Back in September, Dyson announced their long-awaited entry into the robo-vacuum market with the Dyson 360 Eye.
I was delighted to see an old friend of mine, Mike Aldred, in one of the launch videos:
Mike and I were undergraduates at the University of Kent at Canterbury (UKC) together, back in the ‘90s. I was studying European Computer Science (which meant I was on a 4-year degree programme with the 3rd year spent studying Computer Science in Paris) while Mike was studying Computer Systems Engineering. I vaguely recall having common classes together with Mike, although as Mike started at UKC the year after me it’s quite possible I’m mis-remembering.
The main reason I knew Mike at all was actually because we were both members of the university’s hockey club (yes, in England we just call it hockey – rather than field hockey – as we don’t really play ice hockey). Mike was the university’s top goalkeeper for a number of years, and while Mike played at the local grown-up hockey club at weekends we did play together on the same mid-week team from time to time. Anyway, over the years Mike and I got to know each other quite well.
Fast forward nearly twenty years: perhaps surprisingly, we’re both at the same companies we joined straight from University. In my case I joined Autodesk in 1995, in Mike’s case he joined Dyson in 1998 after completing his PhD. Seeing Mike on YouTube brought back lots of good memories, so I thought I’d get back in touch to see whether he’d be interested in talking to us about the Dyson 360 Eye. The interview doesn’t relate directly to Autodesk, as such, but as many of you are aware I’m personally very interested in computer vision and robotics and I’m sure this blog’s readership will also find what Mike has to say to be of great interest.
On a side note, while I’ve had a robotic lawnmower for 8 years, I haven’t yet found a robotic vacuum cleaner that’s interesting enough for me to take the plunge. It’s just possible that the Dyson 360 Eye – with its room mapping technology and cyclonic motor – is that device. Time will tell.
Here’s the interview, with my questions in bold and Mike’s responses in normal text...
How did you end up joining Dyson?
Whilst in my final year of my PhD at the University of Kent my supervisor was approached by Dyson who had decided they wanted to create a new robotics group at the company. They had visited many universities across the country but our practical approach and desire to build machines rather than live in the theoretical domain meant we were the ideal candidates. I started at Dyson in April of 1998 and spent the first 4 months writing behavioural software for the robot during the day and then writing up my thesis in the evening... a tough introduction to the world of work having spent six years at uni!What was the subject of your doctoral thesis?
The title was "Visual methods for robot navigation". It covered many aspects of vision based robotics including work on a vision-based maze mouse but the main focus was on producing a vision system for a robot that could learn for itself to identify where the important bits of information within an image are. It was based on biologically inspired neural networks and enabled the system to automatically break a scene down into its basic parts e.g. "floor", "walls", "door", etc.
What roles have you held at Dyson, over the years?
Initially I was a software developer on the robot projects eventually progressing to senior and then on to robotics manager and electronics research manager. I am now electronics category lead, robotics which means I am responsible for the delivery of all electronics related elements of robotics projects (hardware, firmware, software, algorithms, sensing systems) on time, to cost and to spec.
How big a team has been working on the Dyson 360 Eye project?
The team has varied in size over the years. It started at 6 people and has averaged ~20. In recent years it has continued to grow and is currently 45 people. Over the coming year the intent is to increase the team further by around 20 heads.
I think it’s safe to say that people have been impatient for Dyson’s entry into the robo-vacuum market, especially after the DC06 was announced but never released. What happened to the DC06 project?
We got to a point where it worked as a robot but did not meet Dyson's standards as a product. It was too complex (containing 3 processors and 70 sensors!) and the cleaning performance was not quite where we wanted it. We felt we needed to go back to the drawing board.
Did your team feel much pressure to deliver a product sooner?
We continuously feel a pressure to deliver products but the major pressure has always been to deliver the right product. James [Dyson] will never allow a product to be released which is not better than other products and offers a genuine solution to a problem. Many times we had got to a stage where we had something that could have been released but it simply was not good enough to be called a Dyson.
What convergence of technology advancements has led to the Dyson 360 Eye being viable where the DC06 was not?
When we went back to the drawing two major advances made the difference.
1) We started using vision as our primary sensing system. We had recognised that if a robot is going to be able to behave in a truly methodical manner in real world homes it would have to have a very good understanding of its current location, where it had been and where it still needed to go. DC06 had relied on too many sensors making it impractical and even these did not provide sufficient information. We took the decision to use vision, and not just a standard camera but a panoramic image. This would allow the robot to see all around it in a single image, using a single sensor, which sounds sensible, but when you bear in mind that vision was not common at that time (mobile phones didn't even have cameras at that point) it was a very bold decision!
