Recommendations for the future Neato

[glnc222]

ADDITIONAL PRODUCT: having a top pulled dust bin, the way is open for a separate 120/240v vacuum with robotic connection to the bin top when docked, automatically emptying the bin. Could even connect to a separate customer's vac plugged into a power outlet on the thing switched on when needed. Compete with Karcher...upscale market.

[Wing Nut]

ADDITIONAL PRODUCT: having a top pulled dust bin, the way is open for a separate 110v vacuum with robotic connection to the bin top when docked, automatically emptying the bin.

That'd take care of mainly the US, Canadian and Japanese markets, but the rest of us would like this addition to also work on 240V AC

[glnc222]

The "110" is just short hand for "line voltage", whatever it is. With globalization most things get distributed for the different power systems, and if not, I presume (in my ignorance) transformers are available. (original post now edited with your tip)

[fore]

I revive an old thread to suggest a software improvement that it seems to me could be made very easily. I don't know if this has already been considered or if there would be any undesirable knock-on effects, but.....

When the Neato reaches the end of an outside corner, as soon as the side sensor loses sight of the wall, it turns into and touches the wall. This has 2 undesirable effects, i) it uses the bumper trigger more than absolutely necessary, and ii) the Neato subsequently kicks itself away from the wall, leaving a small area of floor untouched. It always seemed to me that it should delay its turn by some inches. Given some great improvements I see in recent updates, I'm a little surprised to see this issue persist. I can't think of any reason why my suggestion wouldn't work well; after all, the very split second the side sensor loses contact, there MUST still be a wall or obstacle alongside the unit, so why turn in immediately?

[glnc222]

Isn't the new "corner clever"( kudos to marketing), doing this ([corners] ?http://www.robotreviews.com/chat/viewtopic.php?f=20&t=16506 and Firmware 3.1 http://www.robotreviews.com/chat/viewtopic.php?f=20&t=16522)
It presently turns immediately so the head sweeps more area when it doesn't back up. Full coverage requires turning when the inside wheel passes the edge, calculated from when sensed, and then backing up to cover the missed part. Rules for front-head vacuums... But you were saying they still didn't get it quite right?

I just noticed how the back flipping up so easily, and it banging the back up against walls, might be deliberate -- when the back pops up the body length is shortened and it can actually back up a bit farther.

[fore]

Yes, the new outside corner routine catches the missed floor area. But even so, why touch the wall if it doesn't need to? The wall, the bumper and the bumper trigger would live longer without this regular contact.

Even if the pre-3.1 routine has the best possible outside corner coverage acheivable without reverse, it's still not necessary to trigger the bumper to achieve this. It triggers the bumper on every outside corner, so why not just programme it to follow the same path without triggering the bumper, i.e. just let it move forwards a little, like it already does post trigger?

This is not a big issue to be honest, just that so much of the programming is intelligent, but this part strikes me as a little weak.

In regards to your "back flipping up" comment, I saw that on the new videos as well, but my v3.1 hasn't suffered that so much. I saw it happen in just one new place. One thing I did see today was that it pushed through a material flap. This happens because the new 'inside corner routine' waits for a bumper contact which doesn't arrive with loose material, so it continues to drive forwards (like the Roomba). The new mapping logic didn't get confused by this though; all well and good.

[fore]

I came to the conclusion this 'hitting the corner' had something to do with circular or curved surfaces; i.e. if it doesn't make an immediate effort to "re-find" the wall, it would never quite stay close to walls that gradually curve away. Still I guess the infra-red side sensor can determine distance, so should be able to distinguish between immediate loss or gradual loss, so still a chance to solve this.

[glnc222]

the new outside corner routine catches the missed floor area. But even so, why touch the wall if it doesn't need to?

How is it going to know when it has moved to the wall without contact sensing? The laser may not be that accurate. It could use another forward looking wall sensor, but more cost. It is what it is. Compressing the bumper brings the brush another 1/4 inch closer to the wall. The bumper hits so much already, a few times more doesn't matter, and it has a large durable tact switch, inexpensive. It could still limit penetration of a soft barrier according to laser parameters.
Perhaps I don't get the problem fully.

