--- title: "Embodied AI" date: 2026-07-28 description: AI is stepping off the screen and into machines that act in the physical world. Why embodied AI already belongs on the Board's risk and capital agenda. author: Mario Thomas canonical: https://mariothomas.com/signals/embodied-ai/ --- ## What is Embodied AI? {#what-it-is data-toc="What it is"} Embodied AI is AI with a body: systems that perceive, reason, and act in the physical world through industrial robots, humanoids, and autonomous vehicles rather than through a screen. Foundation models and learned control systems are increasingly being applied to perception, movement, and manipulation, allowing some machines to learn bounded tasks from demonstrations rather than through task-specific programming alone. This is the technology changing the [delivering dimension of work](/blog/the-great-remaking/), the manufacturing, logistics, and maintenance work where atoms move rather than bits. Robotics is already deployed at industrial scale: the International Federation of Robotics' World Robotics 2025 report counted **542,000** industrial robots installed in 2024, more than double the figure a decade earlier. What is changing is the intelligence inside the machines, and the change is no longer confined to demonstrations: BMW reports that Figure 02 supported production associated with more than **30,000** X3 vehicles before the next phase moved into logistics work. The term covers more than humanoids. It runs from industrial arms learning broader task sets, through autonomous vehicles being prepared for Level 4 public-road trials, to machines being tested in hospitals, homes, and care settings. What connects them is not the form but the use of models that combine perception, planning, and control, increasingly allowing capability learned in one task or environment to be reused elsewhere. If that transfer becomes reliable, robotics begins to move from a collection of specialist systems towards a more general technology. That prospect is why the question for a Board is less about robots than about which parts of its operation move atoms. ## Why it matters to Boards {#board-implications data-toc="Board implications"} When AI acts in the physical world, some errors become immediate, physical, and harder to reverse. That adds a class of consequence that digital deployment does not always carry and puts assurance, intervention, insurance, and liability on the agenda before any machine reaches a site. The EU's revised Product Liability Directive applies to relevant products placed on the market or put into service from **9 December 2026**, expressly bringing software and AI within the updated defective-product regime. In the UK, full implementation of the Automated Vehicles Act is intended for the second half of 2027, with an authorised self-driving entity responsible for a vehicle's driving behaviour while its self-driving feature is engaged. I flagged embodied AI as one of the [five forces reshaping the Board agenda](/blog/five-forces-board-ai-agenda/) at the start of this year, and the question I posed then still stands: if our AI causes physical harm, who is liable and how are we insured? Liability is only the end of the chain. Assurance also reaches physical safety, human intervention, cyber compromise, vendor dependence, workplace data, and the organisation's ability to operate when the system fails. The capital question follows close behind. [Morgan Stanley's 2025 long-range scenario](https://www.morganstanley.com/insights/articles/humanoid-robot-market-5-trillion-by-2050) puts the humanoid economy, including its supply chain and support services, at as much as **$5 trillion by 2050**. That is a scenario, not a present market size or an agreed industry forecast. Many organisations whose value chains depend on physical delivery now have a build, buy, partner, or wait decision to monitor. The useful work is to make that posture explicit: map where machines already act physically in the operation, settle the liability and insurance position, and use bounded pilots to learn where the technology is becoming operationally credible. ## Questions Boards are asking {#questions data-toc="Questions"} ### Is embodied AI real, or another hype cycle? Both, in different places. Industrial robotics is established at scale, and humanoids are beginning to enter bounded production work. BMW reports that Figure 02 supported production associated with more than **30,000** X3 vehicles before the next phase moved into logistics. The long-range market projections remain projections. My read is that the direction of travel is clearer than the timing, which is a case for early, bounded learning rather than a large commitment made on the strength of a forecast. ### What should management be able to show the Board today? A map, not a strategy. Where machines already act physically in the operation and its supply chain, which of them carry AI, who can intervene and what happens on failure, what the insurance and liability position is ahead of the EU's December 2026 product liability regime, and one or two bounded pilots that generate learning rather than headlines. If management cannot describe the organisation's physical AI surface, that is the finding. ### Who should own embodied AI at Board level? Resist filing it under technology. The exposure is operational and legal before it is technical, so the natural home is the executive who owns physical operations, with the risk or audit committee sighted on liability, insurance, and safety, and the full Board owning any material capital commitment. Where a Chief AI Officer exists, they coordinate; they do not absorb the accountability. ### How exposed are we, and when does timing start to matter? Exposure tracks how much of the organisation's value chain depends on physical delivery. Manufacturing, logistics, construction, healthcare, and energy sit closest to the change. One horizon is statutory: the EU's revised product-liability regime applies to relevant products placed on the market or put into service from 9 December 2026. The second is operational. Pilot economics should be refreshed as capability, cost, insurance terms, and the availability of suitable systems change, rather than inherited from an assessment made several years earlier. ### Does embodied AI matter outside the factory? In my view the larger market is domestic, and it is the one the projections struggle to price. The World Health Organization projects the global over-60 population doubling to 2.1 billion by 2050, with the over-80s tripling to 426 million, against care workforces that already cannot meet demand. A machine that helps someone stay in their own home, safely and with