Quick answer: Producing VR or AR training content and designing the instruction around it are generally unlikely to be core R&D activities. Core R&D may sit where it is not known whether a simulated cue set produces measurable transfer to real task performance, or whether a fidelity reduction preserves it, and only a systematic progression of work can determine it. Section 355-25(2)(d) also excludes research in social sciences, arts or humanities. You self-assess.
18 August 2026 — this article describes the current rules. The 2026–27 Federal Budget announced R&DTI changes that will apply to income years starting on or after 1 July 2028. Until then, the R&DTI continues to be administered under the current legislation.
A training simulator is usually judged on how it looks and how the session felt. None of that says whether anyone got better at the real task, and none of it produces a result that could have come out the other way. The research question is narrower and harder than the build: does this particular set of simulated cues change measured performance on the real task, and does it still do so when the cue set is cut down?
Scope: This article covers training environments and the human-factors measurement used to validate them. It does not cover VR design review and multi-user model collaboration, or the spatial registration and alignment accuracy of AR on site — each has its own Insight on the Insights index, and this one is written to complement rather than repeat them.
The Test the Activity Has to Pass
Core R&D activities are experimental activities whose outcome cannot be known or determined in advance on the basis of current knowledge, information or experience, but can 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 that are conducted for the purpose of generating new knowledge, including new knowledge in the form of new or improved materials, products, devices, processes or services (s 355-25(1), Income Tax Assessment Act 1997; business.gov.au).
One note on vocabulary: the "competent professional" comparison is AusIndustry guidance wording, not statutory language. The statute asks whether the outcome could be known or determined in advance on the basis of current knowledge, information or experience — not whether the team personally found it hard.
Building the Environment Is Not Usually Where the Unknown Is
Modern immersive development is largely assembly against documented tooling. Authoring scenarios, modelling and texturing assets, recording voice, writing assessment items, wiring the module into a learning management system, and adjusting engine settings to hold a documented frame rate are ordinarily activities whose outcome is determinable in advance. They are generally unlikely to be core R&D activities on those facts, subject to the activity's own facts and the statutory tests. The guidance on AI-related activities says plainly that "Using an AI model or technique that is new to you does not, by itself, mean the activity is eligible for the program" (business.gov.au).
Four exclusions in s 355-25(2) sit close to this work and are worth reading against each activity rather than against the project:
(c) Management studies or efficiency surveys
An evaluation of whether the program improves workforce productivity, throughput or cost per inducted worker is aimed at a management question.
(d) Research in social sciences, arts or humanities
Dealt with in its own section below, because it is the one that decides most human-factors work.
(e) Commercial, legal and administrative aspects
Licensing scenario content, or negotiating IP in the simulator, falls on the commercial side.
(f) Statutory requirements and standards
Activities associated with complying with statutory requirements or standards, "including one or more of the following": (i) maintaining national standards; (ii) calibrating secondary standards; (iii) routine testing and analysis of materials, components, products, processes, soils, atmospheres and other things. Some immersive training is developed to meet a statutory requirement or standard. "Associated with" is broad, and whether development or validation work aimed at meeting a standard falls inside paragraph (f) is a question of fact, assessed activity by activity. This article does not assert a carve-out from it, and neither should a self-assessment.
(g) Reproduction of a commercial product or process
Any activity related to the reproduction of a commercial product or process by physical examination of an existing system, or from plans, blueprints, detailed specifications or publicly available information. Both elements are required — the thing reproduced has to be a commercial product or process. A published protocol or an open-source toolkit is not by itself a commercial product or process, so paragraph (g) does not do all the work people expect of it.
Section 355-25(2)(d): The Exclusion Human-Factors Work Has to Answer
Paragraph (d) excludes research in social sciences, arts or humanities from core R&D activities. Human-factors validation sits near that boundary by construction, because the measurement is taken on a person.
Two questions get conflated here and need to be kept apart:
1. What the systematic progression of work is based on: The established-science limb of s 355-25(1).
2. What field the research is in: Paragraph (d) exclusion.
They can point in different directions. Where the activity investigates how people learn, perceive, decide or behave, paragraph (d) may apply even if the research is conducted as part of developing an engineered system. A separate engineering activity directed at the technical performance of the artefact may require a different assessment. The distinction depends on the substance of each activity and must be self-assessed on its facts. Where the object is how people learn, perceive, decide or behave as a subject in its own right, paragraph (d) is squarely engaged. Content, narrative and visual design work carry their own exposure through the arts limb.
