to experiment, observation and evaluation and leads to logical conclusions — and which is conducted for the purpose of generating new knowledge (business.gov.au). The established-science limb bites in point-cloud work: a disciplined loop of retraining a detector and re-scoring it is not by itself a progression based on established science, which here means the physics of the sensing modality, photogrammetric and registration error theory, and the statistics of the evaluation.
Most of a capture workflow fails on the first limb:
Planning and flying a routine capture mission: Where established survey methods and manufacturer guidance allow a competent professional to determine the expected technical outcome in advance, the activity is generally implementation or service delivery.
Registration and georeferencing using established commercial methods: Generally implementation where the required outcome can be determined in advance.
Producing a mesh, orthomosaic or classified cloud with off-the-shelf tools: Configuration, not experiment.
Manual or semi-automated scan-to-BIM modelling: Production work; volume does not convert it into research.
Two of the excluded categories in s 355-25(2) of the ITAA 1997 sit close to this work and are worth testing activities against rather than assuming they fall outside:
Activities associated with complying with statutory requirements or standards
Much surveying and as-built verification exists to demonstrate conformity — to a survey standard, an approval condition, a specified tolerance. Where an activity is associated with complying with a statutory requirement or standard, that subsection excludes it from being a core R&D activity.
Reproduction of a commercial product or process
Any activity related to the reproduction of a commercial product or process by a physical examination of an existing system or from plans, blueprints, detailed specifications or publicly available information. Scan-to-BIM activities should not be assumed to fall within this exclusion merely because they reproduce the geometry of an existing asset. The exclusion applies where the activity is related to reproducing a commercial product or process through physical examination or the other specified sources, and its application depends on the particular facts. The broader exclusion list is on what does not qualify.
Where Core R&D May Arise
Across reality-capture projects the boundary falls consistently in the same place: where achievable performance was the unknown, not where the difficulty was volume, deadline or integration.
Often closer to core R&D
Usually not core R&D
Automated classification of elements from noisy, occluded, partial clouds where the achievable accuracy on that class of asset could not be predicted
Running a vendor's classification tool and correcting the output by hand
Establishing whether deviation can be resolved reliably below the tolerance the instrument and workflow are specified to deliver
Comparing as-built to design at tolerances the toolchain already reports reliably
Defect classification where labelled examples are scarce, failure modes are rare, and it is unknown whether detectable signal exists at all in the captured modality
Retraining a published architecture on plentiful labelled images to a predictable accuracy
Fusing heterogeneous captures (lidar, photogrammetry, thermal, radar) where it is unknown whether fusion yields information neither source contains
Stitching datasets in software designed to stitch them
Automated progress measurement where achievable reliability under real site conditions — clutter, weather, occlusion — is genuinely unknown
Counting placed elements against a schedule using a known method
Two questions separate the columns. Was the unknown about the outcome or about the effort — "we did not know how long it would take" is a business risk, not a statutory one. And would a competent practitioner have predicted the result: if an experienced photogrammetrist would have said "yes, that will work, to about that accuracy", the outcome was determinable in advance, and a documented search of vendor specifications and published benchmarks is what separates a genuine unknown from an unfamiliar one.
Using machine learning does not itself establish core R&D. Where a competent professional could determine in advance that established modelling methods could achieve the required outcome using the available data, the activity is generally implementation rather than experimental R&D.
What This Discipline Actually Produces, and Where It Goes Wrong
AusIndustry expects records created at the time showing that the activities met the eligibility requirements (business.gov.au). In point-cloud work, useful contemporaneous records may include:
The artefacts: A capture log recording scanner or flight positions, the control network used, ambient conditions and the point density actually achieved. A registration report with residuals, not just a pass/fail. A dataset manifest listing every scan epoch, its hash, and which split it was assigned to and on what date. A versioned annotation protocol, with a record of how disagreements between annotators were resolved. A trial log written cycle by cycle, including abandoned approaches — abandoned branches are among the strongest available evidence that the outcome was genuinely unknown.
Where the experiment starts and stops: Contemporaneous records of the technical unknown, hypothesis, evaluation approach and subsequent conclusion can help evidence the scope and timing of the experimental activity. Subsequent productionisation or routine deployment should be assessed separately from the experimental activity, including substantial engineering required to put a working method into production.
