Dear customers, partners and friends,
In early 2019, our deepdive into the autonomous driving (AD) industry led us to discover a massive yet unaddressed challenge, which pertains to the single most important ingredient of ADAS and AD development: data. In the past 10 years, much of the focus in the industry has been centered around the development of next generation sensors, improving state of the art algorithms like perception or path planning and envisioning robo-taxi business models. However, one area that has remained largely unaddressed in all the buzz, is the need for scalable data pipelines. As data volumes for ADAS and autonomous driving continue to increase drastically (every single car produces up to 10 TBs per hour), the current solutions to store, access and manage this data do not scale, because they largely depend on manual work. Large manufacturers are already dealing with hundreds of petabytes of sensor data. However, the tools and processes to access and work with it have not kept pace. This results in the retention of uninteresting and inaccessible data which in turn leads to excessive infrastructure costs (every petabyte costs up to $500k/year today). At the same time, the companies often utilize less than 5% of the data they collect.
At Merantix, Europe’s leading AI venture studio, we aim to build companies that can address these kinds of problems by harnessing the power of machine-learning-based automation (read more about Merantix here). Based on this premise, we started Merantix Automotive in 2019. To tackle the worsening data overload problem in automotive our goal was to develop the first search and management platform for ADAS and autonomous driving sensor data. A search engine would make the data easily and quickly accessible for developers, and would, therefore, save time and simplify data retention decisions.
Search engines make information accessible in two steps. First, they automatically index the data based on some form of meta tags. Secondly, they provide a fast and simple interface for the user to query the index, producing ranked results. This kind of technology typically becomes important when the relevant data increases very rapidly, rendering alternative access options impossible. This is the exact situation that the automotive industry faces today.
Today, we are proud to announce that our initial idea for this kind of search engine has developed into a product: SiaSearch. Our solution allows users to process large quantities of multimodal automotive data and extract queryable metadata. Using this metadata developers can easily find complex situations encountered by the vehicle ranging from lane changes to overtaking to unsafe braking. With fast search, we reduce the time wasted on repetitive data tasks by instantly connecting engineers with relevant data. Beyond access, SiaSearch enables smarter data retention, significant infrastructure optimization, and better informed recording campaigns. The technology runs on any infrastructure, uses very little computational resources and acts as a platform that can be readily extended by our customers.The entire product is packaged into an intuitive GUI and API. With SiaSearch we provide our users an ability to do in seconds what they previously did in hours or even weeks.
After nine months of incubation as a venture inside Merantix, our team and product have come a long way. This is why in November of last year, we took the next step and spun out of Merantix, becoming an independent company. To reflect this change, we will refer to the company as SiaSearch going forward.
Together with our partners and customers, we are very excited to continue on our journey to develop and scale SiaSearch. We will continue to work hard in order to make automotive data accessible and useful. So stay tuned for what’s next!