dignity, answers a need that workforce planning alone will not close. The factory case is about cost; the care case is about demand with nowhere else to go, and demand of that kind tends to find its technology. ### Are autonomous vehicles part of this? They are one of the more mature forms of embodied AI, although conventional industrial robotics is further established operationally. Wayve and Uber have announced plans for Level 4 public-road trials in London under the Department for Transport's accelerated pilot framework, subject to government and Transport for London approval. The UK intends to complete implementation of the Automated Vehicles Act in the second half of 2027, establishing the wider authorisation and liability regime. Many Boards may encounter this first through fleets and logistics, but the application I would also watch is independence: self-driving vehicles could restore mobility to people who can no longer drive. ### What does this mean for the workforce? It moves AI into the delivering dimension of work, the physical work the first wave barely touched, which I wrote about in The Great Remaking. BMW describes its current humanoid deployments as augmentation of physically demanding and repetitive tasks. That is a documented example, not yet a labour-market pattern. The workforce questions nevertheless arrive early: consultation, reskilling, safety representation, and the redesign of operating procedures. They belong on the agenda before a machine reaches the site, not after. ### Where does the supply chain sit? China accounted for **54%** of global industrial-robot installations in 2024, according to the IFR's World Robotics 2025 report. That is evidence of concentration in deployment and manufacturing, not a complete map of the global robotics supply chain. For an organisation planning a physical AI capability, the Board question is where the machines, critical components, software, and models come from, what alternatives exist, and how support and data flows would be affected by trade restrictions or geopolitical disruption. The answer will differ by sector, but the dependency belongs in the capital discussion from the start. ## References - **International Federation of Robotics** (2025): [World Robotics 2025: Industrial Robots](https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years). Establishes the scale of deployed industrial robotics: 542,000 installations in 2024, 4.7 million robots in operation worldwide, and China's 54% share of global deployments. - **PR Newswire** (18 January 2024): [Figure announces commercial agreement with BMW Manufacturing to bring general purpose robots into automotive production](https://www.prnewswire.com/news-releases/figure-announces-commercial-agreement-with-bmw-manufacturing-to-bring-general-purpose-robots-into-automotive-production-302036263.html). /PRNewswire/ -- Figure, a California-based company developing autonomous humanoid robots, today announced that it has signed a commercial agreement with BMW... - **BMW Group** (2026): [BMW Group advances the use of Physical AI in production with Figure 03 project in Spartanburg](https://www.press.bmwgroup.com/global/article/detail/T0458778EN/bmw-group-advances-the-use-of-physical-ai-in-production-with-figure-03-project-in-spartanburg?language=en). First-party confirmation that humanoid robots supported the production of more than 30,000 vehicles over ten months, with Figure 03 now deployed on logistics sequencing. - **NVIDIA** (18 March 2024): [News Archive](https://nvidianews.nvidia.com/news/nvidia-project-gr00t-foundation-model-for-humanoid-robots). The GTC March 2024 announcement of Project GR00T, a general-purpose foundation model for humanoid robot learning. - **NVIDIA** (2026): [NVIDIA Releases New Physical AI Models as Global Partners Unveil Next-Generation Robots](https://nvidianews.nvidia.com/news/nvidia-releases-new-physical-ai-models-as-global-partners-unveil-next-generation-robots). The CES 2026 announcement that brought the foundation-model approach to robotics at platform scale. - **Morgan Stanley** (14 May 2025): [Humanoids: A $5 Trillion Market](https://www.morganstanley.com/insights/articles/humanoid-robot-market-5-trillion-by-2050). The capital-markets case for physical AI: a projected $5 trillion humanoid market by 2050. - **Goldman Sachs** (2024): [The global market for humanoid robots could reach $38 billion by 2035](https://www.goldmansachs.com/insights/articles/the-global-market-for-robots-could-reach-38-billion-by-2035). An earlier, more conservative sizing of the humanoid market, a useful counterweight to the larger projections. - **European Commission** (2024): [Liability for defective products](https://single-market-economy.ec.europa.eu/single-market/goods/free-movement-sectors/liability-defective-products_en). The official summary of the new Product Liability Directive, which extends strict liability to software and AI for products placed on the market from 9 December 2026. - **EUR-Lex** (November 2024): [Directive - 2024/2853 - EN - Product Liability Directive](https://eur-lex.europa.eu/eli/dir/2024/2853/oj/eng) - **Department for Transport** (2025): [Driving innovation – 38,000 jobs on the horizon as pilots of self-driving vehicles fast-tracked](https://wired-gov.net/wg/news.nsf/articles/Driving+innovation++38000+jobs+on+the+horizon+as+pilots+of+selfdriving+vehicles+fasttracked+10062025151000?open=). The DfT announcement fast-tracking commercial self-driving pilots to spring 2026, with full implementation of the Automated Vehicles Act expected in the second half of 2027. - **GOV.UK** (June 2025): [Automated vehicles: statement of safety principles](https://www.gov.uk/government/consultations/automated-vehicles-statement-of-safety-principles). Seeks views on the draft statement of safety principles for automated vehicles. - **Wayve** (June 2025): [Wayve and Uber Partner to Launch L4 Autonomy Trials in the UK](https://wayve.ai/press/wayve-uber-l4-autonomy-trials/). The June 2025 announcement of planned Level 4 public-road trials in London under the UK's accelerated pilot framework. - **Engineering & Technology (IET)** (9 June 2026): [Wayve and Uber say first London robotaxi service is ‘ready to go’](https://eandt.theiet.org/2026/06/09/wayve-and-uber-say-first-london-robotaxi-service-ready-go). Confirms the readiness of the UK's first driverless taxi service in London, and records the union response forming around it. - **World Health Organization** (2025): [Ageing and health](https://www.who.int/news-room/fact-sheets/detail/ageing-and-health). The demographic case behind domestic embodied AI: the over-60 population doubling to 2.1 billion by 2050, with the over-80s tripling to 426 million.