Sergiu Iliev (SI):
I am an aerospace engineer and serial entrepreneur. After graduating from Imperial College London with a 1st class degree I have worked in research and aerospace engineering consulting.
Eric Langenskiöld (EL): I'm a Business Developer and I've met Sergiu from the first couple of days of the Antler startup program in Stockholm. Together we realised we would be a great team to start a business in the space sector.
Arshak Ulubabyan (AU): I've met Eric and Sergiu after they pitched and received funding for Spacept (back then it was called DeepSpace AI) during the Antler startup program. My background is in data science and software development. We decided to team up as equal founders as our aerospace, business and software development skills respectively are a great match for the challenges Spacept aims to solve.
All of us are interested in space and would like to eventually contribute to making humans multiplanetary.
EL: Spacept is a software platform with specialised computer vision systems that process earth observation satellite data of infrastructure to enable safer, more efficient and sustainable operation and maintenance.
All of us are interested in space and would like to eventually contribute to making humans multiplanetary.
SI: Our mission is in line with the #9 (Industry, Innovation, and Infrastructure) and #13 (Climate action) sustainable development goals of the United Nations. Our solution saves lives and eliminates CO2 emissions while also radically reducing time and cost for infrastructure inspection. Thanks to European (Copernicus) and American (Planet and Digital Globe) satellite data providers we can deliver our infrastructure inspection software analysis anywhere on the planet at a highly attractive price: just $5 per kilometer of infrastructure which is half the price of drone inspection and a quarter of the price of helicopter inspection.
Spacept was started in early 2019 by us three, now the team has 35 developers working on a part-time and volunteer basis taking an Open Innovation business model. The main office is in Stockholm and expanding to New York later in the year. Spacept raised $110k at the beginning of the year and we are currently raising $500k as part of our Seed round.
SI: Eric and I worked together with an AI expert to identify business opportunities in the use of earth observation data commonly available from private companies as part of a ‘team challenge’ in the Antler startup program. We knew about the increasing occurrence in forest fires in California and one of the largest fires, that wiped out the city of Paradise had just occurred a few months earlier.
A few months earlier, I watched the live reports on the fire and saw how satellite data was used to coordinate rescue efforts to great effect. Later we found out that the fire was caused by a poorly maintained section of powerline that came in close proximity to trees, which in turn hit and damaged the line in a storm. We decided to look into this and see how we can combine AI, high resolution earth observation data and an open innovation business model. That is how we can up with “DeepSpace AI” which was the prototype business idea of Spacept.
SI: First we did lots of tests and prototyping. We needed to assess if the business was technically and financially feasible and to judge the interest of future clients. Traction is the best performance measure of a good idea, and with that in mind we spent several days on calls with key people in the powerline industry to better understand the issues they were facing and their requirements.
Traction is the best performance measure of a good idea.
EL: Then we discussed with satellite image providers to determine if we can obtain images meeting these requirements for a feasible price. Subsequently we prototyped a best-case-scenario test to see if a common computer vision system is sufficiently capable of discerning trees in satellite images. Finally, we iterated on all fronts to refine and improve the idea.
SI: In a nutshell, the first few months were very exciting and fast paced, but as the company is so dynamic and young we can say that every month feels like a first month as we take it to higher and higher levels of development.
AU: We were part of a great program called Antler, which brings together aspiring entrepreneurs and gives them a small stipend for two months to free them up to develop new businesses. Subsequently, a few chosen ideas receive business funding of $110k out of which $50k is taken back as fees for the program.
SI: With the initial funding we could incorporate the business and acquire the initial satellite data required to train and improve our computer vision systems. We invested the largest amount of funding into the Spacept team and built up a group of 35 developers from 15 different countries to collaborate and build the AI models with an open innovation approach.
EL: Extreme weather, drought and storms around the globe are increasingly becoming the new normal. This has increased demands and risk for energy infrastructure operating in close proximity to vegetation. Infrastructure grids such as powerlines, railways or gas pipelines are the backbones of human society. However, the current means of inspection via manually walking the power lines, or flying over with helicopters and drones is expensive and done too infrequently (on average every 7 years). This leads to power outages and massive forest fires such as the ones that occurred in Paradise, California last year, creating a vicious cycle of climate change effects increasingly nocive to our society.
The current means of inspection via manually walking the power lines, or flying over with helicopters and drones is expensive and done too infrequently.
