GITAI Founding Story
Founding Story of Space Robotics Startup GITAI
Introduction
I am Sho Nakanose, Founder & CEO of the space robotics startup GITAI.
GITAI manufactures space robots that can provide on-orbit servicing and build infrastructure on the moon’s surface, with the goal of reducing the cost of working in space by one hundredth.
GITAI was originally founded by me in Japan, but we have since moved our headquarters and almost all of our functions to the U.S. Myself and other members of the management team have obtained permanent residency status, making us US Persons under U.S. federal law, and we continue to challenge ourselves daily in the United States.
Recently, GITAI raised US$45M in funding last year and received an order from DARPA (Defense Advanced Research Projects Agency).
In addition, GITAI successfully demonstrated its technology outside the International Space Station in March this year. I believe it was a truly challenging task for a space robot, with a manufacturing cost less than one-thousandth (compared to NASA’s robotic arm), which was developed entirely in-house, including mechanisms, electrical infrastructure, and software, to actually perform all complex tasks in space during its first demonstration outside the spacecraft.
We are just at the beginning of our journey as a U.S. space company, but even so, there have been many challenges along the way.
When I built a robot prototype with no background, when I founded GITAI by myself and entered the space industry, when I decided to develop all space robots in-house, and when I decided to conduct a space demonstration, many people told me that it was impossible.
I have faced many unreasonable challenges repeatedly in business development and startup funding in the space industry. I have heard unfounded bad gossip and have been told many heartless things by people who believed the gossip. Many times, I faced situations where the company might go out of business.
GITAI is still a startup that faces challenges, but the reason we have been able to overcome many difficulties and come this far is because of the people who have bowed their heads for GITAI, made passionate presentations for GITAI, and acted on behalf of GITAI.
Personally, I have never written anything like a blog before because I have been focusing solely on the management, business, and development of GITAI, believing that my role is first and foremost to produce results.
However, I have received so many questions about the “startup period” from people who usually support GITAI and who have been supporting GITAI. They ask how an entrepreneur with no background in space or robotics decided to start a startup developing space robots by himself.
Since GITAI has arrived at the space robotics business through various trial and error processes, I have only been able to answer this question in a very straightforward manner.
However, now that GITAI has succeeded in the demonstration outside the spacecraft and is at the starting point as a space company, and is about to move its base to the U.S. and take on challenges in the U.S., I would like to write about how GITAI was born and what criteria are used in decision-making for those who have been supporting GITAI behind the scenes.
It is quite long, but I hope you will bear with me until the end.
Startups Follow the Power Law
I have been vaguely interested in entrepreneurship since I was a student, and after leaving IBM Japan, where I worked as a new graduate, I started my first business in India. The reasons for starting a business in India instead of Japan were: 1) India’s smartphone market was the fastest growing in the world at the time I started my business in 2013, growing at 163% per year, 2) there were several services that I thought I could enter, even with time machine management, to develop Indian market versions of services already existing in the US, and 3) I thought that services for foreign companies operating in India (for example, at the time of 2013, approximately 1,000 Japanese companies had established Indian subsidiaries) would be suitable for building a foothold in India as a small business. The first company was established with no investment from investors and only self-financing, so I spent the first year doing only contracted development. From the second year, we used the profits from contracted development to develop and release our own web services and smartphone applications. After running the first company for about two and a half years, I sold the business. I then returned to Tokyo from India and was looking for my next challenge.
I have loved science fiction novels and anime since I was a child, and as I was regularly researching science fiction-like technologies, I learned about startups that solve major global problems by putting cutting-edge technologies to practical use (so-called Deep Tech or Hard Tech startups), and I wanted to take on this challenge myself.
*A startup is a company that is specifically designed for short-term rapid growth. Please refer to the Y Combinator founder’s blog, “Startup = Growth” for a detailed explanation of what a startup is.
Deep Tech/Hard Tech startups are companies that use cutting-edge technology to solve problems and are designed to grow rapidly in a short period of time.
From the time I decided to take on the challenge of a Deep Tech startup, to the founding of GITAI, to the management of GITAI to the present, my most important decision-making criterion has been that “startups follow the power law”.
The power law in startups means, very roughly, that a very small number of variables have a large impact on the outcome of things. In the case of Venture Capital, for example, the returns on a handful of the startups they have funded significantly exceed the sum of the returns on all other startups.
*Please read Peter Thiel’s “Zero to One” for more information on the power law in startups.
We also believe that the success factors of a startup also follow the power law. In other words, 70% of a startup’s success is determined by the combination of just three variables: market, product/technology, and timing of entry.
Startup success is 70% determined by the market (where), product/technology (what), and timing of entry (when)
Many great entrepreneurs and investors have talked about the importance of market selection and timing, so I don’t think I need to elaborate here.
Whether your company’s sales or market share ranks #1 or #100 in your industry is determined by corporate efforts, but whether the market capitalization of the #1 company in a particular industry is 10 billion yen or 100 trillion yen is determined by the market. Also, for almost identical products and businesses (from search engines to social networking services to rockets), just a few years’ difference in the timing of market entry can make a significant difference in subsequent success or failure.
Vision is often highlighted as a success factor for startups, but we believe that a great vision does not make a great startup. Instead, a rapidly growing market and business will create a great Founder/CEO, who will go on to have a great vision.
In the first place, I think that startups are often described in terms of success factors and advantages in extremely simplified cause-and-effect relationships and stories that eliminate complexity. And in fact, startups also tend to tell their own appealing stories for fundraising and media purposes according to a narrative that can be somewhat accepted by the layperson, so these overly simplified cause-and-effect relationships and stories tend to perpetuate themselves. This seems to be especially true for DeepTech startups, which have a high degree of technical and market expertise.