What gets versioned: Model weights are the least of it. The preprocessing chain, the registration parameters, the annotation protocol and the dataset split all change results, and a result that cannot be tied to a specific version of each is not reproducible by the team that produced it.
Four failure modes recur:
Dual-purpose capture runs: The same flight produces the client deliverable and the experimental dataset. Purpose is a question of fact, and job numbers, scopes and invoices answer it whether or not the answer is convenient.
Leakage between tuning and hold-out data: Repeat scanning of one asset produces near-duplicate observations, so splitting randomly by frame or by point-cloud tile puts the same physical element on both sides of the split and the reported performance measures memorisation. Splitting by asset, or by capture epoch, avoids it.
Ground-truth uncertainty larger than the effect being measured: A proposition about resolving 4 mm deviation cannot be evaluated against a control survey with an unstated error budget, or one no better than ±5 mm. The truth source needs a stated instrument, method and uncertainty materially smaller than the threshold under test.
Unallocated mixed time: The same technician flies the mission, processes the cloud and runs the trial, often in one day. Separation by task code at the time is a different record from an apportionment reconstructed months later.
Supporting Activities: What Happens to the Scanning Itself
An activity that is not core may qualify as a supporting R&D activity where it is directly related to a core R&D activity. Where the activity is of a kind excluded from being a core activity, produces goods or services, or is directly related to producing goods or services, it must in addition be conducted for the dominant purpose of supporting the core activity (business.gov.au).
The dominant-purpose test can be particularly relevant where a capture activity also produces a client deliverable or forms part of the services ordinarily provided by the business. A drone flight billed to a client as a deliverable produces that deliverable; a control survey commissioned specifically to establish ground truth for a trial, which the client did not request and would not have paid for, is a different proposition on the same afternoon with the same aircraft.
What distinguishes them is usually visible in the capture design: additional passes, denser overlap, extra control points, or repeat epochs at intervals the delivery scope does not require. Purpose is documented contemporaneously or reconstructed afterwards, and the two look different. Which activities are registered, and how expenditure is treated, remain matters for the company and its registered tax agent.
A Hypothetical Worked Example
Illustrative only — a hypothetical, not a ruling, and not a statement that any of it would be eligible.
An Adelaide engineering contractor scans a mid-rise frame weekly under a precast façade installation contract. The delivery product is a registered cloud and an as-built comparison against the design model. The workflow's stated accuracy on panel-face position, combining instrument, registration and control-network error, is ±8 mm. The team wants to know whether panel-face deviation can be resolved reliably at a 4 mm threshold — finer than the workflow is specified to deliver — from the clouds it already captures.
The unknown and the search: Vendor specifications state the ±8 mm figure and are silent below it. Published photogrammetry and lidar benchmarks cover sub-tolerance work on flat machined surfaces, not weathered precast concrete with formwork tie marks and surface staining. Whether signal survives at 4 mm on this surface class is not answerable from that material.
The hypothesis, dated 3 March: An edge-fitting method operating on the registered cloud will classify panel-face deviation of 4 mm or greater with precision ≥0.90 and recall ≥0.80 against total-station truth. Below precision 0.75 or recall 0.60 the approach is recorded as failed.
The truth and the split: Sixty panel corner locations are surveyed independently by total station from the site control network, stated combined uncertainty ±1.5 mm — thin against a 4 mm threshold, and the record says so. The 60 locations are split 36 development / 24 hold-out on 3 March, before any trial, with the manifest hashed.
Trial 1: Plane fitting to the whole panel face on the registered cloud. Development-set recall 0.83, precision 0.41. False positives cluster on panels with tie marks and staining, which alter local point density and intensity return. This rules out an intensity-independent plane fit on its own.
Trial 2: Intensity normalisation added, fitting restricted to a 150 mm band inboard of the panel edge to suppress edge noise. Precision rises to 0.72, recall falls to 0.61 — the band excluded the bowing signal, which concentrates near the edge. This rules out the working assumption that edge-proximal points were noise rather than signal.
Trial 3: Fit to the full face, then compare each panel against its two installed neighbours rather than against the design model — relative rather than absolute deviation. Development precision 0.88, recall 0.79.
Hold-out run, 11 April: On the 24 sequestered locations: precision 0.81, recall 0.71 — below the pre-set targets, above the failure thresholds, and at n=24 the interval around recall spans the target. The result is recorded as inconclusive, with the sample size named as the reason.