We started working with Andrew Davison at Imperial college to utilise his vision-based SLAM (Simultaneous Localisation And Mapping) system and take it out of the lab and into the real world. By the time the product was working reliably we had added around five times the number of algorithms on top of the basic SLAM systems in order for us to be able to deal with all of the dynamic complexity and edge cases that the real world throws at you.
2) Dyson developed their own small, powerful and efficient motor ideal for a robot and our handheld machines. The motor is a success story in its own right. It spins at 78,000 rpm and the company has invested £160m in its development over 15 years.
What is it that you feel separates the Dyson 360 Eye from other robo-vacuums?
It is a vacuum cleaner that is robotic, not the other way around. Being a Dyson it has to clean... we concentrated as much on the pick-up, cyclonic separation, and everything to do with getting dirt off the floor as we did on the robotic element. We were not going to produce a gimmick or toy, this had to be something that its owner will use to clean their house.
Robotics is a really interesting field. You guys are using advanced computer vision algorithms, for instance. How much of the tough stuff gets done in software vs. hardware?
Almost everything is done in software. We do have a camera which is very capable in low light to enable us to handle all kinds of lighting conditions, but other than that pretty much all of the work is about taking large amounts of image data and then doing a lot of analysis. As with any mass market product cost is critical, so we have a pretty modest processor which has meant significant amounts of work to take complex algorithms and optimise them so we don’t blow the processing budget.
How closely do you need to follow ongoing robotics research? How much time do you personally need to spend keeping up-to-date?
Not as closely as I would personally like, but I have a very talented team who are very up to speed with everything robotics. We ensure that team members attend conferences around the world and encourage them to read papers and journals to stay at the cutting edge. I have always found it a very difficult domain to keep on top of because of its breadth. In reality we have a group of highly talented engineers, some of whom have very specific expertise, and whose combined knowledge covers most of the field. I personally don't think there are that many people in the world who can honestly claim to be experts in robotics as a whole... it's simply too large a subject!
Now must be a really exciting time to be a roboticist. What kind of advances to do you expect to happen in the industry over the coming few years?
Robotics has always been exciting and I count myself very lucky to come to work every day to look at how we can make our machines better, whether that be making them more reliable, more capable, more cost effective... as an engineer I love every kind of challenge. However, my experience of developing robots over the last 16 years has taught me some bitter lessons in reality. Looking at the cutting edge work being done in academia it would be easy to believe that we are on the cusp of a robot revolution. If these technologies are going to become prevalent, though, they need to be able to work robustly, reliably, repeatably and – perhaps most importantly – safely in the real world. We implemented a basic system from scratch in a couple of years. It then took 6-8 years more, and extensive home trialling, to get the robot to work to a high standard in the huge variety of homes which it is required to work in around the globe. We faced challenges with power budgets, sensing systems, regulations and approvals, cost... you name it, we had issues to overcome.
I am very excited about future products and think that robotics can eventually make genuine differences to everyone’s day to day life. Initially they will be small. I don't think someone will suddenly produce a humanoid servant who relives us of all of our chores (in fact I suspect this may never happen). But step by step we will gradually see more and more of the things around us becoming automated. I think this gradual introduction is also important for user acceptance. A key factor for a robot to be able to do its job is for its user to trust it and to be confident enough to leave to "do its thing". I think a big bang, in your face, humanoid man servant would probably face problems with acceptance and its introduction would be fraught with the need to "educate the user". However, a more gradual introduction of robotics into our way of life should reduce the fear factor and mean that as robots become increasingly capable customers are already asking for the new functions rather than being fearful or suspicious of them.
I understand you may be biased, but do you have a favourite Dyson product?
For me it was the CR01 contra-rotating washing machine. It contained two drums which rotated in opposite directions. This means the washing was "agitated" more than with a standard machine where the clothes just get compressed against the edge of the drum. This "chaotic" motion meant dirt was released more quickly and effectively, resulting in a better clean in a much shorter time. Add in the fact it had the biggest drum of any machine at the time and a very clever auto-balancing system which meant the machine barely moved even when rotating at 1400rpm and you had a perfect example of "Dyson thinking". Unfortunately it was too expensive to make and the company ceased manufacture in 2004. I had my machine for nearly 8 years and have never found anything that comes close to it since.
That’s the end of the interview. Many thanks to Mike for taking the time to share this information with us!
To finish up, here’s a video sharing some additional technical details about the Dyson 360 Eye.