However, the founders of startups need to develop hypotheses of cause-effect relationships/stories based on “insights” such as market changes, challenges faced by actual potential customers, technology bottlenecks and issues, and extensive and deep knowledge of timing, rather than overly simplified cause-effect relationships/stories that eliminate such complexity.
In startups, where the combination of three variables — market (where), product/technology (what), and timing of entry (when) — determines 70% of success, there is a huge element of luck over which we have no control. However, I believe that the probability of success varies greatly depending on whether you leave everything to luck from the beginning or whether you have tested hypotheses about the market, products, technology, and timing of entry to the point where it is just luck.
For a software company like the one I founded as my first company, I believe it is possible to flexibly change the market and products through trial and error and pivots, as exemplified by the Lean Startup method. However, deep tech startups tend to take time and money to develop and commercialize their technology, and it is difficult to conduct trial-and-error and pivots later because assembling a highly specialized team of human resources is necessary. We believe that deep tech startups should only begin in earnest after determining the market, product/technology, and timing of entry.
The market (where), product/technology (what), and timing of entry (when), which affect 70% of a startup’s success or failure and are difficult to change later (especially in the case of a deep-tech startup), should be selected carefully and thoroughly with resources. Once the decision is made, the day-to-day business decisions (how) should be made quickly through a trial-and-error process.
In addition, the serial entrepreneurs I respect in the U.S., who have commercialized cutting-edge technologies and turned them into major businesses, are clearly not only driven by luck, but also by their ability to identify when the technology will be commercialized (technologies whose bottleneck is development rather than research and have not yet been commercialized) and when the market will change drastically or grow rapidly.
I read various books and articles about the market and the latest technologies, but I realized that knowledge obtained from books and articles was not enough to gain “insight” into market changes, issues faced by actual potential customers, and in-depth knowledge about the technologies.
Therefore, I decided to develop my own prototype of the technology I was interested in to learn about the technology, and then to interview potential customers in the market with that prototype.
Is today the timing for the interface of devices (PCs, smartphones, etc.) that augment human brain functions to transition from 2-D to 3-D?
My initial focus was on VR/MR technology. At the time (2015), VR/MR technology, as represented by Microsoft’s HoloLens, Oculus Rift, and HTC VIVE, was attracting a lot of attention and was at the center of a startup/investment boom.
Personally, through technologies such as VR/MR, I believed that the time had come for the interface of devices (PCs, smartphones, etc.) that augment human brain functions to transition from the traditional 2-D display to 3-D.
While using various VR/MR devices as a user, I first created an application for Web3D and smartphone VR in order to learn more about the technology.
Although the development itself was very interesting, I personally found smartphone VR to be a very limited experience, including image quality and frame rate, and I did not think it would become widely used as consumer content.
Next, we developed Windows applications for high-end VR devices like the Oculus Rift DK2 and HTC Vive.
Developing prototypes for high-end VR devices was very interesting, but high-end VR devices were very large and heavy, so I didn’t think they would be something that ordinary non-developer users could work or live with.
At this point, I was convinced that VR/MR devices would not expand as a consumer market unless they offered performance (resolution and frame rate) better than high-end VR devices and were at least as lightweight and small as sunglasses. However, to achieve rapid short-term growth as a startup, the market must grow rapidly within two to three years of entry. Therefore, we set out to verify whether VR/MR devices would be “as lightweight and small as sunglasses while guaranteeing performance better than high-end VR devices” within a few years.
The biggest reason for the large size and weight of VR/MR devices was that CG rendering (drawing process) is performed on the VR/MR device side (including the PC connected to the device). This heavy rendering process requires a high-performance GPU and a lot of power, making the device large and heavy, and requiring a wired, external high-end PC.
We thought that there were two major patterns to resolve this situation and make VR/MR devices “as lightweight and small as sunglasses while ensuring performance that exceeds that of high-end VR devices:
(1) Ultra-high-performance, ultra-power-efficient, and ultra-small GPUs will be developed and marketed within a few years
The most obvious pattern for an extension of the current VR/MR device structure is that an ultra-high-performance, ultra-power-saving, ultra-compact GPU will be developed and sold within a few years, and the VR/MR device itself will be “as light and small as sunglasses while ensuring performance better than high-end VR devices. However, although GPUs were in demand in various industries and R&D was actively conducted, the improvement in GPU performance per unit mass was not exponential but rather linear. Therefore, we concluded that it would take at least 10 years to realize a GPU that could meet the required performance, and it was unlikely to happen within a few years.
(2) VR/MR devices that perform rendering in the cloud and display the resulting rendered image on the device side via streaming will be developed and sold within a few years
The next pattern I thought of was that if all rendering processing could be done in the cloud and the resulting rendered images could be streamed to the device side, instead of doing the heavy rendering processing on the device side, the device functions would be limited to input (hand tracking, HMD tracking, etc.) and output, and it could be “as light and small as sunglasses while ensuring performance that exceeds that of high-end VR devices.”
Personally, I believe that consumer VR/MR devices realized in this pattern in the future will be more likely to form the consumer VR/MR market in the long term.
The technological bottleneck in this pattern is the streaming technology and network infrastructure technology required for streaming the rendering results to the device side. In particular, the amount of video (movie) data that is the result of rendering is extremely large, and VR/MR devices can cause the phenomenon of VR sickness if there is even a small delay. Therefore, “streaming technology and network infrastructure technology to enable low latency transmission of large volume video data” were necessary. Even if streaming technology improves, video compression formats (h.264, etc.) are improved to be more compression efficient, and software delays in video acquisition, video encoding/decoding, etc. are reduced, we believed that the biggest bottlenecks would eventually be the communication speed (bps) and latency of the network infrastructure itself in order to transmit and receive large volumes of data such as high-resolution video data with low latency.
It was at this time that 5G was expected.
