As “robots are moving out of the cages” – toward a geography of robotization

ABSTRACT

There is currently a great deal of confusion about the impact of robotisation, mainly due to the lack of clarity in research on timeframes and technologies. To overcome this, we distinguish between two basic narratives. The first - 'robots in the cages' - is an old story of automation which means that the industrial robots are pre-programmed high-precision machines that are used on a large scale but only in a few industries. In contrast, the emerging story - 'robots are moving out of the cages' - is about the more flexible and autonomous robots at present used on a small scale in the service sector, albeit the application areas are expanding rapidly.The objectives of this paper are first to analyse these two stories in terms of their technological, industrial and organisational characteristics, second, to examine the geographical pattern of global competition for robotisation in these narratives, and third, to discuss briefly the policy challenges involved. For the analysis, we use information from the International Federation of Robotics, the EPO Worldwide Patent Statistical Database (PATSTAT) and the Dealroom.co start-up database.

KEYWORDS:

• Robotization
• robotics start-ups
• artificial intelligence
• territorial development
• Europe

Introduction: two narratives about robotization

The currently prevailing approaches to the impacts of robotization

While there is no doubt that debates about robotization, its impact on employment, productivity and global competition are attracting the attention of the scientific community and policymakers, in three important respects, these debates are extremely confused. The first is the timeframe: What time – past experiences or future expectations – are we talking about? The second relates to technology: What kind of robots – industrial robots, service robots, collaborative robots, software robots or artificially intelligent robots – are we discussing? Finally, there is the question of the analytical perspective from which we are looking at robots: Are robots treated as isolated machines or as part of a broader industrial system and transformation?

Debates over employment impacts illustrates these ambiguities perfectly. Studies on the future of work make predictions for the decades to come, take at face value the technological possibilities that the (expected) development of Artificial Intelligence (machine and deep learning, and even artificial general intelligence, but not of the robotics) may offer, and look at robots as isolated machines assuming a one-to-one replacement of humans’ activity by robots (Frey and Osborne 2013; Arntz, Gregory, and Zierahn 2016; Acemoglu and Restrepo 2017; Nedelkoska and Quintini 2018; Lordan 2018). Their grossly uncertain predictions for replacement range from 10% to 50% of the jobs and popular authors often arrive at the dystopic vision of a “robocalypse” (Ford 2015, 2018; Bostrom 2017; Zuboff 2019; Frey 2019; Acemoglu and Restrepo 2020).

In contrast, many studies analyze the past decades by taking the currently large-scale available technology, i.e. the industrial robots, and look at robots as machines that transform the industrial-economic system. By doing this, they customarily conclude that there is nothing new under the sun, robots are machines as usual and their impact on employment is very close to previous waves of automation (Jäger et al. 2015; Graetz and Michaels 2018; Servoz 2019; Fernández-Macías, Klenert, and Antón 2021).

Thus, it comes as no surprise that the research at its current stage does not provide much help to geographers and policymakers because the timeframe (future vs. past), the technology (Artificial Intelligence vs. industrial robots) and the perspectives (one-to-one replacement of humans’ activity by robots vs. robots as part of industrial-economic transformation) are sharply different.

From “robots in the cages” to “robots are moving out of the cages”

There are a number of ways to overcome these analytical ambiguities and perhaps the easiest starting point is the classic product life-cycle model (Vernon 1966) as at present the different robotics technologies are in very distinct stages. Mass deployment of industrial robots has taken place over the last three decades and with a global stock around 2.7 million they are today in the growth stage. Service and collaborative robots that primarily offer assistance to human’s work are in the innovation/development stage with some examples of more advanced technologies exploring market potential and even introducing commercial products. Finally, fully autonomous and artificially intelligent robots are at present in the extremely initial phase of innovation/development.

As simple as it may seem, this classification is at least as critical because the different stages of the lifecycle reveal two distinct narratives with very disparate dynamics. The first – which can be labeled as “robots in the cages” – is an old and very familiar story of automation, which means that the industrial robots are neither intelligent nor autonomous. In fact, they are high-precision industrial machines working in strictly controlled environment (usually operating in protective and fenced cages), carrying out pre-programmed tasks and are predominantly deployed in manufacturing mass production for relatively few tasks and industries. In contrast, the emerging story could be coined as “robots are moving out of the cages”, meaning that the robots, which are today in the innovation and market exploration stage are more flexible and autonomous machines than the industrial robots. They are able to adapt to changing conditions and operate outside the mass manufacturing, primarily in diverse applications for the service sector or small-scale industrial settings where they are working alongside with humans. Above all, the increased use of sensor technologies is driving this development, which leads to a fundamentally different shop-floor configuration and enables workers to collaborate with intelligent machines.

Main characteristics of the narratives: technologies, industries and organizations

These two stories differ substantially from each other with respect to the technology applied, and the organizations and industries where they are deployed (Figure 1). In this paper, the story of “robots in the cages” relates to industrial robots, and we apply the widely accepted definition of the International Federation of Robotics (IFR based on ISO 8373:2012) referring to industrial robot as “an automatically controlled, reprogrammable, multipurpose manipulator, programmable in three or more axes, which can be either fixed in place or mobile for use in industrial automation applications” (IFR 2020a, 23). In other words, an industrial robot is a machine, which is reprogrammable (the programmed motion can be changed without physical alteration), is multipurpose regarding its functions (it can be adapted to different applications and operations with physical alteration), its mechanical structure allows physical alterations, and is mounted to specify its motion in a linear or rotary mode.

Figure 1. The basic narratives according to their technological, industrial and organizational features.

Source: author’s own design

The present-day industrial robots are predominantly used in industrial automation application for a relatively small and well-defined scope of tasks, such as assembling and disassembling, processing (e.g. cutting and grinding), dispensing (e.g. painting and spraying), material handling operation (e.g. picking, placing, packaging, measuring and testing), and welding and soldering (IFR 2020a; OECD 2019). Similarly, the majority of industrial robots are at work in very few industries, and in 2019 three manufacturing subsectors, automotive (34%), electrical/electronics (25%) and metal (10%) accounted for more than two-thirds of the global stock (IFR 2020a, 84). In Europe, the automotive sector plays a bigger role, as almost every second industrial robot is installed in this subsector, and together with rubber and plastics, and metal products subsectors is responsible for three quarters of all industrial robots deployed. Moreover, in Europe looking at the industrial robot densities across manufacturing subsectors, car manufacturing (60 robots per thousand workers), production of plastic (30) and metal products (11) have the highest densities, while in the remaining subsectors (e.g. textile, paper, plastics and chemicals) the densities are very close to zero (Klenert, Fernández-Macías, and Antón 2020, 22).