This is not a line an article can settle. It turns on what is actually being investigated, it is assessed activity by activity rather than across the project, and the company self-assesses.
What "Based on Principles of Established Science" Looks Like Here
This limb is where immersive claims most often fail, because a team can run a very disciplined loop of build, demonstrate, survey and revise with no established scientific basis at all. A transfer study rests on principles of established science when it draws on bodies of knowledge such as:
Signal detection theory: Separating detection sensitivity from response bias, so a trainee who simply calls more hazards is not scored as more perceptive.
Psychophysics of the cues in play: Binaural localisation from interaural time and level differences; stereoscopic disparity and depth thresholds; the field-of-view and luminance limits of the display.
Motor learning and skill acquisition: Specificity of practice, retention intervals, and how practice conditions relate to later performance.
Display and rendering physics: Motion-to-photon latency, angular resolution and visual-vestibular conflict as measurable quantities, not comfort opinions.
Experimental design and inference: Blinding, allocation, statistical power, and a criterion fixed before the data exist.
Iterating on a scenario until instructors call it realistic, evidenced by satisfaction ratings, is not this: it has no prediction that could fail. The parallel boundary in software is on our R&D for software and AI page.
A Hypothetical Worked Example
Illustrative and hypothetical only. It is not a ruling, is not based on any client, and the facts below do not establish that anything would be eligible.
An Adelaide simulation team works with a rail maintenance contractor on hazard detection for track workers — principally on-track plant approaching from outside the field of view, and underfoot trip hazards in ballast. The high-fidelity rig costs more per seat than the contractor can deploy across regional depots, so the question is whether a cut-down cue set holds the training effect.
Baseline: On a protected mock worksite, newly inducted workers detect a mean 5.2 of 12 planted hazards in a 90-second walk-through, with a false-alarm rate high enough to make raw hit counts uninformative. Detection sensitivity (d-prime) is used instead.
Criterion, pre-registered on 4 May: Locked in a protocol before any condition was run, and not revised. The reduced-fidelity condition must preserve at least 80% of the reference condition's 14-day improvement in d-prime on the field task, with the lower bound of a 90% confidence interval above zero. Eighty-eight trainees, 22 per condition, allocated by a schedule held by someone outside the team; the field assessor blind to condition.
Held constant & varied: Held constant: Scenario content and hazard set, briefing script, instructor, session length, the field transfer task and its hazard placements, the 14-day interval, and the assessor. Varied by condition: Spatialised audio, stereoscopic rendering, locomotion, haptics and asset realism.
Condition A — full-fidelity reference: Room-scale walking, stereo, head-coupled spatialised audio, haptic vest, photogrammetric assets. Reference effect: d-prime improvement of 0.94 at 14 days.
Condition B — the visual-realism hypothesis (Failed): Asset realism was increased (scanned ballast and track, higher texture resolution, dynamic shadows), with the spatialised audio budget reduced to fund rendering. Presence questionnaire scores rose; transfer improvement fell to 0.31, and median detection latency for plant approaching from outside the field of view was worse than baseline for 4 of 22 trainees. What it showed: Increased visual realism did not compensate for the loss of spatialised audio in this condition, and subjective presence ratings did not track transfer performance.
Condition C — cue-preserving reduction: Spatialised audio and stereo retained; haptics, room-scale locomotion and photogrammetric assets dropped for stylised geometry. Improvement 0.81 at 14 days — 86% of the reference, with the confidence bound above zero. Criterion met.
Condition D — stereo also dropped (monoscopic): Improvement 0.62, or 66% of the reference. Criterion not met, isolating stereoscopic depth as load-bearing rather than decorative for the approaching-plant class.
Result actually reached: A reduced cue set meeting the pre-registered criterion was identified for approaching on-track plant. It did not hold for underfoot trip hazards, where Condition C reached only 0.44 against the reference; that class was left unresolved when the program moved on. The knowledge generated is the ordering: for this task, auditory cueing and stereo depth carry the transfer, and asset realism does not.