During the period of GITAI’s founding, from 2016 to 2018, 5G mobile networks were attracting a great deal of attention, and telecommunications carriers were claiming performance that would enable them to send and receive tens of GB of data at 1 m/sec. And although several startups were launched with the basic premise that mobile networks capable of transmitting and receiving tens of GB of data at 1 m/sec (5G) would spread in a short period of time, as far as I could tell, no startups were actually conducting technical verification or research to see if 5G would meet the required performance.
Therefore, we decided to verify by ourselves whether this 5G would meet the following two requirements: 1) it would have the performance to send and receive several tens of GB of data at 1 m/sec, and 2) it would spread throughout Japan within a few years.
The two main things we did to verify the technology were (a) to develop our own communication technology for video transmission and (b) to cooperate with the R&D team in charge of 5G at a certain telecommunications carrier to investigate the actual performance of 5G.
First, regarding (a), since I am not capable of such advanced development as developing communication technology for video transmission, I asked an engineer in charge of developing communication and video transmission technology (*the person who was a development leader at the company I founded before GITAI) to join us in February 2017, and we will be able to develop our own software technology to transmit and receive video data from 360-degree cameras with low latency and low capacity (reduction of software latency, reduction of data capacity by image difference extraction processing, etc.), and middleware and communication technology to transmit and receive video data via wifi/mobile networks (over NAT system, etc.).
(b), After consulting with the 5G R&D department of a certain telecommunications carrier, we were able to conclude a joint research agreement and conduct a joint demonstration using an actual 5G base station. While listening to various opinions of the 5G R&D department, we deepened our knowledge of 5G by connecting an actual 5G base station with a commercially available home wifi router and conducting communication tests for video transmission.
As a result, I learned that the claims made by telecommunication carriers that they can transmit several tens of GB of data at 1 m/sec are only theoretical physical values between one base station and one terminal, and that the actual mobile network will have various bottlenecks, so the values will not be as theoretical as the theoretical values. 5G base stations need to be constructed more densely than 4G base stations, and it is highly unlikely that 5G will spread throughout Japan or the world in a short period of 2–3 years due to the problems of securing base station locations and manpower requirements.
Therefore, I turned our attention to another technology and market.
Is it time for general-purpose robots, which extend physical capabilities in the same way that computers extend human brain function, to become a viable business?
Next, I focused on general-purpose robot technology, familiar from science fiction cartoons and movies. The definition of a general-purpose robot varies, but in this context, it refers to a robot that can autonomously perform several different tasks, like a human, as opposed to specialized robots that repeat a single task, like industrial robots. Personally, my favorite sci-fi anime and movies mainly involve robots, and it has always been one of my favorite technologies. Also, while I was attracted to VR/MR for its potential to shift the interface dimension from 2-D to 3-D, I was drawn to general-purpose robots because they can extend human physical capabilities.
At the time (2016–2017), AI was thriving with developments in deep learning, and the champions of human brain function extensions (PCs, smartphones) (Google, Microsoft, etc.) were competing to create replacements for brain function (AI). However, even then, it had already become a saturated technology and market, with large companies heavily entering and investing, which I thought was not a good match for a startup. On the other hand, in the case of expanding physical capabilities, another human ability, specialized robots had become widespread, but general-purpose robots had not yet been commercialized and competition was not fierce, presenting more opportunities for startups.
The market size for specialized robots (industrial robots) was also significant. This is because, when examining the history and transition of technology and market development for computers, which extend the human brain, there was a pattern where specialized technology (products specialized in calculation capabilities, a part of human brain function, i.e., calculators) first formed the market, and then general-purpose types (general-purpose computers: PCs and smartphones) swallowed the specialized market to become even larger.
If the expansion of human physical capabilities follows the same pattern, specialized robots (industrial robots) currently in widespread use are like calculators in brain function expansion tools, and I thought that general-purpose robots could potentially form an even larger market in the future.
So, I read various books and media articles on robot technology and the market, but I felt that I had only gained superficial knowledge. Therefore, I decided to try building a prototype myself.
However, I had never built any hardware, let alone robot technology, so I started with serial communication from Unity to Arduino for LED blinking.
Once the LED blinking was completed, the next step was to control the servo motors, which gradually made the robot look more and more like a robot. I also synchronized the robot with the hands and arms that can be detected by Leap Motion.
Starting with Prototype No.4, I took on the challenge of making it wireless in a local environment.
I had been developing the robot by myself until the fourth unit, but I realized that I could not develop the communication part for operating the robot via the Internet. So, I consulted with an engineer who had been the leader of development at a company I had previously started and asked him to help me develop communication technology for remote control via Wi-Fi and software technology to reduce data volume and delays in video transmission.
In parallel with the development of the humanoid teleoperated robot described above, I personally wanted to deepen my understanding of autonomous control, so I also tried to develop basic autonomous control.
I started with an inverted pendulum as my first challenge.
I also built a prototype that automates sorting based on supervised data.
I used OpenCV for image analysis, Blob Detection for object detection, SVM for pattern recognition, and Arduino for hardware control. The development environment is Processing.
Can a general-purpose robot replace human labor?
In parallel with developing a prototype that could at least convey what I intended, I also proceeded with researching and selecting markets where a general-purpose robot could be a viable solution or business.
Although the space market was a candidate from the beginning, I began our research and interviews with terrestrial markets that seemed to have demand, such as online conferences, disaster relief, telemedicine, and power plant inspections. Thanks to the Ministry of Economy, Trade and Industry (METI), we were able to visit the Fukushima nuclear power plant, site of the 2011 accident, to explore whether our robot technology could be used for debris removal. We were even allowed to wear protective clothing and go directly under the power plant where the nuclear fuel rods were located. Many people were very willing to assist us in our market research. I am truly grateful for their cooperation.
I researched specific problems that general-purpose robots could solve across various markets, and a common theme in most terrestrial markets was the need for general-purpose robots as a “solution to reduce the cost of human labor.” Essentially, there was a desire for a general-purpose robot that could perform tasks better and more cost-effectively than human workers.