These facts, i.e. limited operational tasks and fewer industries deploying industrial robots, together inherently favor the large-scale manufacturing driven by global corporations. Large companies have significantly higher financial resources than their smaller counterparts, are more experienced with the introduction of new production technologies, and – perhaps most importantly – have higher economies of scale to make the deployment of industrial robots efficient.

The story “robots are moving out of the cages” is centered around the service robots, and again taking the IFR interpretation (based on ISO 8373:2012) a service robot is defined as “an actuated mechanism programmable in two or more axes with a degree of autonomy, moving within its environment, to perform intended tasks. Autonomy in this context means the ability to perform intended tasks based on current state and sensing, without human intervention” (IFR 2020b, 15). Thus, it is the intentional application that classifies a machine as a service robot, and in the words of IFR “a service robot is a robot that performs useful tasks for humans or equipment excluding industrial automation application” (IFR 2020b, 15). With regard to applications, the IFR distinguishes between the service robots for personal use, such as cleaning robots or personal mobility assistance robots, and the service robots for professional use and commercial tasks, such as delivery robots in offices or surgery robots in hospitals, which are usually operated by properly trained operators.

The application areas of service robots for professional use are very wide: field robotics, logistics systems, medical robotics, professional cleaning, inspection and maintenance systems, construction and demolition, rescue and security applications, defense applications, powered human exoskeletons, autonomous underwater vehicles, unmanned aerial vehicles, and robots for public environment (IFR 2020b, 17). Similarly, in terms of the type of robots the spectrum is wide-ranging, such as milking robots, robots for livestock farming, mining and demining robots, space robots, floor, window, tank and tube cleaning robots, building construction robots, autonomous guided vehicles in manufacturing and non-manufacturing environment, cargo handling robots, robot assisted surgery, rehabilitation systems, fire and disaster fighting robots, surveillance robots, unmanned aerial vehicles (drones) and ground-based vehicles for defense, hotel, restaurant and bartender robots.

Nevertheless, the market is relatively small, as in 2019 only 173,000 service robot units for professional use were sold, and three application areas – logistics (43%), robots in the public environment (11%) – accounted for two-thirds of robots installed. In the same year, the market was worth of US$11bn, and again three application areas – medical robotics (47%), logistics (17%) and defense (15%) were responsible for almost 80% of the market value.

Following the logic of IFR classification, collaborative robots are also part of the story “robots are moving out of the cages”, as they – in contrast to traditional caged robots in industrial mass production – are designed to perform tasks in collaboration with workers in industrial sectors and to operate alongside humans in shared workspaces, where human and robots usually perform the tasks sequentially (IFR 2019a). More precisely, according to IFR “a collaborative industrial robot is an industrial robot that is designed in compliance with ISO 10218–1 and intended for collaborative use” (IFR 2020a, 51), usually relieving humans of many heavy, unergonomic and tedious tasks. Hence, in statistical terms, collaborative robots are a subset of the industrial robots.

The latest figures show, however, that their market is still in its infancy, as in 2019 less than 5% (18,000) of the total industrial robots sold and installed (373,000) were classified as collaborative robots. Collaborative robots are usually deployed in small-scale industrial settings and are part of the process by which the traditional manufacturing and digital technologies are merging, commonly termed as Industry 4.0 which includes and combines other digital technologies, such as Internet of Things, 3D printing, augmented reality, cloud services, cyber security, simulation and sensor technologies, and Big Data analytics.

Finally, artificially intelligent robots, which are a synthesis of robotics and Artificial Intelligence (AI), are to be assigned to the story of “robots are moving out of the cages.” In this case, the AI programs are embedded into robot systems that enable robot mobility in unpredictable environments. For instance, AI algorithms can be used as inputs into physical machines, ie. robots, which execute different tasks. Similarly, robots equipped with sensors can be programmed using AI to identify specific objects regardless of their spatial location. Through Machine Learning, which is a technological sub-area of AI, the robot can teach itself how to pick up an object it has not encountered before. However, at present the market of artificially intelligent robots is exceedingly small and they are deployed mostly in the service sector, such as warehousing robots with path-finding algorithms or drones equipped with autonomous navigation. Due to its extremely small market and the fact that currently statistics are unable to capture and classify this type of robot, we omit artificially intelligent robots from the analysis.

In sum, the robots in both stories are understood as machines which have physical form, and which are reprogrammable to interact with the physical world via sensors and actuators and carry out a series of actions autonomously or semi-autonomously. Therefore, we exclude from the analysis the so-called software robots, which are computer programs operating autonomously to complete virtual tasks, and only exist within a computer system. By doing this, we leave out from the story “robots are moving out of the cages” the search engine “bots” or “web crawlers”, Robotic Process Automation, and chatbots. They are often referred to as robots, even though they are software programs with no physical form and thus somewhat misuse the term of robot.

Our main assumption is that these two distinct stories differ from each other not only regarding the technologies applied and the organizations and industries in which they are deployed, but also with respect to their geographical patterns. Hence, in this paper, we next outline our analytical questions and present the main data sources. We go on to analyze the old story of ’robots in the cages’, provide new perspectives on the spatial distribution and employment impact of industrial robots and offer further insights into policy challenges involved. Then, we turn to the emerging story of ’robots are moving out of the cages’. In detail, first we analyze the different dynamics of the robotics industry and Artificial Intelligence technology. The reason for this is the fact that in current debates about the impacts of robotization and automation, most of the popular literature is confused and talks about robots but usually deals with Artificial Intelligence. After this, we scrutinize the two main trends: the expanding scope of application areas of robots in sectors other than mass manufacturing and the complementary roles of large corporations and small start-ups. Based on this analysis, we examine spatial patterns within the framework of “robots are moving out of the cages”, further research needs, and policy issues related to global competition. In the conclusion, we briefly discuss who is leading the global race in the different stories.