[fore]

I talk specifically about outside corners, i.e. a right hand turn. I think you are probably thinking about inside corners, left hand turn, and in that case, I agree, bumper contact is necessary (even good) with the new routine. On a right hand corner, I just suggest it could delay its right turn by a small distance in cases where the LED goes from close contact, to immediate zero contact. It would prevent it touching the wall just before the corner.

It's no big deal to be honest, just seems like it would be so easy to implement that I'm surprised it's not done. Maybe the non-implementation has something to do with surface traction, with it slipping slightly more on carpet, so the wheel rotations necessary are actually different depending on the surface.

[glnc222]

I am indeed thinking of outside corners, which had the obvious poor strategy shown in some videos, slanting off away from a receding wall, sweeping in an "S" curve to reach a recessed wall. I guess I could not believe they would fix that turn with side bumping. It doesnt' even detect a need to turn until the side sensor passes the wall being followed, and you're saying they are turning too soon, before the wheels pass the corner? Never thought they'd do that. The laser should give an estimate of where the corner is after the side sensor passes. Maybe laser data is consulted only in certain functions. Didn't stand out in the video. This needs a graphic drawing of paths to show better than intricate wording.
One reason to turn sooner might be to navigate narrow convoluted paths -- when there is a barrier ahead and it can't get that far? Hard to imagine all the requirements. Like playing basketball. But it's never seemed very clever about tight spaces.

I can see a problem distinguishing wall and furniture corners from furniture legs, where it does its round-about-thing.

[fore]

Yes, that's right; it turns too soon to the right, touching every outside corner. Surely everyone's does this; it can't just be mine?

Such algorithms are the threshold where thought alone just isn't enough; changes need to be seen in practice. I guess then what I'd ultimately like to see in a future Neato software upgrade is easily adjustable algorithms. It'd be great to try out a few of my own ideas, like angular approach to transition strips. Although I guess Neato wouldn't stand by the warranty.

Your suggestion of furniture legs could be a possible reason. The limited turning circle may necessitate an almost immediate turn to stay close to the object. Difficult to know for sure but a good suggestion.

[OntheWay]

In my home, many of the rooms have laminated floors. The problem is with wall skirtings, they are about 45 degrees angled, with 4 cm depth. When neato starts following the edge, one wheel stays on floor the other stays on the skirting, therefore the brush got no contact with ground, a brush-wide area remains dirty, always. Same situation with some of the (larger) furniture.. Although have no idea how Neato can realize that its actually sweeping the air but.. Most of the uncleaned areas after neato pass is such places, this includes the edges of thicker carpets, floor next to carpet edge remains dirty. A roll sensor could help here but this time how to distinguish flat-but-angled ground and one leg is driving high on something.. A couple of proximity sensors measuring the ground distance at around both edges of the brush assembly?

[yakky]

I would love to see some power user tweaks made available. Right now the USB interface is useless for anything beyond firmware updates.

Surely compromises had to be made in programming. It would be nice to tweak parameters.

[glnc222]

I don't think there's anything like that to tweak, and it falls within the closely held program intellectual property core value of their investment. The problem of covering spaces doesn't obviously lend itself to description in terms of some parameters, either, compared to entire procedures and convoluted logic, rule sets. Just an opinion.

[yakky]

Of course there are parameters that can be tweaked. First off is how many charge cycles before abandoning the mission. Second is travel speed. I can keep going.

[yakky]

Few more I thought of:
turn of display and all lights while running
disable sound
maximum tries before asking for help getting unstuck

and a log file dump would be great

[glnc222]

not my specialty, but the labor required to disassemble and interpret machine code without the source usually seems warranted when devices have general purpose, like phones or something. A dedicated function device like this doesn't seem worth it, unless it's a lot easier than I've experienced, maybe newer tools. Most people were interested in the interface to the lidar, which has all been published. There was even a contest or something to get it for product promotion. The money involved is more likely to be spent collecting more robots of various makes than investing in hacking Neato's operating code, that is, unlikely anyone does it. Still anything is possible.