At the time, Boston Dynamics’ video of a humanoid robot doing backflips was widely discussed, and combined with expectations for AI technologies like deep learning, there was a belief that “general-purpose robots with human-like performance could be realized within a few years and, if mass-produced, they would cost less than human labor.” Many people in the market, as well as startups and investors, held this belief.
However, as I became more familiar with general-purpose robotics technology, I began to realize there was a significant gap between the market’s expectations for general-purpose robots and the actual performance and cost of the technology.
The Boston Dynamics video of a backflipping humanoid robot is certainly impressive. However, a successful demonstration in a controlled environment like a laboratory does not immediately imply similar product performance. There is a significant “development and manufacturing” barrier between what is possible in R&D and what can be achieved in mass-produced products, and I believe this difficulty is often greatly underestimated.
AI, as exemplified by deep learning, has led to significant performance improvements in processing 2D information such as text, images, and videos. However, the real, physical 3D world contains too many variables, and the difficulty of “tasks,” especially those involving physical interaction, remains very high and has not seen comparable performance improvements.
Also, in most terrestrial markets, “general-purpose robots as a labor force that performs better and costs less than human workers” were most expected to perform “general-purpose tasks”. In reality, however, even several-million dollar general-purpose robots had previously found it extremely difficult to perform multiple different tasks autonomously, like humans, and had not reached at least the performance of human workers at all.
Even if a general-purpose robot with human-like performance were realized, to “make the cost of a general-purpose robot cheaper than human labor,” it would be necessary to (1) reduce the cost per robot through mass production, (2) ensure stable operation as a product for many years, and (3) minimize human intervention with full or semi-autonomy.
However, hardware mass production is extremely challenging and expensive, and the working performance of a mass-produced machine at a reduced manufacturing cost per unit will almost certainly be significantly lower than that of a one-off prototype costing several million dollars.
Initially, we feel that the difficulty of “development and manufacturing,” including hardware mass production, is too often underestimated by some startups and investors. Additionally, the relative advantages in research and technology that startups claim to possess rarely address the development and manufacturing bottlenecks.
Based on the above, I concluded that general-purpose robots as a “solution to reduce the cost of human labor” still face many technological bottlenecks and are unlikely to replace human labor as expected in many terrestrial markets, at least within the next few years.
Can general-purpose robots become a business with high unit prices as unique products for government sectors in space and defense markets?
Among the many candidate markets, we considered the space market last. The reason why the space market came last was because it seemed too distant to me personally, and I had no realistic image of a market that a startup could enter.
It seemed very counterintuitive to enter the space market as a startup, which until then had been only a subject of personal interest or hobby. However, theoretically, I originally considered the space and defense markets as the most promising candidate markets from the following four perspectives:
(1) General-purpose robots, which are tools for extending physical capabilities, have the same potential as computers, which are tools for extending brain functions, to become a business starting from the high unit prices of single products for the government in the space/defense markets
When we look at the history of computer commercialization as a tool for extending brain functions, we find that computers also started as high-unit-price, one-of-a-kind products for the military and space governments. Their performance was very limited and their cost was very high initially. As performance improved and prices declined, the products gradually transitioned to corporate products, and finally to mass-produced consumer products such as personal computers and smartphones.
I believed that general-purpose robots, which are tools for extending physical capabilities, would follow the same pattern as computers, which are tools for extending brain functions. In other words, I hypothesized that general-purpose robots would start as high-unit-price, one-off products for the government in the space and defense sectors.
Next, we looked into successful examples of startups that have commercialized high-unit-cost, one-of-a-kind products for the government. The company that caught my attention the most was Palantir Inc. of the United States.
Palantir is a startup that provides data analysis and security services to the government. Its business model is not based on selling a single in-house software product but rather a government-oriented business similar to selling very high-unit-price, one-off solutions. About half of its 2018 revenue (US$595.4M) came from government customers (US Military, DoD, FBI, CIA, etc.). In addition, the top 20 customers accounted for 72.9% of sales.
Of particular note is that the average cost per customer for Palantir’s top 20 customers has increased approximately 18-fold (US$1.2M to US$21.7M) over the eight years from 2010 to 2018, and sales have grown by increasing the cost per customer, not the customer volume itself.
The formula of unit price x volume = sales remains the same for any startup in any market or business model. Most robot startups of the time were also trying to increase “volume” by replacing many human workers and thus increase sales. However, as mentioned above, the technology of general-purpose robots is still in its infancy, and we believed that it would be too technically unrealistic to achieve performance sufficient to replace human workers, lower the unit price per unit, and ensure quality and reliability for mass production in large numbers. At the very least, we believe that startups are limited to one challenge of a difficulty level that requires breakthroughs.
In this respect, Palantir’s business model was not to significantly increase the “quantity” of customers, but to significantly increase the “unit price” of customers, since the business was for the government and the number of expected customers was limited.
This point is at least technically very advantageous in terms of commercializing general-purpose robots. In other words, it was a market and business model that did not require the extremely difficult technological challenge of “lowering the unit price per unit while achieving performance sufficient to replace humans, and mass-producing the number of units after ensuring quality and reliability.
(2) The world is heading toward destabilization and a bipolar structure, and the space and defense markets may grow
The global situation at that time (2016–2017) was characterized by the United States withdrawing from its role as the world’s police, increasing instability in the Middle East, and China’s rapid rise to power.
Therefore, if the bipolar structure between the U.S. and China were to accelerate further in the future, the emergence of a powerful and ambitious rival would likely lead to an increase in the defense space budget and market growth, as seen during the previous Cold War era.