What was established: Not that 4 mm resolution is attainable. What the sequence established is that absolute comparison against the design model does not reach the threshold with this workflow, and that a relative-comparison approach warrants a larger validation campaign.
Where the boundary falls: The weekly delivery scans ran regardless of the trial and produced the contract deliverable; the trial reused those clouds. The additional total-station campaign and the annotation of the validation set were conducted for the trial and not requested by the client. Routine client reporting remains delivery; shared software and modelling costs would require activity-level analysis and a supportable allocation by the company and its tax adviser. Whether any of this is registered, and on what basis, is for the company to self-assess.
Where an RSP Fits, and the $20,000 Point
AusIndustry publishes a software development sector guide and an AI sub-guide, and describes Research Service Providers as scientific or technical service providers a company can engage to conduct R&D activities on its behalf, registered in specific fields (business.gov.au). For a capture or digital-engineering team, an RSP may assist with experimental design, technical R&D work and contemporaneous supporting records within the research fields for which it is registered. Earlier involvement can help document the technical question, hypothesis and evaluation approach as the work progresses.
Notional deductions for an income year must generally be at least $20,000 (ATO) — and qualifying expenditure incurred to a non-associate RSP may still form part of the offset where total notional deductions are below the usual $20,000 threshold, provided the services relate to a research field for which the RSP is registered, together with eligible CRC Program contributions as applicable under the substituted base rules in s 355-100(2) of the ITAA 1997. Using an RSP does not guarantee eligibility — you still self-assess. See claiming R&D under $20,000.
Offset rates, the intensity tiers and which entities access the refundable offset are set out on refundable vs non-refundable offset.
Frequently Asked Questions
Q: Is drone or laser scanning of a site eligible for the R&D Tax Incentive?
A: Generally not on its own. Planning a capture, flying it, registering the cloud and georeferencing it use established methods with reasonably predictable outcomes, so they ordinarily sit with service delivery rather than experimental activity. The scanning may qualify as a supporting R&D activity where it is directly related to a genuine core activity and — because it produces goods or services, or is directly related to producing them — conducted for the dominant purpose of supporting it. You self-assess.
Q: Is scan-to-BIM conversion R&D?
A: Routine manual or semi-automated modelling from a point cloud is generally implementation or service delivery where the outcome can be determined in advance. The reproduction exclusion in s 355-25(2) should be considered separately where the activity is actually directed to reproducing a commercial product or process through physical examination or another specified source.
Q: Is automated defect detection from point clouds claimable R&D?
A: It may be, subject to self-assessment against all the statutory requirements and exclusions, where labelled examples are scarce, failure modes are rare, and it was genuinely unknown whether detectable signal exists in the captured modality at attainable performance. Retraining a published model on plentiful data to a predictable accuracy generally will not qualify.
Q: What evidence do you need for a machine-learning inspection trial?
A: You need contemporaneous records sufficient to demonstrate the relevant R&DTI requirements. Depending on the experimental design, useful records may include the technical unknown and prior-knowledge assessment, the hypothesis, evaluation measures, information about the reference or ground-truth method, dataset and version records, trial configurations and results, conclusions including unsuccessful approaches, and a reasonable method for distinguishing experimental work from routine capture and modelling activities.
Sources & Further Reading
legislation.gov.au — Income Tax Assessment Act 1997 — Div 355, incl. ss 355-25, 355-30 and 355-100
Related: R&D for property and construction · R&D for software and AI · what does not qualify · what an RSP is · refundable vs non-refundable offset · R&D Tax Incentive in Adelaide
Talk to Ignition Research if you are planning reality-capture, machine-learning inspection or digital-engineering experimental work and need technical R&D support. As a Registered Research Service Provider at Lot Fourteen in Adelaide, we assist surveying, digital-engineering and construction-technology teams with experimental design, technical R&D work and contemporaneous supporting records within our registered RSP scope. We do not determine R&DTI eligibility or provide tax advice: your company self-assesses and remains responsible for its own claim, with tax advice and lodgement handled by your tax adviser. Get in touch.
This article is general information from a Registered Research Service Provider about the R&D Tax Incentive. It is not tax, legal or financial advice; eligibility depends on your circumstances and you should self-assess and seek your own advice.
Thinking about a project like this?
If you're weighing up an AI, software or technical improvement project and can't tell yet whether it's implementation or research, start with a quick read on where it sits.