Future Made in Australia is an industrial strategy to create competitive advantage, build economic resilience, deliver the net zero transition and diversify exports through next generation, knowledge-intensive manufacturing. But like any strategy, the operational challenge remains: to secure a durable competitive position amidst decarbonising supply chains and rapidly evolving global markets.
AI offers a way forward, but only if it can be adapted to local conditions. This means investing in an AI fast-track leveraging our capability in high-value, mass customised manufacturing. Australia can collect and manage datasets from bespoke, personalised, short-run production to create “embodied AI” models that are increasingly driving manufacturing competitiveness around the world.
Australia will always export commodities. The question is whether we only export raw materials or whether we add intelligence, manufacturing capability and technology to capture value further down the production value stream. The choice is to continue the current model or create an Australia 2.0 where we own the digital intelligence embedded in the machines and processes we use to manufacture our products. Our ambition should be exporting these systems as well as our products to the world.
Embodied AI is Australia’s manufacturing advantage
Embodied AI is artificial intelligence embedded in physical systems. Robots and automation perceive, decide and act in real environments. The global bottleneck is the large data volumes required for machines to learn new tasks. Australia has a unique opportunity to create niche high-value embodied AI models from mass customised manufacturing data and license these models globally alongside manufactured goods and services.
Australia’s experience in mining, manufacturing and agriculture shows that imported robotics and automation platforms regularly require substantial adaptation, ranging from “ruggedisation” to complete workflow redesign, so that they can work in Australian environmental conditions and production patterns. SwarmFarm Robotics offers a compelling example. Queensland farmers built autonomous SwarmBots because imported machinery failed on soil compaction, variable fields and precision needs. It deliveres 95 percent chemical reduction and 35 percent lower emissions tailored to broadacre operations and is now exported as IP.
Failure would mean Australian manufacturers import embodied AI models for mass manufacturing from China or the US and inefficiently adapt them to our mass customised needs. We’ve done this with robotic hardware. Without creating models ourselves, we’ll eventually import mass customised AI models made elsewhere, if Australia still has a manufacturing industry.
The world is industrialising embodied AI
Embodied AI capability builds on dedicated foundations such as trusted industrial datasets, validated reference tasks, safety standards and repeatable deployment pathways. Industry learned from enterprise AI adoption that commercial advantage comes from domain-tuned models trained on high-quality, specific data integrated into real workflows. Frontier LLMs are broad, general-purpose systems; used without domain adaptation or grounding, they can be generic and hallucination-prone for specialised applications (e.g. manufacturing applications) and often are not the most efficient choice compared with smaller or domain-specific models.
Australia cannot build physical AI one pilot at a time. We already have world-class robotics researchers and engineering talent. What we lack is shared industrial data, reusable digital twins and a trusted pathway from research into real factories. That is why leading economies now treat embodied-AI data and test environments as national infrastructure. The US is formalising manufacturing data and benchmarks, Germany is building federated robotics data spaces, Korea is mobilising industry-wide data sharing and China is accelerating through large-scale training centres.
Australia’s advantage lies in high-mix, low-volume and mass-customised manufacturing, exactly where physical AI delivers the most value. Robots in these environments must adapt to variation in parts, materials and processes, not repeat fixed routines. A “ChatGPT for physical tasks” trained on Australian production data would embed local materials, standards, and safety practices, ensuring our automation reflects how Australian industry actually works.
ARM Hub provides the missing link between research and industry. Data captured on-site flows straight back into each company’s enterprise AI systems for planning, quality, and operations. Following Germany’s lead, all data is governed, anonymised and feeds the national embodied-AI model and digital twin library. Companies gain immediate operational benefit and long-term access to shared models they can retrain for their own needs. This turns participation into an investment, not a risk. It builds a sovereign platform Australia can use at home and export to the world.
Embodied AI matters for Australia because it’s increasingly the interface between digital advantage and sovereign capability. It’s the difference between using AI and owning high-value industrial IP embodied in machines, processes, products, and systems.
What is the AI fast track
If Australia wants world-leading manufacturing capability and a growing and diversified export mix, we need to prioritise embodied AI in manufacturing practices, including robot manufacture in Australia. Manufacturing capability is embodied in:
- Production assets operating reliably in extreme (Australian) environmental conditions
- Secure industrial data flows, supply chains, and governance
- Diverse workforce skills to create, deploy, and maintain systems
- Standards for regulation and compliance processes to growing export markets
- Ability to export integrated solutions (machines + software + services), not raw materials
The enabling mechanism exists in the Federal government’s $17 million AI Adopt Program. This is building adoption infrastructure through a network of AI Adopt Centres, including ARM Hub, which support SME AI awareness, training and adoption. While this program provides an important foundation, it isn’t yet sized or structured for the much more challenging scale of the embodied AI race.
Building manufacturing competitiveness
Future Made in Australia (FMiA) established a national focus and investment framework but still requires the implementation architecture that turns research capability and digital know-how into repeatable industrial uplift for SMEs and supply chains making up the real economy. The next step for FMiA is to build the AI fast track as an implementation engine converting Australia’s AI and robotics talent into high-value industrial capability, owned, built and deployed here, and exported globally.
Winning advanced economies over the next decade will treat industrial data, deployment infrastructure and embodied AI learning loops as strategic assets, then scale them with discipline. Australia can occupy a niche high-value role, but only by moving with confidence at scale through a targeted, well-governed, production-oriented program that puts embodied AI at the centre of Australia’s manufacturing competitiveness ambition.
Dr Troy Cordie is Mechatronics Engineer at ARM Hub, Emeritus Professor Roy Green is ARM Hub chair and Professor Cori Stewart is Arm Hub CEO.
This article was produced in partnership with ARM Hub as part of its sponsorship of the InnovationAus Awards for Excellence 2025.
Do you know more? Contact James Riley via Email.