[mehuln]

Hey All, we (maticrobots.com) have built a local-only fully autonomous robot mop and vacuum. We just use 5 RGB cameras with all on the edge device processings. We send even the map data to app using local WiFi. No cloud at all.

Matic builds full matterport like 3D map just using CV. This enables Matic to Precisely navigate without bumping at all.

We have built 10x better vSLAM than the best open source lib like Orb SLAM.

AMA.

[rmeno12]

Would you please expand on your VSLAM improvements? What other implementations are you comparing against and on what metrics/settings?

[mehuln]

The latest NN based approaches still fall short of ORB-SLAM - which is considered the state of the art (in terms implementations). However, it still has only 70% success rate in loop closures.

We tend to measure our accuracy with loop closures and slam graph consistency over time.

Here’s the link of Matic self and exploring and building map on the fly: https://www.dropbox.com/s/e051nmb1ci0o8nu/Auto_Explore_Mehul...

[ranguna]

Is that sped up ?

[mehuln]

Yes. Its at 4x the speed.

[oynqr]

All of the devices that Valetudo supports also only run the actual cleaning logic local-only. That's why Valetudo works in the first place.

[mehuln]

Yep. The diff is dynamic and continuous mapping.

[flemhans]

Does Valetudo remove that? The stock Dreame firmware does that, and I got the impression the core navigation and cleaning routes would stay intact.

[manmal]

How do you handle changes during a session? Eg legs/cats, chairs being moved etc - are those constantly re-integrated in the map?

[mehuln]

Matic keeps both static (historical map) and dynamic constant live mapping. So it constantly observes changes live all the time — just the way we do. And it keeps merging these changes in the slam graph/maps.

[ihattendorf]

Any specific reason to avoid Lidar? Cost?

[mehuln]

Single pixel lidars are not enough. Anything higher is way too expensive.

Also, Indoor world is designed with our vision first perception. We think robots need to be imbued with similar system. And we believe vision is enough (there’s a reason why nature chose vision based systems for us?)

Three bottlenecks in on fully autonomous indoor robots are:

1. Perception and Precise SLAM 2. Affordability 3. Privacy.

All three are feasible with Tesla like cameras only approach.

Problem with indoor robots isn’t sensors — it’s lack of brains/algorithms.

[Rebelgecko]

Probably not, it looks like it costs about 4x what my lidar vacuum did

[krasin]

And there's an annual subscription on top. I really don't understand the value proposition here.

[yoavm]

I don't see an annual subscription. I see one year of "unlimited replacement of HEPA Bags, Mop Rolls, and Brush Rolls". That sounds great actually.

[mehuln]

Thank you. Thats correct. Membership is OPTIONAL. Users are welcome to buy bags and rolls a la carte.

And Membership is more peace of mind with all-you-can-clean like unlimited bags and rolls, extended warranty, and accidental damage coverage.

[mehuln]

Cost is in computes and lack of economies of scale. Have to take Tesla like high-end to more affordable overtime approach.

Disc robots have been around for 20 years. Most are exactly the same from inside.

[hedgehog]

Looks pretty nice. How does it handle cleaning under furniture?

[mehuln]

Thank you. Cleaning head is 2 inches tall and extends up to 3 inches below furniture. It cleans all the visible dirt and goes underneath anything that’s taller than ~14”.

It’s designed to clean visible dirt thoroughly.

Frankly, if primary use case is cleaning underneath furniture than disc robots are great already.

[hedgehog]

Thanks, that makes sense. For context: having little kids we also keep all the crawl-accessible spaces pretty clean which includes under couches etc. Manual vacuum immediately for any obvious mess to keep it from spreading, robot to handle tediously complete levels of coverage.

[mehuln]

Totally understand and the use case and really appreciate your feedback. We hope to resolve that in diff way. Just things kind of like truck vs suv vs sedan — we will just need diff form factor.

With this one, we really want to eliminate day to day use.

[matthew-wegner]

I've used three generations of vacuum robots, with the latter two being Valetudo-supported:

* Eufy RoboVac 11 (no lidar, the "stumble around triggering front bumper" locomotion)

* Dreame Z10 Pro (lidar, base station vacuum emptying, mop attachment)

* Dreame L10s Ultra (lidar/camera, base station vacuum emptying plus clean/waste water for integrated mopping)

I root a second a vacuum for my partner's place, but past that it's very hard to recommend a Valetudo setup to someone remote who doesn't have the technical skills to do the rooting and install procedure. So the Matic is potentially appealing to me, even if I never end up using one myself.

Looking at the Matic page: I think this is aimed at people with very cluttered houses, i.e. folks with young kids? The implicit pitch here seems to be, "it doesn't matter how cluttered your house is; this little robot can get in there to clean without tangling".

Except... that's never made explicit? It's just a lot of photos of very cluttered spaces. I'm left to connect the dots.

