Batteries Not Included
Robotics founders warn that outdated regulation, structural conservatism and an absent insurance market are holding back one of the UK’s most promising technology sectors
The Entrepreneurs Network, in partnership with Mishcon de Reya, recently convened a roundtable bringing together founders scaling robotics companies and the Regulatory Innovation Office (RIO). The event was chaired by the Rt Hon. Lord Willetts, Chair of RIO, so that founders could share their experiences of scaling in Britain and inform the government’s approach to robotics regulation. While the conversation was under the Chatham House Rule, here we pull out some of the themes discussed.
A regulatory system that rewards inaction
A consistent thread throughout the discussion was that the UK’s regulatory architecture tends to actively discourage innovation. Founders described a system in which the entire burden of regulatory risk falls on the innovator. Under the Machinery Directive, companies developing novel robotic systems must self-certify that their products are safe before going to market. If something goes wrong, the regulator retrospectively scrutinises every decision the company made. There is no meaningful way to get advance clearance on whether an approach is reasonable. The result is structural conservatism. As one founder put it: “You’re just encouraged not to do things by the way the regulation is set up.”
This dynamic favours incumbents with deep pockets who can self-insure against regulatory risks, while startups are forced into a conservative posture that stifles precisely the kind of innovation the system should be enabling.
When founders were asked whether the Health and Safety Executive could provide pre-market guidance, the response was telling — nobody in the room knew how to access that support. If it exists, it needs far better publicity.
In contrast, the Maritime and Coastguard Agency was cited as a positive counterexample — an agency that had created a customer-facing, innovation-oriented engagement team, and in doing so had transformed the pace at which companies could work with the regulator.
Pre-market caution vs. post-market learning
A lively debate emerged over where regulatory scrutiny should sit. Drawing on experience in medical devices — a sector in which robotic systems routinely operate in close physical contact with patients and clinicians — one participant argued that the instinct to tighten pre-market approvals after safety scandals is deeply misguided. The real answer, they argued, is stronger post-market surveillance — continuous monitoring, iteration and improvement once a product is deployed in the real world:
“You cannot possibly predict all the things that will happen. The key is to have a regulatory regime that requires you to be continuously monitoring and improving when you’re deployed, rather than believing you can imagine the real world enough to stop bad things happening.”
Agricultural robotics was cited as a British success story of exactly this approach — light-touch regulation at entry, learning from real-world deployment, and fixing problems as they emerged. By contrast, heavy-handed pre-market regulation in California and parts of Europe had, in one participant’s words, “killed agricultural robotics on the vine.”
The additional advantage of post-market regulation, participants noted, is timing. By the time a company is selling products, it has more resources to devote to compliance — rather than being crushed by regulatory costs before it has any revenue at all.
What you call it matters more than what it does
One of the most striking observations came from a co-founder of a surgical robotics company whose product features robot arms that move autonomously near patients and clinicians. Their system is not classified as a robot under the Machinery Directive — it is a medical device. That single classification difference means the regulatory framework asks “how do we enable this dangerous thing to happen safely?” rather than “how do we prevent it?”
Colleagues from industrial robotics backgrounds, the participant said, are astonished by what the system permits simply because the product sits within a risk-benefit regulatory framework rather than a prohibition-oriented one. The implication for the wider robotics sector is significant — how a technology is categorised can matter more than how it actually works.
Similarly, several participants observed that robotic systems operating on road verges, pavements and other ambiguous spaces fall into regulatory gaps where no single body will claim jurisdiction. An autonomous litter-collection vehicle can operate freely behind a highway barrier, but the moment it reaches over to pick up rubbish from the verge — or where no barrier exists — it triggers the same regulatory requirements as a self-driving car, even though it will never enter a traffic lane. As one participant put it: “It’s almost like a designation problem. Nobody will take responsibility.”
Insurance
Insurance emerged as one of the most significant — and least discussed — barriers to scaling robotics in Britain. Several participants reported being unable to deploy robots commercially because they simply cannot get insured, or because the insurance market fails to make distinctions that the regulatory system itself has already drawn.
In warehouse robotics, a technology stalemate exists around safety sensors. Nearly all mobile robots in factories and warehouses use the same basic lidar sensor — a laser that detects objects at a guaranteed distance and triggers an emergency stop. It is not intelligent, but it is certifiable and insurable. Vision-based AI perception systems, which offer far greater capability, are almost impossible to get safety-certified because their probabilistic decision-making does not fit existing certification frameworks. Companies trying to sell vision-based alternatives are told by customers: we cannot get insurance without the laser. Innovation stalls.
Software, not steel
While participants differed on specifics, a broad consensus emerged that the UK cannot and should not try to compete on robotics hardware. China produced approximately 10,000 humanoid robots last year; the United States around 3,500; and the whole of Europe roughly 350. Hardware and mechatronics are rapidly becoming commoditised, with Asian manufacturers dominating on cost and scale.
Where Britain can compete, participants argued, is in software, AI and systems integration layers that make hardware useful. One founder described their vision of building a “spatial AI operating system” — navigational intelligence software that would run on every machine operating within factory and warehouse infrastructure, regardless of which manufacturer built the hardware. Another argued that the real value in humanoid robotics lies not in the metal but in the AI frameworks that control it: “Without software, a humanoid is nothing.”
Several participants observed that UK robotics companies tend to plateau at valuations of £10–15 million, selling a handful of expensive units per year, because the domestic market lacks the depth and the funding ecosystem lacks the patience for hardware-adjacent businesses. The gravitational pull towards the United States — for customers, capital and commercial viability — was described as near-inevitable once a company reaches a certain scale.
Data sovereignty and the Chinese hardware question
As Chinese-manufactured humanoid robots become the default hardware platform, participants raised significant concerns about data sovereignty and security. One participant drew an analogy with 3D printers purchased from China, where the manufacturing data is reportedly copied — and warned that the same dynamic could apply to humanoid robots operating in sensitive environments, capturing video and spatial data.
The group suggested that controlling the software layer is both the commercial strategy and the security strategy — but noted that no clear regulatory framework yet exists for certifying autonomous AI systems embedded in humanoid robots for human environments. Existing CE markings cover industrial robots in controlled settings; they do not address autonomous systems operating alongside people. The EU AI Act introduces relevant requirements, but how these interact with hardware certification remains unresolved.
Standards: essential but double-edged
The role of standards provoked nuanced debate. When a technology is established, standards provide clarity and enable market access. But for genuinely novel technologies — teleoperation with haptic feedback, vision-based autonomous navigation, multi-robot orchestration — existing standards can be actively unhelpful. A company that meets 90 per cent of a standard but cannot meet the remaining 10 per cent due to the nature of its technology is left in a worse position than if no standard existed, because it appears to be non-compliant with a standard it ought to meet.
Next steps
The underlying message was pragmatic rather than ideological. Britain has genuine strengths in robotics — world-class AI research, deep expertise in extreme environments, and a regulatory tradition that, at its best, is respected globally. But the system as currently configured rewards caution over action, favours incumbents over entrants, and leaves founders bearing risks that the regulatory architecture itself should be helping to manage. Until that system changes, it risks a gradual erosion of the UK’s position in a sector that every major economy now recognises as strategically critical.
This was produced in collaboration with Mishcon de Reya, an international law firm. Founders at the session told us how difficult it can be to reach the Health and Safety Executive (HSE) for pre-market guidance. In response, Mishcon de Reya is offering a complimentary 30-minute pop-up session with its regulatory specialists and the HSE, to discuss the challenges your organisation faces in navigating robotics regulation.