Founding Principia Labs
This post marks my official start as an entrepreneur, I am indeed starting a new venture, Principia Labs. It is today the obvious thing for me to do and here I’ll brush through the mission of Principia Labs and the challenges that we are going to tackle.
One of the key learnings from my previous experience at HELIX RE is that digitizing buildings is far from being a commodity. In my view there are still a few challenges to overcome in order to make access to those services ubiquitous across the industry for construction managers, facility managers or asset managers. The obvious reasons that come to mind are:
- surveying buildings and turning them into digital models fit for purpose is still very manual and as a consequence does not scale well;
- quality and accuracy are not always as good as what you would expect;
- the cost and time of scanning is prohibitive for many applications.
The second observation is that surveyors collect a lot of data on site (point clouds, photos and more) but this data is rarely put to work beyond the lifespan of a particular project. By that I mean that very few surveying companies that I know leverage their portfolios to extract additional value from aggregated information. Why is that? Isn’t it possible to address the challenges above by leveraging data informed technology such as Deep Learning? How can we use those technologies to accelerate the creation of Digital Twins of buildings?
With Principia Labs I am exploring those ideas further with customers that are looking to leverage Deep Learning technology to enhance their raw survey data. As a starting point, in collaboration with Thomas Chaton, we gathered some of the best Deep Learning models out there into a unified framework that allows simple benchmarking of those models against common visual tasks (object classification, semantic segmentation etc…). In the context of building survey data, Deep Learning can help with a wide range of tasks such as identifying objects, classifying points based on which class they belong to (wall, table, furniture etc...) or registering multiple point clouds together. This has been a fascinating journey, and we plan on releasing this in the coming weeks. One thing became very apparent though, doing Deep Learning on large 3D datasets is still a challenge and there will probably be a couple of years before it reaches the maturity that we can see for Deep Learning on images. By maturity, I mean two things, first the tech is simple to leverage on custom datasets and second there are a broad range of proven business applications of the tech.
I personally believe there is tremendous value in using those large survey datasets when it comes to facilitating the creation of digital twins and that is the reason why Principia Labs exists today.