Research questions and data

The basic narratives about robotization are distinct not only because of their subject and the stage they are in the corresponding life-cycle – industrial robots in the growth stage and service and collaborative robots in the innovation stage – but also regarding the hotly debated issues such as employment and productivity impacts, territorial diffusion and global competition. While all of these aspects are important, this paper focuses on the global contest with a particular stress on Europe and aims to answer the following main question: How is the global competition evolving in the different robotization narratives? In addition, we touch upon the main drivers and governing rules beyond the global race and discuss briefly the policy challenges that have been raised in the framework of different robotization narratives.

For the narrative “robots in the cages”, the principal source of information is the International Federation of Robotics (IFR), which provides consolidated measures of industrial robot stock by country, year and industry breakdown, and calculates robot densities measured as the number of industrial robots per 10,000 persons employed in respective industries.

For the story of “robots are moving out of the cages”, we retrieved and analyzed patent data from the PATSTAT 2019 Spring edition, as maintained by the European Patent Office (EPO). One main advantage that patent offer in the analysis of technological innovation developed for commercial purposes is due to the International Patent Classification (IPC) scheme, which is a hierarchical classification system used primarily to classify and search patent documents according to technical fields. In this work, we also used the Cooperative Patent Classification (CPC), which is an extension of the IPC and jointly managed by the EPO and the US Patent and Trademark Office (UPSTO). We focused on patent families pertaining specifically to robotics, following the methodology developed by the UK International Patent Office (UKIPO 2014), and replicated by the World Intellectual Property Organization (WIPO 2015).

More specifically, PATSTAT was queried for patent documents with IPC/CPC classes pertaining to robots and the term “robots” or “robotics” in the title and the abstract of the document. Given that, documents from all intellectual property offices were retrieved and in order to avoid double counting the unit of our analysis was the extended patent family (International Patent Documentation, INPADOC). The patent families in this analysis are fractionally counted according to their year of worldwide first filing, and the patent assignee data from PATSTAT were matched with data from ORBIS (a proprietary database maintained by Bureau van Dijk, which contains information on more than 365 million companies) at the level of individual companies (including subsidiaries where available) using a series of probabilistic string-matching algorithms. In total, 6,223 robotics-related patent families were identified for the period of 1995–2016.

While this method cannot tell the difference between the robotics patents for robots with a physical shape and those for software robots, we must emphasize that the software robot is a relatively new phenomenon, and our patent analysis covers the period from 1995 to 2016. Therefore, robotics patents for software robots, which are software programs that only exists in computer systems and as such are excluded from both robotization narratives, play a relatively minor role in the patent sample. In addition, a detailed analysis of the patent sample, broken down by patent assignees and industries according to NACE REV 2 classification shows that almost 75% (4,662) of the robotics patents were filed by companies, and two-thirds of these companies have manufacturing activities (3,079), meaning that they are producing physical objects. Moreover, almost every fifth robotics-related patents (1,239) were filed by the eight largest industrial robot manufacturers (Fanuc, Yaskawa, Seiko-Epson, Kuka, ABB, Kawasaki, Denso, Mitsubishi).

With respect to start-ups our information source is Dealroom.co, a leading global platform for intelligence on startups, which provides comprehensive data on venture-backed companies in every country and ecosystems throughout the world with a very detailed breakdown by location, industry, technology, funding and market value. We used the predefined taxonomy of Dealroom to filter all start-ups in robotics industry and Artificial Intelligence technology and, in this way, 4,000 robotics start-ups and around 14,000 start-ups engaged in AI technologies were included in the analysis.

Similarly to the patents sample, the Dealroom pre-classified data cannot distinguish between the start-ups that deal with robots in physical form and those that deal with software robots. This is, however, less relevant to the current paper, because our main aim is to highlight the different dynamics of start-ups in robotics and Artificial Intelligence. Furthermore, only around one in four robotics start-ups falls in the cross-section with AI (936 of the total of 3,998 robotics start-ups), and thus we can assume that the majority of robotics start-ups focus on robots in physical form. Start-ups in the cross-over between the robotics industry and Artificial Intelligence technology account for 56% of total funding of robotics start-ups (US$18.2 bn) which indicates significant differences between the dynamics and scaling of robotics start-ups mainly dealing with physical robots and start-ups dealing with AI.

In terms of territorial scope, this study focuses on the large economies – US, EU, China, Japan and South Korea – which currently take the lead in robotization, but also pays special attention to countries that successfully specialize in robotics in terms of global comparison, such as the United Kingdom, Switzerland, Canada, Israel and Singapore. Within the European Union, we provide additional information on the largest economies, Germany and France, the Nordic countries, Denmark, Sweden and the Netherlands, and Czechia, which represents the countries of Central and Eastern Europe.

“Robots in the cages”: research evidence, new trends and policy challenges

Research evidences

The old narrative is fairly well documented, not least because today the use of industrial robots has become the manufacturing standard in some industries. Studies highlight that the deployment of industrial robots has caused an adverse impact on manufacturing employment (Acemoglu and Restrepo 2017; Waldman-Brown 2020), though at a manageable scale (Craglia 2018; European Commission 2018), while recent studies using econometric models have found neutral or slightly positive impacts on total employment (Dauth et al. 2017; Klenert, Fernández-Macías, and Antón 2020). As a trade-off to this, scholars agree that the deployment of industrial robots has triggered significant productivity growth (Dauth et al. 2017; UNCTAD 2017; Kromann et al. 2020) by an order of magnitude similar to that of steam engine technology during the first Industrial Revolution in Victorian Britain (Graetz and Michaels 2018).

Concerning global competition, the deployment of industrial robots is heavily concentrated in five large economies – US, China, Europe, Japan and South-Korea – and China is now the global leader with respect to the absolute number of industrial robots deployed (Table 1). In terms of robot density, however, which shows in a more precise manner how deeply robotization transforms the manufacturing industry of the economy in question, China is still lagging behind its peers, even though the gap has started to close. In contrast, smaller countries, such as Denmark, Sweden and Singapore have very high densities, and the latter currently has the world’s highest figure.

