Tools for the Commons
How ROBOTICS can strengthen the essential systems society depends on
Robotics is now meaningfully reshaping the way labour and capital are deployed across our economy. And no, this is not so much about the C-3POs of the future, but instead functional, often specialised, robots that get the job done.
This acceleration is being driven by two forces. First, hardware deflation; sensors, motors, and components have fallen in cost by up to an order of magnitude. Second, breakthroughs in embodied AI and spatial intelligence are making it possible to operate robots outside of clean, structured factory lines, into the unstructured, complex environments that make up most of the real economy. We believe this is the moment where robotics becomes critical infrastructure, across a very broad set of industries & tasks.
The excitement around ‘physical AI’ is well deserved, but general-purpose robotics is still early. Most systems today require structured oversight. The frontier teams are those building human-supervised deployments that deliver value now while gathering the real-world data needed to push toward autonomy (e.g. in cleaning, construction, inspections). This shift, from hand-coded behaviour to adaptive, data-driven systems, is what makes the current moment feel like a genuine inflection.
At Revent we believe that robotics has the potential to create real value for society. Let me explain.
👁️There are the obvious high impact cases. Robots that automate unsafe tasks in manufacturing, mining - software encoded safety. Robots that improve patient outcomes from more accurate surgery, or better end to end care. Robots that reduce pesticide use, reduce field worker exposure to pesticides and diseases and increase crop yields. Robots that increase the efficiency of resource and critical mineral use by orders of magnitude.
👓 Then there are the less obvious ones. Automating repetitive, manual tasks in sectors that are over-burdened - think, hospital operations, schools, the energy grid, our food system. We are seeing systemic labour shortages in critical sectors across the economy (European Commission dashboard on where these are most prominent).
🕵️Then there are the hidden ones: the tools and systems that enable us to scale these technologies faster and in a more secure and safe way. Think, financing, safety insurance, cyber security, circularity & maintenance, responsible supply-chain and labour mobility.
We are bullish on robotics across a wide range of uses, but we are not blind to the employment impacts of automation. Clearly, this technological shift will change the shape of our job market, in the same way that major advances in manufacturing technology have in the past. This transition will require deliberate work from both the public and private sectors, to give people agency and unlock effective labour mobility.
We’re looking for teams who see robotics not as novelty, but as a lever to solve some of the hardest, most consequential problems in the real economy.
As we’ve explored this space over the past few months, we’ve been asking :
Where does robotics create opportunities to improve efficiency in sectors that are central to society (surgery, the energy grid, water, education, food)?
What are the safety implications of widespread robot deployments?
How will the robotics economy operate? How will robots be owned and financed? And how will robots be serviced, maintained and repaired?
In which sectors are labour shortages both most severe and most amenable to automation?
How can individuals affected by automation retain agency in the transition?
The State of Play in Robotics
Robotics and physical AI is moving fast. European robotics funding is surging, with 2025 on track to exceed the €761m raised in 2024 (Sifted).
This has been heavily driven by hardware costs continuing to collapse: humanoids have fallen from $2m (Boston Dynamics Atlas, 2013) to $90k (Unitree H1, 2024) to $5.9k (Unitree R1, 2025). LiDAR has dropped from $75k in 2015 to $7k in 2024. China dominates these hardware supply chains, with 451,700 intelligent robotics companies registered by end-2024 and over 6.4 trillion yuan in capital. At the same time, AI is speeding up development through generative design tools (DiffuseBot, URDFormer), natural-language programming and spatial intelligence models that cut programming costs by 50 to 70 percent (IFR).
Despite the excitement, real-world deployments of autonomy remain limited. Most environments are nothing like Amazon’s controlled warehouses, and fully autonomous factories are still rare. Philips’ Drachten site uses 128 robots with limited human presence, and Siemens’ Amberg plant remains close to autonomous but is still supervised. Yet meaningful progress is happening in unstructured settings. Starship Technologies only scaled after years of stalled trials. Loki delivers commercial cleaning via full tele-operation. Monumental deploys bricklaying robots in a 1 to 1 human robot pairing that already works economically while generating the diverse data needed for autonomy later. All whilst reducing the long-term injuries associated with masonry work.
