Robot Architect

  • Authorandyro
  • Date 2 January 2018
  • CategoryEconomy Planning Research Residential

If anyone had told me last year that starting in 2018 I would be using a robot to create measured plans of existing buildings, I would think that either a) they are whackos that over-estimate the capabilities of current technology or b) they must assume that I have unlimited funds to buy such equipment.  Neither statement is actually true however. Last September I purchased a robot vacuum – you know – like a Roomba. Only this one was just a little different, and was designed and manufactured by a company called Neato Robotics. The thing that appealed to me, was that it had a laser scanner mounted on top in what the company calls a ‘turret’. With this added feature, the robot scans the spaces it enters for obstacles and walls, and sets up a smart combing algorithm that traces the perimeter of each room, and then zig-zags the centres with a slight overlap, so that coverage is 100%, and unnecessary ‘crossings’ are eliminated. This means that the job is completed in much less time than any other vacuum, and wear and tear on the machine is reduced. Neato indeed.

Left: The Leica Disto D2 we use to measure spaces to millimetre tolerances, together with the Neato’s laser turret to right.

The Neato D5 Botvac. Controlled remotely via smartphone or tablet.

But that’s not all! The Neato Robot will even create a ‘coverage map’ of where it’s been, and what it has covered. While this is handy to know – let’s say a door was closed and so a room was missed, cool, you can see a dark region that shows where the bot didn’t go, but that wasn’t my primary interest. I wanted to see what kind of floor plan it made – and how accurate it was. Now as an architect – I create floor plans for a living. This usually means I show up on the site of an existing building, with a pad of graph paper and a measuring tape, and create a sketch of the space before arriving at the office afterwards to recreate this facsimile in the computer. Well, I used to work that way 20 years ago. Now of course I arrive with a laptop and a laser measuring tool, and I create a digital floor plan right on site, saving a step and improving accuracy as I can gather all of the information while I am in the actual space!

Now typically, I charge around $1/sf of space for this service – and this is just for 2D information, aka. ‘plan’ information. So, a 6,000sf house – to measure, can cost upwards of $6k.


Any technology that saves me time on site, or that accelerates the capture of this information has my full attention, provided it is reasonably priced. Less of my time on site saves my clients money.

However, a full Leica point cloud laser scanning station, even the base models, are well over $40k – and the conversion to simplified architectural plans and sections and elevations is still cumbersome and complicated at best. Now there are less expensive options, like Occiptal’s Canvas application that leverages Apple’s VRKit, and allows the user to ‘sweep’ a modified iPad or iPhone camera around a space and capture a digital facsimile – however – one then has to pay a third party (Occiptal) to turn these 3D scans into useable architectural models – and the process goes room-by room. This work incidentally, is not mastered by an AI agent, but is done by an actual human being at a computer workstation somewhere, and so comes in at a cost of approx. $300 per typical room. I have tested the workflow and the quality of this process and service and I can only say it was not as helpful as one might imagine, and certainly far from affordable. Dollar for Dollar the Canvas approach is no better than my old laser workflow, and at least I draw things the way I need them represented. What I really need is the ‘big picture’ – a simple, accurate sketch locating all walls within the floor plate – so that I can quickly add relevant information on top of this ‘base’.

Here is the plan my Neato Robot made for me today, in the space of 1 hour, and at a cost of $0 in terms of my time:

As you can see, I get the overall perimeter captured and I can ignore what I know to be furniture and cabinets. The dark area to the left is the stairwell – nope – ‘Rosie’ can’t navigate stairs yet. The orange block is where the robot is docked to recharge. The dimensions I added in my CAD software, and are verified using a Leica D2 laser. The white dots of the Neato Coverage Map, can be resolved to a single pixel, and the accuracy of the Coverage Map vs. the laser measured points has a tolerance of +/- 30mm. Now given that the Robot is measuring baseboard dimensions and not to the surface of walls, this tolerance may be even closer to 10mm, as the baseboards at my house are approximately 15mm thick (15mm + 15mm = 30mm).

Does this level of tolerance matter? Typically, I work to around 5mm tolerance for new design, but 30mm is acceptable on an existing building, and I would only use the coverage map as a kind of ‘sketch’ anyways, to accelerate floor plan drafting, and then I will shoot a few ‘calibrating’ laser lines for the record, and the usual photographs of the space, just to be sure it all checks out and is to scale.

In conclusion – is current Robot technology useful to help generate architecturally accurate floor plans of existing buildings in a way that allows my firm a competitive advantage? The answer is a resounding yes – it will become a PART of our workflow, where we will ‘pre-measure’ with the Neato, and return to calibrate the Neato’s coverage map with a few real laser dimensions. This will help us get in and out of a typical house in under 4 hours. And as an added bonus, we’ll even vacuum your space for free!




















Write a comment