Table 1. Development of the operational stock of industrial robots between 1995 and 2019 (1,000 industrial robots) and robot density in 2019 (industrial robots per 10,000 employees in manufacturing) in selected countries of the world economy.

	Number of industrial robots installed (1,000)				Robot density
	1995	2010	2015	2019	2019
United States	57	173	234	300	228
China	0	53	256	783	187
European Union	109	324	388	505	N/A
Germany	51	148	183	222	346
France	13	34	32	42	177
Denmark	1	4	5	7	243
Sweden	4	9	12	14	277
Netherlands	2	5	10	14	194
Czechia	0	2	11	19	147
Japan	387	308	287	355	364
South-Korea	18	101	210	319	855
United Kingdom	8	14	17	22	89
Switzerland	3	4	6	10	161
Canada	N/A	N/A	12	25	165
Israel	0	1	1	2	49
Singapore	3	4	9	22	918
Rest of the World	16	66	161	379	N/A
World total	601	1,048	1,581	2,722	113

Source: author’s calculation based on data from IFR.

In most of Europe rising robot stock has gone hand in hand with robot density well above the global average throughout the past two decades, and since the mid-2010s there are also signs of convergence among the latecomers (Jungmittag 2020). For instance, while in the 2000’s there were hardly any industrial robots to be found in Central and Eastern Europe (Czechia, Hungary, Poland and Slovakia) and at the end of the decade the combined robot stock only comprised 11,000 robots, in 2019 around one in ten European industrial robots were deployed in the region (52,700 respectively). Yet, only one manufacturing subsector is responsible for most of this increase as between 2014 and 2019 almost 36,000 industrial robots were newly installed, of which around 53% (19,000) were put at work in car manufacturing.

New trends to examine

Studies of the old narrative are usually based on aggregate data, though, taking into account the practical and everyday use of industrial robots, three additional factors need to be examined:

• Firstly, at current level of robotics technology, the deployment of industrial robots is limited both in tasks and industries concerned, and these facts together intrinsically favor the large-scale manufacturing driven by global corporations.

• Secondly, the implementation of industrial robots is very often coupled with the establishment of a new system of production or a new plant and the cost of robots makes up just a fraction of the total investment (OECD 2019). Since implementation of industrial robots in production processes in existing factories is associated with very high costs and risks, it is more advantageous to set up new factories, which are designed specifically for industrial robots (Waldman-Brown 2020). Therefore, the use of industrial robots only affects a relatively small part of the total employment in the new plant.

• Thirdly, while studies are focusing almost exclusively on the deployment of industrial robots, the industrial robots themselves are also embedded in a broader system of the robotization chain which connects robotics developers, robot manufacturers and companies that deploy the robots (Cséfalvay and Gkotsis 2020). Thus, to capture all the impacts of the industrial robots, the parts of the chain prior to use by companies in different manufacturing processes must be analyzed too.

These features are mostly overlooked in research, although they significantly influence the issues at stake and, in many cases, sharply contradict dominant research assumptions. For instance, due to the limited scope of the types of application, industrial robots are increasingly used in industries with middle or higher skills requirements and – contrary to widespread expectations (Autor, Levy, and Murnane 2003; Acemoglu and Autor 2010; Autor and Dorn 2013) – to a very limited extent or not at all in industries where low-skilled workers are employed and where the routine tasks carried out can easily be replaced by robots, even with present-day technology (UNCTAD 2017; OECD 2019).

Similarly, since the deployment of industrial robots is currently limited to only a few industries, industrial robots are heavily concentrated in countries which specialize in these robotized industries and where shortages in the manufacturing labor force and/or high wages act as incentives for companies to deploy robots (Atkinson 2018; Cséfalvay 2020). Especially for the effect of wages, there is a clear relationship in Europe: the higher the labor cost, the higher the robot density (Figure 2). In the top three countries with the highest robot densities (Germany, Sweden and Denmark) the labor costs in manufacturing are also the highest at over 40 euros per hour, while in the bottom three countries with lowest densities (Romania, Poland and Portugal) the labor costs were less than 12 euros per hour.

Figure 2. Robot density and hourly labor compensation cost in manufacturing, selected EU member states with industrial robot stock > 1,000 robots, 2019 (Pearson’s r = 0.7928).

Source: author’s calculation based on data of IFR (2020a) for robot density and EUROSTAT for labor cost.

By the same token, robot installation is in many cases only part of a larger and newly established production system, which in turn generates jobs as a whole that should be included in the calculation of the net employment effects of robots. Moreover, this fact should be incorporated in the analysis of the time-space diffusion of industrial robots too, as the diffusion is also very often directly linked to the localization decisions of global firms and coupled with foreign direct investment (FDI) streams. For instance, in the 37 emerging economies covered by the European Bank for Reconstruction and Development (EBRD), a regression analysis suggests that a 1% increase in FDI in a given sector and a given country results in a 12% increase in the industrial robot stock (EBRD 2018, 36). Similarly, the recent increase in the number and density of industrial robots in Central and Eastern European countries is largely due to the relocation of production from Western European (and overseas) car manufacturers to the region (Pavlínek 2017).

Finally, while industrial robots were originally implemented in factories to carry out high-precision tasks performed by humans using previous technologies, industrial robots are also embedded in the broader robotization chain, where robotics development and robot manufacturing are also creating jobs and opportunities for employment. Again, and contrary to studies on the future of work, this fact should be included in the net employment impact assessment. In addition, the competitiveness of countries is highly dependant on their position in the different parts of the robotization chain, with the leaders strongly involved in all three parts – robotics development, robot manufacturing and deployment of robots – while the countries which are only present in the final part of the chain where the deployment of robots in the economy occurs are lagging behind in the race

Since roughly half of the robotics-related patents were filed by manufacturing companies (3,079 of the total 6,223 patents) which develop and file patents for robots in physical form, the number of these patent families and, above all, the patent densities (measured as patent families per 100,000 employees in manufacturing) are good approximations for capturing countries’ engagement in parts of the robotization chain prior to the deployment of robots (Table 2). Here, too, the five large economies are leading the global race, in particular, Japan and South-Korea, though China, as a latecomer, is lagging far behind its counterparts, both in terms of number of patents and patent density. Some smaller economies, such as Switzerland and Israel, are successfully specializing in robotics development and/or robot manufacturing, while in Europe, more than half of the patent families were filed by German companies and Sweden also has a very high patent density, albeit the patent density across Europe is below the global average.