Supporting infrastructure is also shifting. Data remains the bottleneck, with projects like Open-X Embodiment collecting more than 2.4 million episodes and new companies building synthetic data and tooling. Robotics data farms are emerging in the US and China to gather rich, diverse demonstrations where sim to real transfer still lags.
These developments set the context for the real question: which use-cases stand to matter most as this transition accelerates?
Data Flywheels
Autonomy in real environments is still limited because robots lack one thing at scale: high-quality, real-world interaction data.
Simulation helps, but sim to real transfer is unreliable when the robot hits bad lighting, unexpected materials or edge-case object geometry. Even when relying on human-in-the-loop or tele-operated demonstrations, the resulting data can still be messy and inconsistent, often requiring significant smoothing and fine-tuning before models can use it effectively. The companies making the most progress accept this reality and design for it. They create value today through human in the loop or tele operated systems, while turning every deployment into training data - some key examples mentioned already in Loki and Monumental.
Neither needs full autonomy to win. What matters is that each job produces real demonstrations in messy, unstructured settings. These datasets compound and gradually push the model toward reliable autonomy for that specific use-case.
The key is to look for use-cases where:
Automation is currently low
You don’t need to remove the human to make money
Each deployment builds data for future autonomy
Two development paths are emerging. One is large generalist models that lay a foundation, from companies like Genesis, Flexion, Skild.ai and the Physical Intelligence Company. The other is use-case specific models/datasets trained directly from deployment data. Covariant’s work before its Amazon acquisition showed the power of this approach, mapping human motion sequences onto different robot bodies and environments. NVIDIA’s Omniverse and Cosmos platforms are becoming the simulation and spatial-intelligence backbone for many teams, while DeepMind’s MuJoCo and others shape the underlying physics models. These tools accelerate the journey, but they do not replace the need for large volumes of domain-specific real-world data.
Architecture and hardware choices shape this flywheel. Staer’s Jan Erik recently talked about how teams are shifting toward smaller, modular models that load only the skills required, for example manipulation or mapping, rather than relying on a single oversized network. Hardware costs continue to fall and many components are trending toward commoditisation, particularly across Chinese supply chains. But for many use-cases (especially manipulation) teams still need to vertically integrate key components (for example, grippers or specialised end-effectors), so hardware simplification is real but not uniform.
For Europe, this creates an opportunity to combine cost-efficient Chinese platforms with local deployment expertise and domain-specific software.
Credit: Agibot
In an attempt to hack data gathering, we’re seeing the rise of dedicated ‘data farms’ often run by major robotics manufacturers, or sometimes dedicated training companies, where human tele-operators train robots on thousands of tasks. These factories exist because nothing replaces large volumes of real-world demonstrations, and they have become one of the fastest ways to build the breadth and diversity of data needed for generalisation.
Value concentrates where data accumulates. Strong robotics companies will be those that gather the right demonstrations quickly, prove generalisation in the field and build a go-to-market motion that reinforces both.
A New Wave of Automation Use-cases
Clearly there is opportunity to identify credible robotics use-cases by focusing on where deployment can work today and where data flywheels can compound into autonomy.
We look for areas where current automation is low, where tasks are repetitive and structured enough to automate over time, where the societal value is high and where the space is not already crowded.
This lens naturally favours specialist systems over generalist humanoids in the short term. We tend to see that the most effective teams focus on a narrow beachhead and build outward. Monumental is a good example: a large vision for construction automation, but an exceptionally focused first robot (fetching mortar and brick-laying) in a sector facing deep labour shortages, where the average mason in Europe is now over 50 and replacement workers are increasingly scarce.
The table below highlights some example use-cases we think are compelling, and where genuine automation is currently relatively limited.
These use-cases share a similar pattern: potential for real customer value today through supervised or semi-autonomous deployments, and a credible path to autonomy as data accumulates. They are examples where we believe robotics can generate meaningful societal benefits.
To illustrate the societal value of automation, consider the impact on workplace injuries. Each year, EU manufacturing, transport and storage experience more than 761,000 non-fatal and 956 fatal injuries, while UK warehousing alone records around 7,000 injuries annually. Even a modest 10% reduction in fatal injuries across these sectors would deliver more than €200 million in societal value, based on standard QALY valuations.
