The other week, a few of us here had an illuminating chat with Stuart Maggs, CEO and Co-Founder of Scaled Robotics - a company which was formed out of the frustration of not being able easliy and accurately check if the physical installation on site had been installed as per the design back in the office. Since, "Whilst we design digitally, we still construct manually" for the most part. Sure there are scans you can do to check installations, but these take time and are costly - work on site moves quickly and you don't want the project to be put on hold for something that might not even be a problem. However, on the other hand you do absolutely want to identify problems that need to be rectified as early as possible so that they don't have a meaningful impact on programme and therefore, cost.
So what is the solution?
Scaled Robotics' product is not just a software package, it's also a bit of hardware. Initially, a robot with a LIDAR scanner that would travel around your building and produce a point cloud of whatever was there. The technology in lasers and photogramatry for digital scanning has been developing at a pace over recent years. As such, Scaled Robotics have been developing their product around the LIDAR technology to a handheld device that any worker can walk around the site with to produce the scan. This (perhaps sadly), removes the need for the robot.
Nevertheless, once the scan has been carried out, it produces an accurate point cloud survey that is then put through their platform where the AI engine gets to work on filtering out everything that is not part of the completed design. For example, all the temporary works, the workstations and general rubbish around site, so you are just left with the installed physical assets.
The filtered point cloud is then compared to the 3D Digital Design Model for tolerances. This is done by colour coding and heat maps to give an easy visual on whether something is in the right place or if it's even there at all.
If you can't measure it, you can't manage it
Stuart says that as well as being able to identify early any errors made during construction, what they have found is that it gives contractors and their sub-contractors data on their performance over time. So for example, if they see the same or similar mistakes occuring multiple times over the project or different projects, they have real measurable data that they can use to drill into why this keeps happening. This brings a new dimension to the 'lessons learned sessions' that we never really learn it seems and they can occur at the point where they are most needed - at the time of the error, not after the job is over. With almost realtime and a transparent record of what has actually happened, the trends can be seen and any action taken can be truly measured against that trend.
By tracking quality in this way, it also allows the inevitable impacts in both time and cost of rework due to late changes to design or poor installation, the opportunity to be mitigated more effectively.
Stuart points out though that they aren't trying to be the platform that manages the issues raised, but give people the information they need to raise and manage issues by integrating with snagging and QA platforms that customers are already using. Their focus is to do their part really well, to feed accurate information to the management platforms that do their bit well, to give a better result allround.
There are others in the space that provide similar solutions to that which Scaled Robotics seeks to solve. For example Buildots is a leader in the space where workers can attach one of their cameras to their hard hat and capture a scan as they walk around the site. We haven't spoken to Buildots about their system, but according to their website the scans are analysed by their own algorithms which compare the status of installed elements to the design and programme to assess the overall status of activities and the project. So where Buildots initial direction was in programme management and perhaps leans towards Quality Control, Scaled Robotics' initial direction was the other way around. Inevitably, they will be both offering very similar solutions to both problems.
The big difference between between the two is the technology used in the scans. Buildots' uses photgrammetry whereas Scaled Robitics have focused on LIDAR (Light Detection and Ranging). Stuart believes that LIDAR gives them an edge as it results in accuracy to the millimeter meaning that you can have higher confidence in the data that is captured in terms of positioning in a digital model. Not only that, but the price of laser technology is going down as the quality is going up.
As with all solutions in development, it's never as easy as just plug and play as there are still challenges to address with the quality of data at the start of a project. For example, they have had to overcome issues with the 3D model files and how the information is produced within these to start with. This is not uncommon as designers do not always do what's ideal for all uses when drawing up 3D information. To solve this, it needs better definition in the Project Execution or BIM Execution Plans to get aligned with all technologies that the project wants to use. In anycase, Scaled Robotics have tried to overcome some issues with machine learning which goes someway to ironing things out to fit in with poor 3D information.
With this in mind, to feed the AI platform, they need more data which will only serve to improve the platform. It's likely that reality will never reflect the design with 100% accuracy - nothing is perfect and as we say here, it doesn't need to be so. However, with more data it enables you to make better decisions to get as close to the designed solution as possible and have an accuratly managed digital asset to prove it. So for example, when your insurance company asks for the QA records or evidence that the fire protection to the building was installed correctly, you can be fully transparent giving everyone confidence. Who knows, they may even lower your premium!
It is worth noting also, that whilst the solution offered works very well in buildings, it is not very strong on infrastructure projects - so anything that you might consider is more 2D really.
So when will we see more of this?
When talking through this with Stuart, it seemed to me like this was a no-brainer solution for a lot of building sites who want to manage quality and programme more effectively. So where have they been, why is this not out on more projects?
Over the last few years they have been focussed on the technology, getting the AI to a point where there is next to zero manual interjection required by a human in the filtering process. They have worked the proof of concept on some small scale projects and are currently working on 6 projects with a couple of larger projects in place for the summer of 2021.
Personally, as with all things that look to use technology on site to remove the admin burden to managers and workers, I'm excited to follow how things develop and hope to be involved on a project using the platform soon.