Table 2. Development of the stock of robotics-related patents families between 1995 and 2016, and the patent density, 2016, in selected countries of the world economy (patent density = patent families per 100,000 employees in manufacturing).

	Number of robotics related patent families				Patent density
	1995	2010	2015	2016	2016
United States	5	359	988	1,193	0.70
China	0	32	181	580	0.03
European Union	3	445	896	1,011	0.35
Germany	1	278	514	576	0.72
France	1	51	119	133	0.37
Denmark	0	6	16	22	0.63
Sweden	1	50	110	121	2.05
Netherlands	0	13	27	31	0.33
Czechia	0	0	1	1	0.02
Japan	18	610	1,332	1,519	1.33
South-Korea	6	625	1,306	1,410	3.27
United Kingdom	0	22	50	58	0.15
Switzerland	1	88	165	189	3.01
Canada	1	19	46	57	0.25
Israel	1	12	38	44	1.07
Singapore	0	1	7	8	0.21
Rest of the World	0	35	108	154	-
World Total	35	2,248	5,117	6,223	0.52

Source: author’s calculation based on data from PATSTAT and ORBIS for patent stock and EUROSTAT for employment.

Policy challenges for Europe

Because of these specific features, three main policy challenges can be identified in Europe in the framework of “robots in the cages.”

• First, due to the increasing deployment of industrial robots, the highly robotized European economies are facing a trade-off between potential job losses in manufacturing and a strengthening of their industrial base. On the one hand, automated manufacturing requires a smaller but more highly qualified workforce, but on the other hand it increases productivity and industrial output. Therefore, the key question is how to strike the right balance and upskill the labour force.

• The second policy issue is upgrading along the robotization chain. While in Europe there is a time-space diffusion of industrial robots under way, economies that entered the robotization process with a time-lag of more than one decade, lack important parts of the robotization chain, such as robotics development and robot manufacturing. Without strengthening these parts of the chain countries might run the risk of being trapped in the so-called “dependent robotisation” (Cséfalvay 2020), meaning that robotization in these economies relies excessively on the localization decisions of global firms in just one or a few manufacturing sectors. Supporting the robotics ecosystem, including start-ups, and targeted government incentives and procurement policies to source intermediate components domestically could help reduce this risk.

• The third challenge is the reshoring of previously offshored production processes, as robot-based automation increases the possibilities of manufacturing geographically closer to the consumer markets of developed countries (Propis de and Bailey 2020). Nevertheless, at present, reshoring is relatively low and in Europe varies between 3% of the companies that report reshoring in Germany, over around 6% in Belgium and France up to 9% in Sweden, while almost 10% of companies are still offshoring their production (Kinkel, Pegoraro, and Coates 2020).

As “robots are moving out of the cages”: new dynamics and spatial patterns

Robotics versus Artificial Intelligence

Expectations for the emerging story are very high, in particular, when it comes to combining robotics industry with Artificial Intelligence technologies (Bostrom 2017; Franklin ed. 2017; Ford 2018; Susskind 2018; Agrawal, Gans, and Goldfarb 2019; Clifton, Glasmeier, and Gray 2020). Similarly, recent publications highlight that in the future the widespread use of intelligent robots will open up a number of new geographic agendas and topics that go beyond the traditional ones (Bissel and Casino del 2017; Macrorie, Marvin, and While 2021).

Yet, and in stark contradiction to these popular expectations and bestsellers, the artificially intelligent robots are the story of the very distant future rather than current reality. Therefore, by analyzing the narrative of “robots are moving out of the cages” we make a crystal-clear distinction between the robotics industry, which results in robots with physical shape and the Artificial Intelligence technology, which is a branch of computer science that creates software and algorithms capable of problem-solving and learning similarly to humans. Furthermore, this distinction between robotics and Artificial Intelligence is critical not only because of the differences regarding technologies and properties of their products – robots in physical form versus algorithms in computers – but also with respect to their distinct dynamics.

At present, robotics lags far behind Artificial Intelligence in terms of many measures, such as the related patents (WIPO 2019), the investment of large companies and the engagement of start-ups. For instance, regarding venture capital backed start-ups, which play a crucial role in innovation, development and market exploration, especially in niches not covered by large companies, the latest figures show that AI outperforms robotics by a factor of four in both the number of start-ups and their total funding (Table 3). Start-ups in AI also have significantly more scaling options than in robotics, as indicated by the number of the scaleups (start-ups with funding more than 1 M EUR), the “future unicorns” (start-ups with market value between 200 and 800 M EUR) and the “unicorns” (start-ups with market value more than 800 M EUR).

Table 3. Main characteristics of start-ups engaged in robotics and Artificial Intelligence, 2020.

	Number of start-ups	Start-ups > 1 M EUR funding	Start-ups with market value 200–800 M EUR	Start-ups with market value > 800 M EUR	Total funding M EUR
World total	472,639	49,316	4,456	1,715	N/A
Artificial Intelligence	13,849	4,224	300	114	128,670
Robotics	3,998	796	48	27	32,540

Source: author’s calculation based on data of Dealroom.co, retrieved in May 2020.

In addition, AI start-ups are more widely distributed around the world than those involved in robotics (Table 4). While the leading economies, the US, China and the EU together account for around 65% of AI start-ups, three out of four robotics start-ups worldwide are located in these economies. American, Chinese and European start-ups combined received 84% of the total global funding in AI and 88% in robotics. Within this triad, there are, however, strong differences, and particularly the European start-ups are facing a serious funding and scaling gap. With respect to the number of start-ups, the EU has figures in both AI and robotics that are comparable to the US. Yet, when looking at the funding the start-ups received, the EU is an order of magnitude behind the US and China is also surpassing the EU. In other words, while in Europe there is no shortage in innovative ideas and entrepreneurial spirit, the European start-ups lack the venture capital for financing these ideas and the regional ecosystem for scaling up the companies.