SI: Last year’s Camp Fire in California resulted in the loss of 86 lives, with $16.5 B in resources lost and 18,804 buildings destroyed along with the release of several megatons of CO2 into the atmosphere. Spacept’s vision is to eliminate CO2 emissions for all global infrastructure inspection and proactively prevent deadly accidents and fires.
SI: We analyse multi spatial satellite imagery from ESA Copernicus, DigitalGlobe and Planet using our computer vision system trained and deployed on AWS cloud computing platform (Amazon SageMaker, EC2, RDS, Lambda). The team is simultaneously exploring several design paths for Spacept’s machine learning solutions, dedication full resources only to the ones that best serve our clients.
Among the frameworks we use are Mask R-CNN, a model that generates a set of bounding boxes that possibly contain the trees. The second step is to color based on certainty. U-Net was initially used for biomedical image segmentation, but because of the good results it was able to achieve, U-Net is being applied in a variety of other tasks. It is one of the best network architectures for image segmentation. We applied the same architecture to identifying trees and got very encouraging results, even when trained with less than 50 images.
AU: To date we labeled over 1,000 images. A large group of people makes it not only faster but also more accurate through our peer-to-peer review process. Furthermore, a set of image processes has been built, different combinations of filters were used and a basic pipeline to automate the process was implemented to test out the combinations. All in order to preprocess the set of labeled images to achieve more accurate results with the AI models.
Our model has reached the same level of accuracy as the NDVI industry standard for vegetation detection. Over the next few months, the Spacept product will be perfected to automatically perform high accuracy risk assessments. The ready system will then be able to distinguish between safe and unsafe vegetation regardless of season and climate.
SI: Starting a business is like a roller coaster with its ups and downs. The peaks we reached so far were very rewarding highlights such as the initial business pitch in front of an investment committee of 15 VCs and fellow entrepreneurs, some of which were joining in remotely. Subsequently, there was the second pitch for the seed round delivered by my co-founder Eric in front of hundreds of investors at Musikaliska in Stockholm, on the same stage the first Nobel Prizes were awarded (video here).
More recently the formation of the AI collaboration team from a selection of hundreds of interested participants and seeing hour our results and work progresses from month to month.
SI: In the powerline industry the business to business sales cycles defined as the time of initial collaboration contact to the time of incoming revenue from the contacting customers are very slow. This means that the iterative loop on the product is prohibitively large and quite incompatible with a fast-paced AI technology.
We took an innovative approach to compensating for this by partnering with fellow startups to reach our initial customers on a shared revenue basis so we can quickly iterate on our product. This enables us to test and prototype very fast whilst still accessing impactful contracts on the long term.
EL: In the next 6 months we are primarily focusing on improving the early version of the analytics product tailored for powerline maintenance companies.
SI: Ideally Spacept would be ready to provide predictive maintenance solution in time for the next dangerous fire season in California to prevent accidents like the one that affected Paradise last year.
EL: In the longer term we aim to grow the customer database and start analyzing millions of kilometers of infrastructure on a recurring yearly basis. Subsequently we would like to grow Spacept to provide a more diverse portfolio of satellite analysis solutions.
SI: In my opinion, the most important aspects in starting a company is the passion of the founders for the idea and the impact it can bring to society. This is the foundation upon which everything else is built and there is no substitute for passion and potential impact. Then it is paramount that the founders keep in close touch with the problems of their soon-to-be customers and have an intimate understanding of them.
Only then innovation on the technical solution they are developing can reach its full potential fast enough to make a positive impact on the world. Fundamentally, I think these are the key elements that everything boils down to. There are of course thousands of lessons we learned along the roller-coaster ride so far and we look further to learning more to come, but at the core of everything the above is essential.
The most important aspects in starting a company is the passion of the founders for the idea and the impact it can bring to society.
Working remotely is a great experience and team diversity is of unspeakable value, however we slightly underestimated the amount of work required to establish efficient processes for managing and coordinating a team of 38 from 16 different countries. For example, just setting up a meeting taking into consideration the various time-zones for plenum meetings is a logistical problem and takes a couple of time-saving digital tools to do efficiently. In retrospect, this could have been improved to minimize time spent on administrative tasks.
The most impactful way a reader can help is by joining our mission! We are always looking for highly-passionate people interested in learning and applying AI methods make an impact on Earth leveraging the opportunities offered by our growing space infrastructure. Simply drop us a line at firstname.lastname@example.org.