Robot Reviews

Those of you who have upgraded to 3.1 what are your spontaneous reactions to the new software?
http://www.neatorobotics.com/eur_en/sup ... e-updates/
I'm upgrading my neatos tonight

So far I like
It is not perfect at handling corners. If the "wall" is too short (ie a box) it will try to do the new cornering, but give up, and do as always. So not quite enough of a try-retry to it.

I could see some houses having problems with runtime - I would expect a significant increase in battery drain (takes longer now), which for some may cause problems.

It does not fix the major flaw we have - that it gets caught on chairs with the legs being like a U. It will get stuck, asking to clear it's path (even though it CAN manage to get itself free if it tried a few more times).

All-in-all only downside I can see, would be added drain to battery. Not that ours have ever had to do a mid-clean recharge.

Besides the new cornering system I noticed a potential improvement in docking. When my Neato XV15 was rotating in preparation for the final docking maneuver, it hit one of the legs of the little table that "houses" the docking station underneath. With the firmware 2.6 this would typically result in the Neato aborting completely the docking and complaining about the path being obstructed whereas today with 3.1 it kept rotating, even pushed the table a bit out of its place, then docked successfully. I hope this continues to be the case over the next runs cause failed dockings was one of my two biggest complains so far. Fingers crossed

Yes that "fix" is also seen in my 3.0 firmware neato.

Just finished cleaning my house...
I noticed a great improvement in closed corners (when it turns left) but I have a mixed feeling about the open corner (turning right).

It seems to be turning right much closer to the wall now then before. After that, it returns some inches to attack the corner again. I think much of the improvement was in the first attack, making the second attack almost useless.
When it does a right turn and there is a left turn within 5 inches, the Neato executes the old algorithm to the left turn. In other words, it didn't improve in these places.

For the first time the neato pushed the curtains instead of going around. I think it is an unintended consequence of the left turn algorithm that keep going until the front bumper be activated.

I haven't seen any kind of wall climbing so far.

To conclude, it was a good update but there is still room for improvements in the future.

Backed into charger and stopped three inches away, refusing to charge. Never happened before in 2.6.
Potentially very serious damage, but will have to see.
Battery has faded a bit and the new extra run time probably pushed it just the right amount. Could be that when capacity is not original size, the reserve calculation used just doesn't allow for that and will conk out at the oddest times. Apparently it will use a fixed amount of charge cleaning regardless of the battery capacity, so reserve is reduced as the battery ages. An old observation. Battery dynamics are well known, so they need to work on this.
Will see if this is fixed by a much larger battery on order.

glnc222 wrote:Backed into charger and stopped three inches away, refusing to charge. Never happened before in 2.6. Potentially very serious damage, but will have to see.

My XV-11's upgrade from 2.6 to 3.1 went smoothly, as did its first few runs - a couple of spot cleans, a half house and then just one room. It may have used more battery charge on the half house vacuum than in the past, asking for the charger once it finished - the charger was elsewhere in the house. Running locked in the room with the docking station, however, it found the charger OK. I'll send it off again to do the other half of the house soon and then see how its return to base goes.

The new corner cleaning regime is much better, though not brilliant. Doesn't seem to do two smart outside corner manoeuvres in quick succession - say, around a stub wall or wide open doorway. Still, it now requires far less pre-sweeping by me!

So far, so good

I would like to update but how easy is it to downgrade firmware if things do not work out in my home?

I read that people have downgraded on the VR100 but that Neato does not support the procedure on the Neato?

The US Neato cannot revert to the previous firmware version. However there is a bootloader process that does allow the Neato to be reverted to a backup firmware file that can then be used to attempt to download the failed firmware version again with help from Neato.

My upgraded XV-11 just finished a full run that included a normal (as for v2.6) mid-task recharge. All went well - didn't really get stuck on anything at all in 8 rooms and a hallway despite many traps - a very thick mat with tassels, chrome chair legs and a fireplace hearth it can sometimes climb onto.

However, on its return to base, it managed to nudge into, rather than over, the cat-frayed corner of a door mat - a carpet off-cut. It kept shoving the corner further into the mat itself, building an increasingly high loop of carpet ahead, until the whole thing flicked back and over the Neato, covering it completely.

Then as it tried reversing out, the cat climbed on top of the whole arrangement - the mat and the Neato, which was still humming away underneath! Amazingly, despite the turret completely covered, it reversed out, pulling the mat and the cat off, and then continued back to base. Wish I'd had my camera handy.

Maybe this is an undocumented feature of the new firmware? A cat dumper

The LCD actually had a message recommending moving the charger to a different place to avoid the docking failure. It is obvious that putting it in the center of a suite of rooms will minimize return time. Just not the best parking place here. Old versions are of no interest though. The newest Neato's sold have had this software for months, so what's not to like.

After a first run with 3.1, I can say that everything is working fine
Could be my imagination but I think it navigates in a better (smarter) way, its definitely run a different pattern.
Cleaning corner is, as mentioned earlier, much better

Improved navigation observed with an overhead pinch blocked by bumper height addition: instead of repeatedly probing the front bumper hit as before, with long contemplation similar to mag strip encounters, the vacuum turned at once and smoothly traversed the front of the sofa with side inserted as deeply as practical, no repeated probing, impressive.