Table 4. Global pattern and main characteristics of start-ups engaged in robotics and Artificial Intelligence, 2020.

	Robotics			Artificial intelligence
	Number of start-ups	Total funding (M EUR)	Funding (M EUR) per 1bn USD GDP	Number of start-ups	Total funding (M EUR)	Funding (M EUR) per 1bn USD GDP
United States	1,401	21,355	1.00	4,973	77,772	3.64
China	242	4,672	0.33	457	18,860	1.31
European Union	1,354	2,735	0.18	3,612	11,048	0.71
Germany	196	362	0.09	667	2,677	0.70
France	219	409	0.15	689	3,119	1.15
Denmark	165	458	1.32	101	246	0.71
Sweden	46	80	0.15	168	369	0.69
Netherlands	148	89	0.10	439	1,058	1.16
Czechia	12	0.2	0.00	37	129	0.52
Japan	58	314	0.06	94	730	0.14
South-Korea	32	373	0.23	98	579	0.35
United Kingdom	201	1,101	0.39	1,117	7,329	2.59
Switzerland	96	161	0.23	183	943	1.34
Canada	89	220	0.13	393	2,199	1.27
Israel	120	1,130	2.86	937	6,116	15.48
Singapore	30	162	0.44	165	1,140	3.06
Rest of the World	375	317	N/A	1,820	1,954	N/A
World Total	3,998	32,540	0.37	13,849	128,670	1.44

Source: author’s calculation based on data of Dealroom.co for start-ups, retrieved in May 2020, and World Bank for GDP, retrieved in October 2020 (indicator code: NY.GDP.MKTP.CD, https://data.worldbank.org/indicator/NY.GDP.MKTP.CD?end=2019&most_recent_value_desc=true&start=1960&view=chart

Taking into account the size of the economies in question – and normalizing the funding of start-ups with a measure of total funding per US$1bn of GDP – countries outside of the triad are more intensively involved in AI than in robotics. In robotics, only Israel’s start-ups have the value more than 1 M EUR funding per US$1bn GDP, but in Artificial Intelligence, Singapore, the United Kingdom, Switzerland and Canada have also exceeded this threshold, and Israel stands out by more than 15 M EUR per 1bn USD GDP. Within the EU, the robotics start-ups in Denmark and the AI start-ups in France and the Netherlands have the value more than 1 M EUR funding per US$1bn GDP, although in funding of the start-ups Europe as a whole is lagging behind the US and China in both robotics and AI when compared to the size of the economy.

There are two main reasons for these differences in dynamics. The first relates to the specific feature of innovation economics, as there are more incentives to finance a technology with a wide domain of application, like AI, than to develop an industry in a relative narrow domain, which is currently the case with robotics. Robotics is an industry, and its products are manifested in physically existing tools and machines that perform complex operations automatically and with a certain degree of autonomy. AI is, by contrast, a technology producing highly scalable software and algorithms and promises to become the decisive General Purpose Technology (GPT) of the future with a variety of applications to transform the economy and society (Trajtenberg 2018).

The second factor is the specific business model of the technological start-ups, as they are more engaged in industries and technologies where knowledge and human capital are the assets leveraged to take on larger and more established incumbents or to carve a market niche like in AI. On the other hand, start-ups are less present in sectors that require relatively higher up-front capital investment, as is often the case in robotics. Start-ups engaged in software technologies, such as AI, can scale much faster than those involved in hardware, i.e. robots, due to the lack of manufacturing and associated costs.

In conclusion, there is currently a lot of hype about Artificial Intelligence, but this should not be confused with the rather modest development of robots outside of industrial mass production. At present, the story of “robots are moving out of the cages” is primarily not about the merging of robotics and Artificial Intelligence, but about exploring new application areas for robots in sectors other than industrial mass production and the emerging complementary dynamics between large corporates and small start-ups in this process.

Exploring application areas of robots in the service sector

The old story of industrial robots is characterized by mass deployment albeit in a limited number of application areas, while in the new narrative, there is a wide range of application areas for robots with relatively low deployment. The annual installation of service robots for professional use compared to industrial robots shows clearly that service robots are lagging behind and the number of robots installed annually has increased to over 100 thousand units only in the past few years (Figure 3). The discrepancy between the expanding scope of application areas and the relatively low scale of deployment is partly due to the fact that the implementation of service robots requires not only relatively high up-front investment, but also redesigning and restructuring the business process and reskilling the workers, while the productivity gains – following a J-curve – come only with a considerable time-lag (Brynjolfsson, Rock, and Syverson 2020).

Figure 3. The development of annual installation of industrial robots and service robots for professional use, 2010–2019, thousand robot units sold.

Source: author’s calculation based on data of IFR.

The market structure of the robot manufacturers indicates also very clearly these differences. Leigh and Kraft (2018) note that the 28 industrial robot suppliers which provide data for the IFR are headquartered in just 12 countries, and only four countries worldwide are home to three or more manufacturers; Denmark and Switzerland each have three, while Germany and Japan each have six companies. For the service robots, however, in 2018 the IFR listed more than 750 manufacturers, including 160 start-ups. According to their market, 178 manufacturers produce robots predominantly for personal/domestic use, 510 manufacturers supply robots for professional use, and 64 manufacturers are active in both market segments. Taking the service robot manufacturers producing for professional use, their leading market segments are logistics systems (e.g. cargo handling, autonomous guided vehicles), medical robotics (e.g. robot assisted surgery, diagnostic systems), and field robotics (e.g. agriculture, greenhouse and livestock farming) (Figure 4).

Figure 4. Number of service robot manufacturers by application areas, professional use, 2018.

Source: author’s calculation based on data of IFR (2019b)

In addition, while the market of industrial robot manufacturers is dominated by Japanese and European companies, about 60% of service robot manufacturers are located in the US and the EU (Table 5). In terms of service robot manufacturer density – measured as number of manufacturers per million employees in the service sector – it is very striking that China, which is the world leader in absolute number of industrial robots deployed, lags far behind its competitors in service robot manufacturing. On the other hand, Switzerland, Israel, Singapore and Canada, as well as in Europe, Sweden, Denmark, France and the Netherlands are more specialized in manufacturing of service robots than might be expected given the size of their service sector.

