We’ve mapped billions of light-years above our heads, but we know little about what’s even a meter below our feet. At Mapxact, an in-house R&D team, together with software engineers from Alten, are creating the next generation in subsurface mapping. By combining research into geophysics and algorithm development with the latest software development tools and high-performance AI, a tool is being built that allows every radar operator to perform advanced geophysics processing with real-time AI assistance.
The core technology for underground mapping is ground-penetrating radar (GPR). Of course, you could always just start digging, but for large-scale mapping, that’s not particularly practical. GPR is the tool of choice because it’s non-invasive and, as it requires no digging, there’s little risk involved. Not to mention that under new Dutch regulations, non-invasive surveying is required anyway before you can dig into the ground.
GPR works like any other radar, sending out 10 MHz to 2.6 GHz pulses and measuring the signal’s reflection. What makes it different from other types of radar is that it can be used to image materials in the ground, including ice, rock, concrete or plain old sand. Acquiring the data is the easy part; interpreting the underground imaging is where it gets fun. Each project provides a new challenge, from sorting out the tree roots and rocks from the gas pipes and electrical cables to calculating the signal propagation rate in a substrate that changes from sand to clay after the first meter.
Mapxact specializes in interpreting subsurface data. Recently, they decided it was time for a new tool for the job, something befitting the modern world, where everything is connected and everyone expects to have all information they need at the tips of their fingers at any moment. By reimagining and redesigning the software and hardware that makes up their system, they’re not just building a new tool for underground mapping, but they’re also integrating the latest components, frameworks and technologies to do so.
This new platform utilizes sensors, including high-accuracy RTK-GPS units to localize the device position within 2 cm, wheel odometry and IMUs for correcting positioning, and the radar unit itself. GPR contains two parts: a sender that generates a directed electromagnetic radar pulse and a receiver antenna that measures the returning reflection. Different substrates, such as sand or concrete, have distinct conduction and dielectric properties that affect their returning signal and allow for identification. In particular, scanning perpendicular to a cylinder-like object such as a pipe, cable or tree root creates conic cross sections in the radar visualization, or simply, downward opening curves. Such elements can be visually distinct, and their identification can provide important information about the contents of the ground below. Combined with localization from the other sensors in the device, radar operators can accurately identify objects and their exact positions meters underground.
Engineers don’t often get the chance to play with the latest languages or shiniest frameworks. And ultimately, there are unavoidable growing pains in working with new technologies. It’s always harder to do something you’ve never done before, let alone something nobody else is doing yet. However, for Mapxact, the rewards are worth it. Using a carefully chosen set of modern tools, they’re creating a user-friendly and technically advanced product that’s likely unique in the market.
Unlike existing systems that cater to single user groups, the new platform is for anyone who wants to learn what’s beneath their feet. This goes beyond having a good user experience or intuitive controls. Mapxact is working with Alten engineers to build an AI that automates the most challenging and repetitive processes and internalizes the most intricate concepts, freeing the user from the complexities of the process and allowing them to focus on results.
The platform incorporates tools like Apache Kafka, Kubernetes and React. While popular for creating sleek UIs for smartphone applications and consumer websites, such tools have yet to become as common in scientific software, let alone for hardware user interfaces – these types of software are usually at the back of the line to get the latest components. Mapxact, however, is leveraging Alten’s engineering knowledge from other industries to build a user interface based on the same framework as most large video streaming platforms. Not just for looks, but also to take advantage of the optimized low-latency data visualizations integral to video streaming.
What’s more, the tools are employed to do some really cool data processing in ways almost no one else is doing. GPR is used in environments from remote grass fields, underground mines and tunnels to highway bridges. To support such a varied application, the data needs to be processed in a similar variety of ways. Thanks to the Nvidia SoC Jetson family hardware, computation isn’t dependent on the cloud. Mapxact’s devices are built to utilize edge computing, not only to speed up the process but also to operate entirely offline, wherever they’re needed. The devices run a tailored set of services to extract the maximum power from their hardware, with full parallelization, GPU computing and near real-time results.
The choice of the Julia programing language to build the platform has been essential to achieving these goals. Julia is a high-performance just-in-time compiled language released in 2009. While not widely known, it’s very powerful. Among other advantages, it allows a synergy between research and production code. Because it’s both a scripting and an application language, Julia solves the two-language problem, enabling researchers and engineers to work in the same formalism.
For engineers, a one-language-for-everything approach like this has further advantages when it comes to development. Traditionally, performance-critical code is written in C/C++ and data science programming is done in a high-level language like Python. Moreover, algorithms are often first created in Matlab or Python by researchers and then converted to C/C++ for production. Using Julia cuts through these circuitous processes, letting researchers write their scripts in the same language the engineers need for production, thus speeding up development time, decreasing errors and improving the quality of life for everyone.
Because it’s been written in Julia, Mapxact’s data processing pipeline is significantly more flexible and customizable than most two-language alternatives. It facilitates creating an AI tailored to help map the underground. With this platform, many of the tasks that a data analyst usually does can be automated and machine assistance is provided where automation alone is inadequate.
A cool AI is far from the only goal. Mapxact is promoting academic research to create innovations to advance its products, from funding university projects to the work of in-house geoscientists. Novel algorithms in academic geoscience provide possibilities that go well beyond industry-standard functionality. By maintaining a close connection to the academic world, we can bring theoretical advancements into industry and to production software.
The new platform isn’t just benefitting from a few pieces of work. The computational systems are designed in such a way that the AI can be continuously extended and enhanced. Research scripts can be written in Julia and are immediately ready to be converted into production functions. In turn, modular computation clusters can have functions switched out or new functions added with ease as new methods are tested and improvements are made.
The platform is currently being prepared for its alpha release this year, with the new hardware following shortly after. Meanwhile, we hope to quickly bring even more advances to this project. These include a myriad of hardware and software improvements like an expanded sensor selection for the radar and augmented reality subsurface viewing.