Enablers & the Robotics Economy
Scaling robotics is a systems-level challenge. Beyond the robots themselves, the enabling layers of financing, supply chain access, maintenance, safety and workforce transition need to mobilise for deployments to reach meaningful scale. Friction remains in these layers today, and it’s where we think some of the most important businesses in the robotics transition will be built.
If we are to believe that robotics is a critical part of our future, then the companies that get us there quicker represent big impact opportunities.
Some enablers sit ‘inside’ robotics, for example servicing networks and cyber-physical safety. Others sit ‘outside’ robotics, such as retraining workers affected by automation or helping employers adjust job design as robotics enters their sector. As is the case for all major technological shifts, large-scale adoption of robotics will require both technical and social infrastructure.
One important class of ‘inside’ enablers is the build-out of robotics foundation models and spatial-intelligence platforms (e.g. Staer, Yaak). These platforms provide essential building blocks, from perception and mapping to control systems, dramatically reducing the time and cost to build domain-specific autonomy.
Another is insurance. GE reportedly took out a policy with AIG to cover potential injury or damage from robots operating in its facilities. We see significant potential for robotics data to power more intelligent underwriting. A relevant parallel is Tesla’s launch of its own insurance product, which uses real-time driver behaviour to price policies dynamically for its semi-autonomous fleet.
We are seeing increasing excitement around the financing of robotics, as new players enter to disrupt the traditional asset finance providers. We anticipate a blurring of the line between the financier and robotics provider, as attractive returns incentivise robotics companies to embed financing and securitise their portfolios into the capital markets.
Below are some of the areas we think are needed:
Academic Hubs & Team DNA
Robotics is still an R&D-heavy industry, and the most successful teams tend to have strong roots in academia. These hubs concentrate resources: frontier research, exceptional engineering talent and dense networks of professors and labs.
In Europe, ETH Zurich remains one of the strongest examples, with recent spinouts Flexion and Mimic Robotics. Across the continent, precision engineering hubs in Germany and Switzerland, sensor and actuator clusters in Scandinavia, and robotics institutes at TUM, EPFL, TU Delft and Lund all act as magnets for founders and early technical hires. The UK has similar concentrations at Oxford, Cambridge, UCL and in Edinburgh. In the US, Stanford, MIT CSAIL, CMU, Berkeley’s RAIL and Princeton continue to set the pace in physical AI, with the RAI Institute linking Boston and Zurich as a cross-Atlantic bridge.
Key academic hubs for Robotics (let us know what we’ve missed)
This is what we often look for in founding teams: First, deep ecosystem connections, because selling robots is complex and requires navigating integrators, suppliers and deployment partners. Second, strong academic ties, which give teams early access to emerging methods and a pipeline of world-class talent. Third, exceptional talent density inside the company, similar to what we expect of top R&D labs across AI and cyber. We think these factors help teams to execute at the frontiers of robotics.
One overlooked challenge we see for deployment companies is attracting senior generalist software engineers. Several experienced operators noted that robotics companies are flooded with robotics PhDs but often struggle to hire the seasoned software talent who often think “they don’t fit”. With developments in physical AI, we are seeing this start to shift, with a new wave of engineering talent entering the robotics domain.
Beyond technical excellence, we look for teams that understand the deployment landscape intimately. Selling robots is hard: it is rarely a simple direct sales motion and often involves integrators and complex operational stakeholders. The strongest founders have unfair domain knowledge and a differentiated go-to-market strategy. Vertical robotics companies often win through distribution, operational insight and execution. (thanks to Vincent Faber for helping to bring these thoughts together)
Market Map
We’ve built a database of the most compelling companies operating across humanoids, spatial intelligence, full-stack deployments and enablers. To help make sense of the landscape, we’ve mapped them across two simplified axes: software vs. physical deployments, and generalist vs. specialised. It’s not a perfect system, but it highlights where companies operate in specialisms vs. across the stack.
Rather than being exhaustive, this map reflects the companies that have made a notable contribution to their segment:
We are excited about a new generation of robotics companies building critical infrastructure for our economy. If that’s you, we’d love to hear from you. Reach out on albert@revent.vc
Over the last few months I’ve been able to speak to a host of brilliant people who are building companies, investing or researching in this space. Thanks in particular to Pieter Senster, Sharon Chen, Hannes Bergkvist, Advika Jalan, Vincent Faber, Csaba Hartmann and Estia Ryan for being generous with your time and ideas!