Interestingly enough, for robotic vacuums _low_ weight is also a problem: the better your suction, the worse your ability to drive!

The tank-tread wheels on Dyson's model are meant to help mitigate against this, but I gather from (now ex-) insiders that this is still a buggy sticking point (ha!) that needs to be solved before release.

ex-dyson robotic engineer here.

Correct - low weight means you have less force to apply to the wheels and hence achieve grip. The design of the aperture where the air goes into the vaccum (the soleplate) has to be optimised between flicking the carpet fibres in the right fashion to achieve good dirt pickup and not impeding sliding across the carpet too much.

The design of the tyres is critical. If there is any slip in the tyres you have a problem. Even though the attitude and trajectory of the robot is governed by the information coming from the camera it does also rely on feedback from the wheel speed sensors. If you have slip on one side and not the other then you will tend to spin around and not drive in a straight line. Adding more power to the slipping wheel merely exacerbates the problem.

You'll notice the treads on the tank tracks are chunky. This is to give maximum grip on thicker pile carpets, however this kind of carpet tend to be "squishy" and allow the wheels to float somewhat on the carpet surface no matter what design of tyre. So this is probably where they are having problems. Also you'll notice the wheels are not rigidly mounted to the robot chassis, they have suspension. But unlike car suspension the bottom of the robot does take some of the weight and it touches the floor. If you have more mass in the robot it gives you more latitude to distribute it between the wheels and the soleplate.

Its a careful balancing act and all of these variables have to be tuned based on the amount of suction your vacuum motor is providing.

At the Homebrew Robotics Club we had a number of sessions on 'localization', which was as you might expect figuring out where you were in a room. Carpet was a real pain. One of my robots (wheeled) had excellent sensors and no wheel slip but I discovered that the 'weave' of the carpet would bias the whole robot to travel slightly sideways to the direction of travel, it was so annoying. I put 'pingers' in my lab, basically synchronized sources of sound and tried to do triangulation based on arrival time with limited success, the camera should be able to do much better than that. Especially if there are already 2 or three distinctive features (lights, doorways etc).

Look into methods like extended kalman filters to mitigate some of this. It's a smart way to blend multiple sources of information and checking amount of error in real time to weight / bias one source of information over the other.

Add suction to the treads? A few air holes routed to the side of the tread where there's a vacuum pickup? I bet it would be tricky to get a seal that has low enough friction to allow the tread to slide past but still a good enough seal to get good suction from the bottom of the tread. Oh and it's got to last and last. Yikes... back to coding I go...

I have great ideas, which are actually terrible ideas, that's why I stick to coding.

Hello. Sorry about the offtopic question. How does one become a robotic engineer. What was you career path/education?

I studied Manufacturing engineering and got a bachelors degree. This is very similar to ordinary Mechanical Engineering. I happened to be working on laundry products leading up to my time on the Dyson Robot. Its just a diverse enough company that they also operate in robotics. I now work for a medical devices company still working on motors.

My personal interests have been influenced by my father and elder brother who are both electronics engineers and I developed a keen interest in electronics and software from as far back as I remember and which I still have to this day. The ability for a mechanical engineer to fully understand the implications of the integration of mechanical systems, electronics and how they are prototyped and manufactured is key.

For instance when developing the wheels we were able to achieve minimal size by integrating the motors directly onto the PCB inside the gearbox and the PCB was an integeral part of the gearbox housing. This came about because all members of that part of the team understood each others domains to an extremely high degree.

Thank you for your detailed answer.

Electric Engineering is certainly a good option. Another is Mechanical Engineering, if you're more interested in the motion, industrial aspects and the construction of robots. And a bit of Computer Science is another option, if you're interested in software aspects, firmware development, machine learning to understand where the robot is, etc.

can you not get any propulsion from a beater bar?

Not in a sufficiently predictable way. Also, over time, you might not notice but the beater bar wears away to the point that after a couple of years it no longer touches as much as it did when it was new. Worth replacing every year to maintain cleaning performance.

Wouldn't high suction give you better traction (ie. by adding an additional downward force to the force of gravity) and hence mitigate any disadvantage of low weight ?

Also with sufficiently well-controlled motors you should be able to optimize suction and drive, say to minimize power consumption or maximize speed/vacuum power.

"Wouldn't high suction give you better traction"

Reminds me of the F1 Brabham BT46B "fan car" that sucked itself down to the road using a large engine driven fan at the back of the car:

http://en.wikipedia.org/wiki/Brabham_BT46#Brabham_BT46B.C2.A...

Or better, modulate the suction based on your desire to adjust course. Suck more on the left, less on the right, and heave on the rudder for a hairpin. :)

It'd give somewhat better traction but at the same time you increase the likelihood of stalling your motors. How does the machine know if the motors are stalling because there's something jammed in the wheels or because the vacuum is sucking stuck to the floor? You can design algorithms or sensor mechanisms to try and determine this and also tune the motors but now you have even more problems!

The tank treads are sprung, so I think it actually just sucks the sole plate onto the ground rather than the tracks.

I'm not sure the ability to drive is that much worse -- for constant curvature surfaces it should just increase traction. I guess the suction could be regulated if a certain varying inclide causes trouble.

> varying inclide

them bumps