Activity boundary: For this example, the candidate experimental activity is documented from the formulation of the hypothesis and experimental approach through to the trials, evaluation and recorded conclusion. Scenario authoring, artwork, voice recording, LMS integration and the rollout of Condition C as the standard depot induction module sit outside it — the rollout is delivery, notwithstanding that it came out of the trials. Roughly 310 hours sat inside the trials against about 1,400 hours of content production and delivery, recorded per run as the work was done.
Supporting Activities Around a Validation Study
Activities that are not themselves core may be supporting R&D activities where they are directly related to core R&D activities. Under s 355-30(2), if an activity (a) is an activity referred to in s 355-25(2), or (b) produces goods or services, or (c) is directly related to producing goods or services, it is a supporting R&D activity only if it is undertaken for the dominant purpose of supporting core R&D activities (business.gov.au). Each limb is tested against the particular activity, not against the simulator or the business.
Where asset or scenario production, or the running of training sessions, produces or is directly related to producing goods or services, the additional dominant-purpose test applies if those activities are being assessed as supporting R&D. Building the four experimental conditions, setting up the blinded field task and analysing the results are distinguishable from producing the module the depots will use; where an activity does both, dominant purpose is a real hurdle, not a formality. Ineligible work generally is covered on our what does not qualify page.
Where an RSP Fits
AusIndustry describes Research Service Providers as scientific or technical service providers, registered in specific fields, that a company can engage to conduct R&D activities on its behalf (business.gov.au). Qualifying expenditure incurred to a non-associate RSP may still be eligible for the offset below the usual $20,000 threshold, provided the R&D activities are within a research field for which the RSP is registered. See claiming R&D under $20,000 and what an RSP is — and using an RSP does not guarantee eligibility — you still self-assess. Offset rates and how they apply are covered on our refundable vs non-refundable offset page.
Ignition Research works with simulation and training teams on the research side of this problem: defining the real-task measure, choosing a sensitivity metric that response bias cannot inflate, and fixing the criterion and conditions before any trainee is booked. As a Registered Research Service Provider at Lot Fourteen in Adelaide we supply and structure research capability; we are not a registered tax agent, and your company self-assesses and remains responsible for its own claim. Talk to Ignition Research before you commit a training validation budget — get in touch.
Frequently Asked Questions
Q: Is developing VR safety training content eligible for the R&D Tax Incentive?
A: Producing scenarios, assets, voice and assessment items, and integrating a module into a learning platform, are ordinarily activities whose outcome is determinable in advance, and are generally unlikely to be core R&D activities on those facts, subject to the activity's own facts and the statutory tests. Core R&D may sit in experimental work where it cannot be known in advance whether a particular set of simulated cues produces a measurable change in performance on the real task. Note also that where the training activity is undertaken to comply with a statutory requirement or standard, and "associated with" is broad. You self-assess.
Q: Does s 355-25(2)(d) exclude human-factors research?
A: Paragraph (d) excludes research in social sciences, arts or humanities from core R&D activities, and human-factors work sits near that boundary because the measurement is taken on a person. Where the investigation is into how people learn, perceive or behave as a subject in its own right, the exclusion is squarely engaged. Where the investigation is into whether a specific engineered configuration produces a specified performance outcome, the human measure is arguably the instrument rather than the subject. It is a question of fact assessed activity by activity, and the company self-assesses.
Q: What is transfer of training and why does it matter for eligibility?
A: Transfer of training is the measured change in performance on the real task after training, as distinct from performance inside the simulator or how the session was rated. It matters because a core R&D activity needs an outcome that could not be known in advance and a progression of work that leads to logical conclusions. A simulator score or a presence questionnaire generally cannot falsify a claim about the real task; a pre-fixed criterion measured on the real task can.
Q: Can reducing VR fidelity while keeping the training effect be a core R&D activity?
A: It can be the shape of one, where it is genuinely unknown whether a reduced cue set preserves a measured transfer effect and only a systematic progression of work based on principles of established science — signal detection theory, psychophysics of the cues, motor learning, display physics — can determine it, conducted for the purpose of generating new knowledge. Where the reduction is known in advance to work, or is simply an engine or hardware setting applied within documented limits, that reasoning does not get started. You self-assess.
Sources & Further Reading
legislation.gov.au — Income Tax Assessment Act 1997 — Div 355, incl. ss 355-25 and 355-30
Related: software and AI · what does not qualify · what an RSP is · under $20,000 · offset rates · more Insights
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.