Table 5. Global pattern of service robot manufacturers, 2010–1018.

	2010	2018		2018
	Total number	Total number	Start-ups	Manufacturers per 1 million employees in service sector
United States	64	218	53	1.7
China	1	55	14	0.2
European Union	72	229	44	1.6
Germany	29	65	14	2.1
France	6	52	16	2.4
Denmark	5	9	2	3.9
Sweden	13	16	3	4.0
Netherlands	3	17	3	2.4
Czechia	0	2	2	0.6
Japan	25	47	5	1.0
South Korea	20	32	4	1.7
United Kingdom	11	37	10	1.4
Switzerland	10	33	9	9.2
Canada	14	32	2	2.1
Israel	3	14	4	4.7
Singapore	0	8	1	2.8
Rest of the World	4	47	15	N/A
World total	224	752	161	0.5

Source: author’s calculation based on data of IFR (2019b) for service robot manufacturers and ILOSTAT for employment (Employment by sex and economic activity – ILO modeled estimates (November 2019), indicator code: EMP_2EMP_SEX_ECO_NB_A).

The dynamics between large corporates and small startups

In case of the old story of industrial robots, large corporations dominate almost everywhere across the entire robotization chain (particularly in robot manufacturing and deployment of robots) and small companies and start-ups play a minor role only as robotics developers or as intermediaries and system integrators. Although, as “robots are moving out of the cages” the combination of the big-firm innovation driven by digital giants and the bottom-up entrepreneurial innovation triggered by small start-ups becomes more and more the prevailing model. The battle between the old and the new, the big and the small is, undoubtedly, very spectacular, yet in the real world the giant corporates and the small start-ups play rather complementary roles (Baumol, Litan, and Schramm 2007). Large companies deliver incremental and continuous innovation, access to global market and worldwide scaling, and most importantly, devote immense resources for R&D, while start-ups in the niches not covered by large companies, especially with radical, disruptive innovations that require predominantly intellectual assets and relatively low capital investment, are making a huge contribution to the discovery of new application areas of robots.

Definitely, the start-ups have many other means to protect their intellectual property, though the analysis of robotics-related patents shows that of the almost 4,000 robotics start-ups only 103 start-ups filed patents (as a cross-section by matching the Dealroom startups data and our sample of robotics-related patents) and together they filed just less than 4% of all robotics-related patent families (Table 6). Patenting is more common for start-ups in the growth stage, as start-ups that filed patents received more than 10% (Euro 3.8bn) of the total funding of robotics start-ups. In contrast, looking at the top R&D investors, for whom the EU Industrial R&D Investment Scoreboard provides the most relevant data from 2,500 companies worldwide and 1,000 European companies in addition (Hernández et al. 2019), 167 large companies make up about half of the robotics-related patent families (as a cross-section by matching the Scoreboard data and our sample of robotics-related patents). Large R&D investor companies play a particularly big role in robotics-related patent filing in Japan, Switzerland, Germany, and Sweden, while the US is the only big economy where start-ups have significant patenting activity, as every tenth patent family is filed by start-ups.

Table 6. Global pattern and main characteristics of the robotics-related patent families.

	Number of assignees	Total number of patent families	Patent families filed by companies listed in the EU Industrial R&D Investment Scoreboard	Patent families filed by start-ups	Patent families filed by other assignees (companies, universities, governmental organizations, private persons)
United States	473	1,193	401	129	663
China	250	580	24	9	547
European Union	370	1,005	536	45	424
Germany	152	576	373	6	197
France	78	133	35	18	80
Denmark	10	22	-	6	16
Sweden	26	121	97	3	21
Netherlands	12	31	20	-	10
Czechia	2	1	-	-	1
Japan	179	1,519	1,332	6	181
South Korea	276	1,410	567	1	842
United Kingdom	35	58	14	8	36
Switzerland	26	189	167	7	15
Canada	45	57	-	6	51
Israel	30	44	13	5	26
Singapore	7	8	-	2	6
Rest of the World	95	160	37	3	120
World total	1,786	6,223	3,091	221	2,911

Source: author’s calculation based on data from PATSTAT (1995–2016) for robotics-related patents, Dealroom (2020) for start-ups and EU Industrial R&D Investment Scoreboard database (2019) for large R&D investors.

Concentration is also a defining characteristic of robotics start-ups, although it occurs primarily at the level of regional ecosystems and less at the country level. The main reason for this is the specific nature of venture capital investments, which follow a rather skewed distribution with only a small fraction of start-ups responsible for the vast bulk of innovation and funding, while the overwhelming majority of start-ups tend to receive low or even no funds at all (Figure 5). Of the 4,000 robotic start-ups the top ten accounts for more than 40%, and the top 1% (40 start-ups) for almost two-thirds of the total funding, and only one in five (803 startups) received more than 1 M EUR funding.

Figure 5. Distribution of robotics scaleups (start-ups with more than 1 M EUR funding) according to the size of their funding, 2020, (M EUR).

Source: author’s calculation based on data of Dealroom.co (accessed May 2020)

The spatial pattern follows this trend, with just five ecosystems – San Francisco Bay Area, Greater Boston, Beijing, Shenzhen and Pittsburgh – concentrating more than two-thirds of total funding of robotic start-ups. With regard to scaling options, there are less than 30 ecosystems where at least one start-up can be classified as future unicorn (market value between 200 and 800 M EUR), while only 12 ecosystems have at least one robotics unicorn (market value more than 800 M EUR), and they together add 86% of the total funding (Figure 6).

Figure 6. Leading robotics start-ups ecosystems according to their funding (bn EUR) and scaling performance (number of future unicorns and unicorns), 2020.

Source: author’s calculation based on data of Dealroom.co (accessed May 2020)

The San Francisco Bay Area is home to 16 future unicorns and 10 unicorns and with 12.8 bn EUR has almost 40% of the total global funding of robotics start-ups. Despite considerably lower funding of 2.1 bn EUR, Greater Boston also shows good scaling performance with a high number of future unicorns (6) and unicorns (3). Regarding scaling options, Chinese ecosystems also excel, as Beijing, Shenzhen and Shanghai together have nine future unicorns and six unicorns, with a total funding of more than 4 bn EUR. While in the Israeli ecosystems, the funding is relatively low (around 1 bn EUR combined), the robotics start-ups also scale well and in Tel Aviv-Yafo, Jerusalem and Haifa together there are two future unicorns and two unicorns. In the same way, the ecosystems in the United Kingdom received low funding (combined 1bn EUR), though London, Cambridge and Bristol each has one robotics unicorn.

