Imagine walking into a factory, showing a robot how to assemble a component just once, and watching it repeat the task without additional programming or training. What sounds like science fiction may be getting closer to reality.

London-based robotics startup Machenta has raised $40 million in seed funding, co-led by Creandum and Union Square Ventures, to develop artificial intelligence capable of learning physical tasks after observing a single demonstration. Founded by researchers from Imperial College London, the company is challenging one of robotics’ longstanding limitations: the considerable time and data required to teach machines new skills.

Learning to learn

Traditional industrial robots excel at repetitive, precisely programmed operations. However, adapting them to unfamiliar objects, changing production lines or new tasks often requires extensive programming, demonstrations, and retraining.

Machenta is developing a different approach known as one-shot learning. Its AI model is trained entirely in simulated environments, where virtual robots encounter numerous objects, movements, and physical interactions. Rather than memorising individual tasks, the system learns how to interpret demonstrations and translate them into actions.

According to Machenta, robots using its model can perform unfamiliar tasks, including packaging, assembling, and organising objects immediately after a single demonstration. The company has also demonstrated robots responding to changes in object positions and recovering from certain unsuccessful movements without additional instruction.

From virtual worlds to real factories

Simulation-based training could dramatically reduce the expense of developing intelligence robots. Instead of collecting thousands of hours of real-world demonstrations, researchers can generate diverse training experiences inside virtual environments. However, transferring these capabilities into physical machines remains challenging.

Known as the sim-to-real gap, the problem arises because simulated environments cannot perfectly reproduce real-world conditions, including friction, unpredictable object behaviour, sensor inaccuracies, and physical disturbances. Machenta argues that training its model across highly varied simulations helps robots adapt to unfamiliar real-world situations.

Other researchers are exploring similar approaches. A 2025 study presented at the Conference on Robot Learning introduced Human2Sim2Robot, a framework that uses a single human demonstration video and simulation-based reinforcement learning to develop robotic manipulation skills. It is worth noting that the method still requires training in simulation and it is not equivalent to immediately learning a new task without further optimisation.

^{Could manufacturing change before 2030?}

Analyst firm Oxford Economics estimates 20 million manufacturing jobs around the world could be replaced by robots by 2030.

The potential implications extend beyond faster robot programming. Factories increasingly need flexible automation capable of accommodating smaller production batches, changing product designs, and customised manufacturing. Robots that can acquire new skills quickly could make automation more accessible to smaller manufacturers, where conventional robotic systems may be too expensive or inflexible.

Yet, significant obstacles remain.

Machenta’s technology is still at the research stage. Precise manipulation, complex movements, and dependable performance across unfamiliar environments remain difficult. Its demonstrations are promising, though, but they do not establish that robots can reliably perform arbitrary tasks under industrial operating conditions. Manufacturers also require rigorous safety validation, predictable performance, and integration with existing production systems.

By 2030, one-shot learning could begin transforming selected manufacturing processes, particularly repetitive tasks requiring frequent adjustments. Whether it can support truly general-purpose industrial robots remains uncertain.

Nevertheless, the direction of robotics research is changing. The next major advance may be that an industrial robot will acquire a new skill almost as quickly as a human followed by having the capacity of performing thousands of predefined tasks. And that could fundamentally change how factories approach automation.

Susan Fourtané is a sci-tech journalist who has been reporting on startups, industrial manufacturing, robotics, AI, and automation for nearly two decades.

Subscribe now

Get 50% off for 1 year

Latvia is taking a significant step towards strengthening its defence technology sector by selecting 12 startups for its first national defence pre-acceleration programme, DIAL (Defence Innovation Accelerator for Latvia).

Announced on 20 October 2026, the initiative will provide each selected company with up to €120,000 in funding, alongside commercialisation support, international mentoring, and opportunities to test technologies with military end users. Developed by Latvia’s Ministry of Defence in cooperation with UniLab, an accelerator established by the country’s research universities, the programme aims to transform early-stage innovations into commercially viable defence technologies. The initiative also reflects Latvia’s ambition to become a competitive player in the international defence technology market.

AI, drones, and autonomous systems take centre stage

Under the theme ‘Autonomy for Superiority’, the programme focuses on autonomous systems, unmanned vehicles, advanced sensors, and artificial intelligence.

Among the selected startups, Detexity is developing OrbiScan, a 360-degree drone detection system combining event-based cameras with edge AI. Unlike conventional radar-based detection, the passive technology identifies drone propellers through their optical signatures without emitting signals. This approach could prove particularly valuable against fibre-optic-controlled drones, which are difficult to detect using conventional radio-frequency monitoring.

Another startup, Unfazed, is developing Sense On, a portable acoustic detection system designed to provide early warnings against first-person-view (FPV) drones.

Meanwhile, Ardlat is working on a compact high-power microwave system capable of damaging drone electronics, offering an alternative to conventional signal jamming.

Other innovations include Comlab’s Hardmode, which enables rapid deployment of private 4G/5G networks when conventional communications infrastructure becomes unavailable.

SUBmerge Baltic is developing autonomous mission-control software for unmanned maritime vehicles, while Ascent Lumina is creating synthetic datasets to improve AI-powered computer vision systems.

These technologies demonstrate how advances in artificial intelligence, robotics, telecommunications, and sensing are increasingly converging within the defence sector.

From university research to international markets

According to The Defense Post, the programme attracted over 30 applications, exceeding initial expectations. Although the original plan was to support up to 10 startups, the Ministry of Defence expanded the first cohort to 12.

Three selected companies, SUBmerge Baltic, Tactical Avionix, and CapitalG, emerged from UniLab’s university pre-acceleration activities, illustrating the potential of academic research to generate defence-oriented businesses.

Over the next ten months, participants will receive guidance from specialists across more than 15 countries and opportunities to validate their technologies with Latvia’s National Armed Forces and international partners. Commercialisation and export developments will be central objectives, with particular attention to North American markets.

What this means for Europe’s defence technology future

Latvia’s initiative comes as European countries increasingly recognise the importance of technological independence, military readiness, and rapid innovation. For smaller countries, investing in specialised defence startups offers an opportunity to develop advanced capabilities without competing directly with established military manufacturers.

The programme could also strengthen Latvia’s position within NATO’s defence innovation ecosystem, particularly by preparing promising startups for initiatives such as NATO’s Defence Innovation Accelerator for the North Atlantic (DIANA). However, developing promising prototypes is only the beginning. Their success will depend on operational validation, procurement opportunities, investment, and the ability to scale production.

The bigger question is whether Latvia can turn its emerging defence technology ecosystem into an internationally competitive industry. If successful, DIAL could offer a model for how smaller European nations can transform university research, entrepreneurial talent, and emerging technologies into strategic defence capabilities.

Susan Fourtané is a sci-tech journalist who has been reporting on startups, unmanned vehicles, sensors, robotics, AI, and automation for almost two decades.

Get 50% off for 1 year