(3) Humans, who have expanded horizontally over the past 5 million years, may become a vertically expanding species in the next few decades
Like other organisms, humans have an instinct encoded in their genes to explore unknown places for reproduction. Having originated on the African continent and expanded horizontally over the past 5 million years, we may now be at a critical turning point in our history to see if we will become a species that expands vertically. In other words, I believe that the reason why the world’s top entrepreneurs and millionaires have recently been targeting space is not merely a trend but an irreversible movement, driven by fundamental desires such as curiosity and instinct.
(4) SpaceX’s transportation revolution may shift the bottleneck in the space industry from transportation costs to labor costs
The initial bottleneck in the vertical expansion of mankind was the “cost of transportation to space.” At that time, SpaceX had already reduced the cost per kilogram of transportation to space to a few tenths of what it was a decade earlier, initiating a transportation revolution. I believed that once the “cost of transportation to space,” which had been the primary bottleneck in the space industry, was dramatically reduced, making it feasible to transport goods and people into space, the next bottleneck would become the “cost of labor in space,” and demand for this service would start to emerge.
Although I theoretically saw the potential in the above four points, I did not know anyone in the space industry and had no idea how to conduct market research and interviews. At that time, I happened to have an opportunity to stay at NASA’s Ames Research Center, where I conducted interviews with people in the U.S. space industry.
Thanks to the opportunity to talk with people in the space industry for the first time, I gained a deeper understanding of the market and gradually became more convinced of the above four hypotheses. At the very least, I thought, “It is highly likely that the need for a means of work in the space industry will emerge within a few years, which will be a major challenge for the vertical expansion of humanity.”
Therefore, we looked into the existing means of working in space and found that there are two main methods:
(1) Human astronauts
One way to perform work in space is for human astronauts to do it. As mentioned above, humans are the primary workforce for general-purpose tasks in most industries on the ground. However, there are many safety issues in space due to the vacuum, extreme temperature changes, and radiation levels that are hundreds of times stronger than on the ground, and the cost of safely transporting humans into space is approximately several hundred times higher than the cost of transporting the same weight of an object into space. For example, NASA astronauts cost US$130K per hour. Also, due to the effects of cosmic radiation, etc. (*unless an environment that can block most of the cosmic radiation, such as an underground base on the moon, can be constructed), the total period of stay in space is limited to a few years.
Although the workforce of human astronauts may increase in the future through commercial astronauts, we do not expect a significant reduction in work costs due to safety issues and restrictions, and it is unlikely that human astronauts will be able to perform hazardous work.
(2) Robots in the space industry
The second way to work in space is to use robots in the space industry. In the space industry, robots such as the Kibo Arm of the International Space Station (ISS), the Canadarm, and the robotic arm attached to the Mars Rover have been developed and used for various missions for a long time.
However, compared to general-purpose robots on the ground, robots in the space industry are limited in performance, especially in autonomous control (due to constraints such as the need to ensure safety and reliability in an environment outside the spacecraft), and above all, they are overwhelmingly expensive and have a long lead time. For example, one Canadarm that is still installed and operating on the International Space Station cost about US$1.2B.
So, when I investigated why robots and spacecraft in the space industry have high costs and long lead times, I found that the space industry is dominated by waterfall development, which is based on the assumption that all components and parts are procured from various space component suppliers.
The space industry up to now has had extremely few opportunities to actually launch into space, and because most missions were led by space agencies operated by public taxpayers, it was difficult to take the risk of failure, and the development method prioritized “no failure,” even if it took a lot of time and money. From the standpoint of suppliers of spacecraft components, spacecraft are basically one-of-a-kind items and their components are sold only occasionally, so suppliers of spacecraft components have to make large margins on parts. This also requires a long lead time because of the build-to-order production system. In addition, spacecraft used for space agency missions are subject to an important process called safety review to ensure safety, which requires certification of safety at the raw material level and specification changes as necessary, each time requiring the suppliers of space components to respond. Such a situation had accumulated for hundreds of components. In addition, waterfall development required tremendous man-hours for defining development requirements, process management, and progress management, and it was a situation in which several thousand people were responsible for the development of a single spacecraft project. As a result, the development cost of a single spacecraft was several hundred million to several billion dollars, and the lead time was 10 to 20 years. Space industry startups tried to reduce manufacturing costs by using as many terrestrial consumer products as possible for spacecraft components and by developing with as few people as possible. However, in the end, it seemed difficult to significantly reduce costs and lead times because key components are procured from individual space component suppliers and spacecraft are developed by waterfall development.
From the above two investigations, we learned that there are existing means of work in the space industry, but due to either cost, lead time, safety, or other issues, we did not believe that they would be the primary means of work as a solution even if demand for such means of work were to emerge in space.
At this point, I roughly thought, “Let’s bring the technology of general-purpose robots on the ground and agile development methods to the space industry and take on the challenge of creating space robots that are much more versatile, inexpensive, and safe than the existing robots in the space industry.”
How will GITAI, a startup, develop a general-purpose robot for space use?
Through the above process, we have determined to some extent the “market (where), product/technology (what), and timing of entry (when)” that will have the greatest impact on the success of the startup.
Next, I roughly considered the “how” of how GITAI should develop a general-purpose robot for the space market and what steps should be taken to develop and commercialize it. In particular, I considered the following four long-term and important policies:
(1) Perform vertical integration and agile development through in-house production to reduce costs and lead times, and improve performance.
As mentioned above, most companies in the space industry were developing spacecraft through waterfall development by procuring components from various suppliers.
However, SpaceX, which had already achieved the world’s highest performance and lowest price for a rocket at the time, had dramatically reduced intermediate costs and lead times by producing all components in-house and using as few suppliers as possible, and had rapidly improved performance through agile development (a development method that improves performance in a short period by repeating trial and error of making, testing, and breaking products) made possible by in-house production of all components.
Therefore, GITAI also decided to produce all the necessary components for a general-purpose space robot in-house and to proceed with development in an agile manner to lower costs and lead times over the long term and improve performance.