Mag strips, on the other hand, really annoy it, like it can't really believe not supposed to cross, stopping a while to recover its equanimity and then repeatedly probing to the point there could be a trap at corners. Where a strip forms a right angle into a wall, like a virtual wall, it fusses at the vertex enormously compared to regular corners. V2.6 did not keep trying as much. It acts like encountering a mysterious invisible force field warranting exploration -- surely you left a hole somewhere. "warning. you have reached the periphery barrier of Vortex 4..." (Sean Connery)

Broad furniture legs were detected by laser, moving around without touching, also impressive.

Docking was effective cleaning a smaller area than before; the failed run also may have started in a mid-run state from some prior fiddling, as it did not pull away from the base but immediately moved to the side commencing peripheral circuit without the usual contemplation from mid-room -- like it's judging your taste in furniture. When the start was normal, docking completed at finish.

There is still a missing step in handling larger scale layout of sections: when starting a narrow strip it often criss-crosses the narrow dimension making many more turns than were it go the long way. Would not seem all that hard to calculate that advantage, so a puzzling inefficiency. More important than ever with reduced run times.
The process of improvement continues...

Even on a long, straight mag strip it keeps going back at it - I think it had decided from the room scan that the opening I put the mag strip in was the doorway to another room so it was determined to get to it - too determined!

glnc222 wrote:There is still a missing step in handling larger scale layout of sections: when starting a narrow strip it often criss-crosses the narrow dimension making many more turns than were it go the long way. Would not seem all that hard to calculate that advantage, so a puzzling inefficiency. More important than ever with reduced run times

As mentioned in the past, our Neatos occasionally seem to suffer schizophrenic tendencies and run their routines a little differently for a while before switching back to 'normal'

Mine always cleaned the hallway in long runs parallel the walls when running on v2.6 FW, then a few months ago decided to frequently do a pattern 90 degrees to this - short runs across the hall with lots of turning. It also occasionally decides to do a diagonal pattern in a bedroom for a few days, then goes back to paralleling the walls and furniture. Gives the carpet a rest, since wheel ruts were forming, but I doubt it's that smart

Not sure what pattern it did down the hallway yesterday on v3.1. I'll keep watching next time.

To be honest, on carpet, I much prefer these occasional "inefficient" runs. Whilst the neato is turning on the spot the vacuum and brush bar are still running so gives the carpet an extra workout. Especially useful at getting out stubborn fluff in the pile in the hallway and giving the carpet an extra "lift".

glnc222 wrote:Broad furniture legs were detected by laser, moving around without touching, also impressive.

Doesn't a neato always do this anyway? Or are you saying you've noticed increased sensitivity with the new firmware?

Has anyone who has upgraded from a 2.6 to 3.1 noticed reduced run times? My x25 was a ver 3.0 so runtimes pretty much the same as before. On another thread there was speculation that 3.0 (and above?) implemented reduced runtimes to preserve the longevity of the battery, thereby reducing the need for replacement.

increased sensitivity with the new firmware?

-- detecting furniture legs. Probably not new -- am usually out when it runs, and it bounces off so many things hard to keep track.
Always impressed by how it sweeps by vertical blinds without touching.

Randomness in some features may result from varied position starting an area, from accumulating small differences in turns, travel times, and bouncing around in tight spots. Also limited lidar precision and range. But the pattern variation seems awfully large. Would need precise maps of its motion to analyze fully. Deliberate randomizing makes it more interesting, but would they go to all the trouble to program that, with enough else to do?

Closer observation of the mag strip problem may relate it to climbing mentioned by others. I find it so serious may have to replace a strip with a wooden barrier the bumpers and wall sensor can see.

When following a wall and hitting a strip in front, it does not do its new or even old cornering thing. Instead it backs into the wall trying to turn, being prevented by the wall, then rolls back onto the strip to be rebuffed back into the wall, back and forth. It manages to turn just a bit each time but in some cases is trapped. There can be a difference between whether following a wall with that sensor, or approaching on the other side, which can be worse, will have to document.

This is unacceptable and should be fixed. Obviously if it hits a strip it should reverse straight back, along the wall it was following -- which I've seen it do otherwise -- and make a proper inside corner turn before probing the strip again (a different issue), and it is not doing this.

Personally I find 3.1 strange in how deals with corners.
• First it goes all the way into the corner until it bumps (great)
• Reverses (great)
• goes forward a little (Why?)
• Turns left (great)

It is the third step I believe which occasionally causes it to misjudge and run out of space to turn which in turn causes the bot to reverse "climb" onto the wall. I cannot see a point in doing that. If it was removed or the reversing is extended slightly then the likelihood of climbing is reduced.

inside corners -- laser vs bumper or boundary strip

The trap severity observed cannot be easily reproduced and depends on random factors. The difference between the inside corner turns in V3.1 when walls are visible to the laser and when only a bumper or mag strip is detected is shown in three brief videos:
Normal inside wall corner, visible to laser: http://youtu.be/LyQZvPL2lrI

Inside corner invisible to laser, bumper detected: http://youtu.be/5CsRFzcMO9M

inside corner invisible to laser, boundary mag strip: http://youtu.be/seQ6GpEY5PY

The new backing behavior is seen only when the walls are visible to the laser, and the bumper and magstrip cases are the same -- backing into the wall. Sometimes that blocks the turn and a trap occurs, difficult to reproduce.
At least when following a wall with the wall sensor, the unit should have info a wall is present it should not back into.