In stark contrast, Europe is lagging behind not only in funding with around 8% share in total global funding of robotics start-ups (2.7 bn EUR respectively), but also in scaling performance, as it has only one unicorn in Bucharest, and one future unicorn in Odense and in Augsburg. Even in the top European start-up hubs, such as Paris, Berlin, Helsinki-Espoo and Stockholm, the robotics start-ups received very minor funding (between 60 M EUR in Stockholm and 240 M EUR in Paris), while in all other start-up hubs, the robotics start-ups have to live on extremely minor funding.

What we still do not know

The new story of more autonomous service and collaborative robots raises many further research needs, mainly due to the lack of available data. Firstly, while there are trustworthy data on service robot manufacturers, robotics patents and robotics start-ups, information about the deployment and territorial uptake of service robots are sparsely available. As a consequence, apart from case studies, very little is known about the impacts of service and collaborative robots on employment and productivity. We can only assume that their deployment is to lesser extent exposed to one-to-one replacement of humans’ activity by robots, since the main application mode is to assist humans’ work and help to undertake tasks that humans physically are not able to do (e.g. drones, demining robots) or robots can do in a more precise way than humans (e.g. robot-assisted surgery). Thus, the emerging story of “robots are moving out of cages” might be more about collaboration between humans and robots, and much less about the displacement of humans caused by the deployment of robots.

Secondly, the global race in this new story is not as clear-cut as it is in the development, production, and deployment of industrial robots dominated by large corporations, because both the digital giants and the small start-ups are strongly concentrated in few regional ecosystems and the winner-takes-all logic rules the global landscape. In the background, however, there are two different forces at play. The growth of digital giants is driven by the specific nature of the information economy (Shapiro and Varian 1998), the increasing share and return of intangible assets (Haskel and Westlake 2017), the platform-based business model (Parker, Van Alstyne, and Paul Choudary 2016), and the network effects (Tirole 2017), and all these together pave the way for the “rise of superstar firms” (Autor et al. 2020). On the other hand, the growth and scaling of start-ups is driven by the peculiar investment model of venture capital as it attracts many but disproportionately finances only a few promising start-ups that are clustered around large cities in some regional ecosystems, leading to the “rise of the global startup cities” (Florida and Mellander 2016; Startup 2020). Yet, it is safe to assume that the competitive edge is at those countries and regions where digital giants are active in robotics development and service robots manufacturing and are able to leverage their products at global scale, and at the same time where there are vibrant robotics startup ecosystems with wide scaling options. Since talents are rare resources, however, the digital giants are often more interested in harnessing the start-up talent pool (hence the word “acqui-hiring”) than their innovative products or services (Coyle and Polsky 2013).

Policy challenges and opportunities for Europe

Taking into account its specific features, the new story poses at least three major policy challenges for Europe:

• First, while it is true that the winner-takes-all dynamics rules the landscape, particularly in robotics development and robot manufacturing, the deployment of robots in services and small-scale industrial settings is highly local in nature and that paves the way for regionally targeted policy measures. In Europe, there are many such policies in place, e.g. Industry 4.0 and Smart Specialization, and their increased focus on robotics is vital to Europe’s success in the emerging story.

• The second challenge is the scaling of robotics start-ups. Europe traditionally lags behind in the global start-up scene, though Europe does not lack innovative ideas and start-ups, it is the weak risk capital funding and less dynamic ecosystems that prevent the start-ups from growing. Creating and nurturing thriving ecosystems for robotics start-ups, where knowledge (universities and research), entrepreneurial talents (start-ups), finance (venture capital), and supportive institutions (accelerators, co-working spaces, etc.) can meet and reinforce each other’s effects is critical to improve Europe’s competitive position.

• The third policy challenge is to increase the involvement of the public sector. In contrast to the old story of industrial robots, in the case of service robots the public sector could play an active role on the supply side by exploring new application areas, as well as on the demand side by implementing the robots (e.g. health, transport and defense). Since in Europe the public sector has a stronger role than in many of its global counterparts, this challenge offers an opportunity for it to improve its competitive position.

Conclusion

Policy making thus needs to begin with an adequate identification of the challenges, particularly with regard to making a clear distinction between the two basic stories – “robots in the cages” and “robots are moving out of the cages” – because they are very different in terms of their driving forces, competition dynamics and geographical patterns. This distinction is also critical because of the strange contradiction in the current literature, as the vast majority of books and studies confuse robotics industry with Artificial Intelligence technologies and talk about (artificially intelligent) robots that at present barely exist, while empirical studies deal with the reality of industrial robots, although the latter are not new in many respects.

The old story of industrial robots is about a handful of economies, more precisely, the large companies of these countries, which operate in the few industries where industrial robots have become the manufacturing standard, and as their value chains are restructured, this is the story partly about the countries in which automated manufacturing processes are being relocated. Hence, the competitive advantage rests with those countries which develop, produce, and use industrial robots, while economies which only deploy industrial robots lag behind in the race.

In contrast, in the emerging story of more autonomous robots, competition becomes more global with more competitors, and alongside large economies, small countries and regional ecosystems can also achieve a strong competitive position through specialization in robotics development and/or robot manufacturing. In this case those countries and regions with both the digital giants and the small start-ups have the competitive edge, as those lacking one of these elements are left behind. However, partly due to the lack of reliable data, the spatial diffusion and uptake of service and collaborative robots remains a largely uncharted territory to this day.

Acknowledgments

The author wishes to thank Petros Gkotsis, with whom he made a previous study on robotisation and to whom the author is grateful for retrieving and analyzing of the robotics related patents from PATSTAT. He is also grateful to the anonymous reviewers for their extensive and very helpful comments and suggestions in finalizing the manuscript. All errors and omissions are the author’s own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Pallas Athéné Domus Meriti Alapítvány (PADME, Hungary).