(2) The first step was to develop a remote-controlled robot to raise the level of hardware, and then develop autonomous control software to make it an autonomous robot
The entrepreneurial method of startups is to grow by making progress and raising funds approximately every 18 months, but the in-house development of a general-purpose space robot is likely to take a very long time, so I thought that managing progress along the way would be a major issue.
In order to develop a robot that can perform general-purpose tasks in an environment as harsh as space and without network infrastructure, it is necessary to achieve not only the robot’s work performance but also exospace measures as a spacecraft, safety, and reliability as a product. To develop such a robot in-house, we needed to develop all of the mechanisms, electrical infrastructure, and software (including autonomous control software) in-house, which we thought was impossible to do in a year and a half.
In addition, the performance of the autonomous control software would be bottlenecked by the upper limit of hardware performance, so if the hardware and autonomous control software were developed simultaneously, it would not be possible to develop a robot with a competitive advantage (especially in terms of work performance) in the limited amount of time available, and as a startup, we would risk not being able to make progress (= not being able to raise funds).
Therefore, I decided to abandon the development of autonomous control software for the time being and focus on the development of a general-purpose remote-controlled robot to build a technological competitive advantage, and while enabling demonstrations for business development and fundraising, I decided to proceed with in-house development of various components in the background. I decided to begin development of autonomous control software when the performance of the hardware developed in-house had been sufficiently improved, and to go through the steps of making the robot autonomous at a later stage.
(3) Aiming to provide robotics as a service (RaaS), rather than selling the robots themselves
In the U.S. space industry, when NASA procures something from a private company, it is shifting from a procurement method known as parts procurement (example: contract to purchase the rocket itself) to a method called service procurement (example: contract to purchase launch services using the rocket, not the rocket itself).
In this service procurement model, NASA minimizes risk by fixing the price and making payments at each achieved milestone, while private companies increase their profit margins by lowering costs.
SpaceX has also made great strides by maximizing the cost advantages of in-house production through this service procurement contract (service to transport supplies to the ISS).
In the long term, GITAI has decided not to sell robots itself but to provide robotics as a service in space. However, since in-house development of a general-purpose robot for space use is extremely time-consuming, GITAI decided to adjust its business model in the following three steps, roughly corresponding to the degree of progress in in-house development:
(ⅱ) Phase of in-house development of robot arms: Sales of robot arms as a subcontractor.
(ⅲ) Phase in which the entire general-purpose robot for space use, including the moving mechanism, is developed in-house: Robotics as a Service (RaaS) is provided as a prime contractor.
(ⅰ) Phase of in-house development of components: Contracted development of space robots.
(ⅱ) Phase of in-house development of robot arms: Sales of robot arms as a subcontractor.
(ⅲ) Phase in which the entire general-purpose robot for space use, including the moving mechanism, is developed in-house: Robotics as a Service (RaaS) is provided as a prime contractor.
(4) Maximize the output generated by agile development for business development and fundraising
Because waterfall development was the mainstream in the space industry, it was difficult to show actual products and demonstrations, and business development and fundraising in the space industry were mainly conducted using computer graphics (CG) images or videos and PowerPoint documents. This is still a common scene in the space industry, but to me, entering from other industries, it seemed very peculiar that business discussions about spacecraft costing several hundred million to several billion dollars were conducted using computer graphics images and PowerPoint documents, not actual products. I believed that even in an industry as specialized as space, potential customers and investors would want to see actual products and demonstrations (even if they were product prototypes or ground demonstrations), just as they do in other industries.
Therefore, GITAI decided to focus its business development and fundraising efforts on in-person robot demonstrations and demonstration videos, rather than relying on CG or PowerPoint documents.
This approach to business development and fundraising, centered on actual prototypes/products and demonstrations, was a particularly good match for GITAI’s agile, in-house development strategy, and was a major factor in making GITAI stand out in a space industry that was dominated by CG and PowerPoint documents.
Although we had already established a corporation and raised funds from Skyland Ventures during the technology and market validation described above, we decided to raise another million dollars to fund the hiring of engineers and the development of a general-purpose robot. We then raised about US$1.4M with ANRI as the lead investor.
Formation of a team of engineers for GITAI
Thanks to the investment we received, we now have a budget and have decided to create a team to develop a full-fledged general-purpose robot for space use.
I founded GITAI alone in July 2016, and in February 2017, an engineer responsible for developing communication and video transmission technology (who was the development leader at the company I founded before GITAI) joined GITAI. However, from then until April 2018, it was just me and the engineer, and we did not do any hiring at all.
I decided that we should not hire or build a team until I, as the Founder, determined the market (where), product/technology (what), timing of entry (when), and especially the development policy (how).
This is because we believe that what defines “excellent people” for a particular startup depends greatly on the startup’s business model, and business and development policies. For example, a high level of education and extensive work experience at a large company may be very attractive in the hiring process, but if a person who has been involved in supply chain management and quality control of waterfall development at a large company for many years moves to a startup that aims for in-house production and agile development, he/she may not be able to demonstrate his/her experience and skills effectively, which could lead to a mismatch. Of course, this varies from person to person, but we decided not to conduct hiring activities or build a team without first clarifying the definition of “excellent people” for GITAI.
Then, as mentioned above, after deciding on a combination of “market (where), product/technology (what), and timing of entry (when)” and a rough business and development policy for GITAI, we began to create a team of engineers for in-house and agile development of general-purpose robots for the space industry.
Our target was “researchers and engineers who are engaged in in-house and agile development of general-purpose robots (especially humanoid robots) in the terrestrial industry, not in space”.
Since we started recruiting in earnest, we have met some truly wonderful researchers and engineers, but in particular, the encounter with Dr. Kozuki (currently GITAI CTO) had the greatest impact on GITAI.
While researching various robots for recruitment, I found an amazing humanoid robot that was doing push-ups while sweating. It was a robot named Kengoro from the Information Systems Engineering Laboratory of the University of Tokyo. The University of Tokyo’s Information Systems Engineering Laboratory is a world-famous laboratory that has produced SCHAFT, a bipedal robot startup that was acquired by Google after winning first place in the preliminary round of the DARPA Robotics Challenge, and MUJIN, a robot that aims to make factories unmanned, and is building robots at a tremendous pace.
I immediately contacted Dr. Toyotaka Kozuki, who was the lead author of the paper based on this robot, and asked him to visit me in my small one-room office I was renting in Shibuya at the time. After that, I showed him a demo of the robot and discussed the above information with him until late at night for about three days, and he agreed to join GITAI. Thanks to Dr. Kozuki’s participation, GITAI has grown under a system in which Dr. Kozuki oversees the technical development and I oversee everything else.
Dr. Ryohei Ueda (currently VP of Software at GITAI), a former Software Engineer at Schaft (Google) and a senior member of Kozuki’s lab, also came to our office many times and after heated discussions, he decided to join GITAI.
Furthermore, I happened to hear that almost all of Google’s robotics divisions were going to be shut down due to a change of CFO at Google at this time, and that Schaft, which had developed bipedal robots as one of Google’s robotics divisions in the U.S. after its acquisition by Google, was also going to be disbanded. I thought this was a great opportunity and immediately contacted Dr. Yuto Nakanishi (currently GITAI Chief Robotics Officer), who was the Founder & CEO of Schaft, and asked him to come to my office for a visit. After many heated discussions late into the night, Dr. Nakanishi agreed to join GITAI.
After that, people from the University of Tokyo’s Information Systems Engineering Laboratory joined GITAI one after another, and they have been playing a central role in the in-house and agile development of GITAI’s general-purpose robots for space use. It has been 5 to 6 years since they joined GITAI, and none of them have quit yet; they continue to be the core of GITAI’s technology development.
The people who had been researching and developing general-purpose robots in the world’s best and most advanced environments, such as Schaft (Google) and the Information Systems Engineering Laboratory at the University of Tokyo, joined GITAI for a variety of reasons, but they all had one thing in particular in common: they were convinced that GITAI’s development and business plan was based on “actual technology and market insights,” not “an extremely simplified causal relationship and story that eliminates complexity.
Consumer general-purpose robots that could replace humans experience a boom every decade or so, attracting startup investment and media attention. I think such challenges and investments are great in and of themselves. Also, general-purpose robots that can eventually replace humans will surely become a reality. However, we need to calmly determine whether this is a boom based on the premise that technological and business bottlenecks have been resolved enough for realization at this time, or whether it is a boom based on the fact that most people are not aware that there are still significant technological and business bottlenecks requiring major breakthroughs.
During the period when GITAI was founded, there was also a boom in consumer-oriented general-purpose robots in the U.S. and Japan, with many startups launching and attracting investment from investors and media attention. Under such circumstances, GITAI was not taken seriously by most investors at the time, as it was competing in the exact opposite direction, focusing on unit price instead of volume, space instead of terrestrial, and government instead of consumer. (*Despite this situation, there were investors who believed in GITAI and invested in it, which is why GITAI is what it is today.)
However, people who had been researching and developing robots in the world’s best and most advanced environments, such as Schaft (Google) and the Information Systems Engineering Laboratory of the University of Tokyo, strongly resonated with us in that we avoided the extremely difficult technical challenge of “achieving performance sufficient to replace humans, lowering the unit price per robot, ensuring quality and reliability, and mass-producing the robot in large numbers”, and instead turned general-purpose robots into a business in the high-unit-price, one-off business for the government in the space and defense sectors.
And rather than me telling you why they joined GITAI, you can watch a video of Dr. Nakanishi, former Founder & CEO of Schaft and current Chief Robotics Officer of GITAI, talking about why he joined GITAI in his keynote speech at the World Robot Summit in 2021. It’s about an hour long, but I recommend you watch the whole thing because it’s very interesting.
With the establishment of GITAI’s engineering team, the challenge of developing a general-purpose robot for space use and in-house production of all necessary components began in earnest, leading to a successful demonstration inside the ISS in 2021 and a successful demonstration outside the ISS in March 2024. GITAI developed all components in-house to significantly reduce costs and lead times, and the space robot, whose performance was enhanced through agile development, has actually successfully completed all tasks in space.
Business development also began with a joint research contract with the Japan Aerospace Exploration Agency (JAXA), which led to orders from various Japanese commercial space companies such as TOYOTA, orders from various U.S. commercial space companies, and orders from DARPA.
This is the story of the founding of GITAI.
One point I would like to add is that the actual startup period was chaotic, with multiple hypotheses being tested simultaneously, so the steps were not neatly separated into “A is done, then B is done,” and so on. Therefore, the timeline is sometimes non-linear, and there are many more hypothesis tests that could not be fully described above.
With the cooperation of many people, various twists and turns and hypothesis testing led to the birth of the space robot startup GITAI.
About the management of GITAI
Let me also describe a little bit about how GITAI is currently managed based on the decision-making criteria.
As I mentioned in the founding phase, the decision-making criterion that I emphasize above all else in the management of GITAI is that “start-ups follow the power law.” This means that a very small number of variables will have a large impact on the outcome of things. When applied to the decision-making criteria of startup management, it means that in order to maximize the return on investment for greater results, we identify the few things that are most important and focus the majority of our resources on those few things.
For example, in the space and defense space market, we have positioned two of the most important markets, the on-orbit service market and the lunar infrastructure construction market, both of which are in the defense space domain, where the space robot market is expected to grow especially with the destabilization of the global situation. We have been concentrating our technological development resources on these markets since about two to three years before the needs for space robots and operations became apparent in each of these markets. Furthermore, in these two space domains, we have been developing the entire spacecraft to become a prime contractor that provides work agency services directly to NASA, the Space Force, etc. under service procurement contracts, rather than a supplier to any one company. For the lunar infrastructure construction market, we have been developing not only a robotic arm for work on the lunar surface, but also a lunar rover for movement on the lunar surface, to be manufactured in-house starting in 2021. For the on-orbit service market, we have been working since 2023 on in-house development of not only a robotic arm for work in space but also a satellite for movement in space.
The project from DARPA (Defense Advanced Research Projects Agency) awarded at the end of 2023 is a mission to build infrastructure on the Moon using GITAI’s lunar rover and robotic arm. This is a result of the timing of the demand from DARPA due to global instability and competition, and the timing of the provision of labor by the lunar rover and robotic arm, which GITAI has been developing in-house since 2021.
In parallel with the above, we have also been concentrating our resources in the United States. Originally, I thought that in order to reduce the cost of work in space by 1/100 over the long term, it would be an important step to gain the top share in the U.S. market, which is by far the largest market in the global space and defense market. However, many Japanese startups, not limited to DeepTech, have tried to enter the U.S. market, but at least as far as I have been able to find, there have been almost no successful cases of startups investing part of their resources in the U.S. by establishing subsidiaries or other means. This may be due in large part to the fact that in Japan there are many localized versions of businesses that are popular in the U.S., but I also felt that the seriousness and commitment of the management team to the U.S. market had a very large impact on this.
In addition, the U.S. space and defense space industry is subject to extremely strict laws and regulations, including overseas export controls, and in order to do business directly with U.S. government organizations such as NASA and the Space Force, it is necessary to meet various requirements, including the country of headquarters, development bases and suppliers, and corporate ownership. Therefore, even if a U.S. subsidiary of a non-U.S. company or a non-U.S. management team entered the U.S. space and defense space market (unless it was a U.S. subsidiary of a large company), they could hardly get involved in major projects.
Therefore, instead of dividing resources halfway between the U.S. and Japan, we chose to focus all major resources on the most important market, the U.S., to increase our chances of winning in the U.S. market.
As a result, as of the end of 2023, we have moved our headquarters and almost all of our functions, including technology and business development, from Tokyo to Los Angeles in the US. In addition, as of April 2024, all members of GITAI’s U.S. relocation target (25 in total), including all management team members who were originally in Japan, have all obtained U.S. working visas with the support of GITAI and immigration attorneys and have completed their relocation to the United States. Furthermore, all of GITAI’s CxOs, including myself, obtained US Permanent Resident status and became US Persons under US federal law.
GITAI also stopped all PR and recruiting activities in Japan about a year and a half ago and has thoroughly concentrated on business activities, recruiting, and PR activities in the United States. As a slight aside, sometimes investors tell me that DeepTech startups will sell on their own as long as they have excellent technology. However, I have never met such a DeepTech startup. Also, GITAI’s robots have (unfortunately) never sold on their own. I believe that the idea “if the technology is good enough, it will sell” is another prime example of an extremely simplistic cause-and-effect relationship/story that eliminates complexity.
For a DeepTech startup like GITAI, it is as important as hiring the best people and keeping them on board for the long term, developing technologies that give it an edge, developing products that meet performance, quality, safety, and reliability requirements by developing all components in-house, anticipating the complex space and defense market and making decisions to meet the various requirements of the U.S. space and defense market, lobbying and top selling to government agencies and commercial space companies to win missions are all equally important and equally challenging.
What GITAI’s business development and recruiting activities have in common is the importance of first thoroughly clarifying one’s target audience, creating the most attractive environment for that target audience in the market, and making this known, rather than the individual operational-level sales and recruiting activities themselves.
I won’t bore you with the details because it would be too long, but GITAI has been focusing on building such an environment in the U.S. space and defense space market, especially over the past year and a half. As a result, GITAI’s presence in the U.S. space and defense space market is rapidly increasing.
As an example of our recruiting activities, GITAI USA currently receives over 1,000 job applications per week via Direct Application (direct job applications via GITAI’s website or LinkedIn), not via agents or other means. Since mid-February of this year, the number of applications has never fallen below 1,000 per week, has recently surpassed 2,000 per week, and is continuing to grow. Currently, GITAI has a mixed team, with members transferred from Japan and members hired locally in the US.
In addition, business development in the space and defense market is an industry where lobbying and top sales by management are extremely important, and GITAI is currently building a network that allows us to have daily meetings with senior officials from NASA, the US Space Force and Air Force, and CxOs and executives from US private space companies.
Finally
I founded GITAI through a series of twists and turns because of my love of science fiction novels and anime, my love of technology, and my desire to do something that would make a significant contribution to humanity in my once-in-a-lifetime life.
However, the journey was truly unimaginably difficult and painful. There were many times when I thought that GITAI might go bankrupt. I still have to deal with hard and unreasonable things every day.
However, when I was about to leave the office in the middle of the night, exhausted by the hardships and unreasonable things, I was saved by the sight of the space robot that I had dreamed of in the early days of the company actually working, the engineers working hard on its development, and the engineers enjoying their experiments.
I am also thrilled every day when I actually receive an order for a mission from an organization that I admired during the startup period, or when we jointly conduct a space demonstration or have a business meeting.
GITAI is still a startup in the midst of challenges, and we are just at the starting point as a U.S. space company. There are still many hurdles to overcome, and we are struggling every day.
However, thanks to so many people, we are now able to take on challenges in the U.S. as a U.S. space startup from Japan.
I myself am excited every day despite the difficulties. I also still have exciting dreams.
We at GITAI will continue to make our dreams come true. Let us continue to challenge ourselves to dream big. Thank you for your continued support of GITAI!
GITAI Founder&CEO
Sho Nakanose