Customers aren't going to care about "Real-time 3D floor mapping" or "Cutting-edge vision software". They want "Won't trip over your (sometimes literal) shit".

Vacuum feature wise, it seems like table stakes and not much past that? I only just upgraded to the L10s Ultra, which has a larger base station that includes two water containers--one for clean water, and one for waste water. The robot returns to the base station to cycle water and clean the mop pads every X square meters (configurable). This does such a better job actually mopping, compared to the Z10 Pro's mop attachment. It lifts the mop pads when it crosses carpet, so it can even mop the other side of large area rugs that fill rooms. I suspect the Matic's mopping will be only marginally okay, especially with no mechanism to automatically clean the mop pad?

Rubber roller seems to be common on newer robots too.

Having a charger-only base station seems really limiting (versus emptying on-robot waste into a larger bag on the base station itself).

I picked up a second L10s Ultra for my partner's place for $630 during the black friday period, so price-wise it's going to be an uphill battle. Honestly, though, I don't think people willing to pay $1k for a cleaning device are all that price sensitive. I'm trading money for convenience, and I would absolutely trade more money for more convenience if I clearly understood the convenience on offer.

Lack of Home Assistant integration makes it a no-go for me personally, as a technical user, but I realize the average person isn't going to care (and probably you shouldn't either). If you did want to support Home Assistant, I think the shortest path would be MQTT support. You don't need to do a custom integration. It's fully discoverable and automatic if you adhere to their expected structure: https://www.home-assistant.io/integrations/vacuum.mqtt/

Anyway, no idea if any of that is useful, especially since "nerdy person who roots consumer hardware" isn't exactly your target market, but maybe something in that brain dump is useful info!

[mehuln]

Thanks for the detailed feedback and thoughts.

1. Yes, our core users are young families: working parents with young children and pets. Clutter and rate of entropy is just higher. :)

2. The fundamental reason for reengineering and reinventing robot floor cleaners with Matic is that disc robots are relatively inferior and suck (literally and figuratively).

Few issues that I as user encountered: 1) constantly getting stuck, chewing wires, dog toys, etc. 2) small bin size and my wife hates the big docks - she just thinks they are ugly. (People go out of their way to hide appliances behind cabinets, so why do we need to tolerate these ugly bricks. 3) Noise. Not only vacuum is noisy but docks are like rocket ship taking off - doesn’t work with pets or young kiddos. 4) Can’t get to sides and corners - circle shape. Vacuum at the bottom is literally 2 inches away from the side 5) and, can’t tell it where, what, how to clean. Most times I don’t want the whole home cleaned or whole room — I just need to clean kitchen are clean where we cook/chop veggies etc.

** However, it just doesn’t make sense to why we are tolerating these Gen 1 robots that were invented in early 2000s. They were great for that time. But they are Nokias and Blackberries. Now tech and AI is so much better, so it’s time to reinvent and build iPhone of home robots.

(200+ self driving car start-ups, same amount building industrial robots, none in home space…why?)

2. We believe that just like self-driving cars need Google Street View maps (which Sebastian Thrun built first) and GPS, fully autonomous indoor robots need precise SLAM and high fidelity 3D maps. With Matic, we are letting robots build maps on the fly and remember its location in precise manner.

3. You are right that mostly we have built table stakes features. However, the it’s about HOW we have built them. With our vision-first approach, disc robot ceiling is our floor. That’s just the foundation.

And HOW is about making robot that actually works and is intelligent.

How can we call robot intelligent if it continuously needs to bump? If it doesn’t even know what’s in front of it?

A robot that can navigate our home the way we do is in itself a huge step forward and for that it needs precise and dynamic maps (we constantly observe when things move).

We have done few things: - reinvented sweeping and vacuuming to adjust suction, brush roll speed, height of CH etc. based on type of surface and type of dirt. (We don’t take vacuums over rugs with frills and we mop wine stains - why can’t robot do that).

- first if it’s kind self-cleaning mop that doesn’t just drag dirt with it like mop pads. Instead we squeeze dirty water and dirt out the bin with every turn.

- a completely new mobility system invented for the modern homes. disc robots don’t climb shag rugs. We do.

- quiet. Vacuums at 55dBA and mops at 52dBA. This is really important for robots to do things on our behalf in homes. They can’t be noisy.

- with our ID not only families love looking at it but kids and pets are not afraid of it.

And, we are just getting started. By end of the next year we will be adding embeddings and simple chat like command/control based on visual maps.

Stay tuned!

[matthew-wegner]

Sort of a strange question on a Valetudo thread, but have you used Valetudo on a lidar-enabled robot? You can easily clean specific rooms, and I've never seen a lidar robot get lost, and very rarely stuck (usually something like managing to close a bathroom door behind itself).

But anyway: I do think it's worth tripling down on your "average busy person with a cluttered house" pitch. 3D mapping is a huge distraction, although I get that it's unique and your team is proud of it. You're selling the features it enables!

If your site's first content was a time lapse video of Matic navigating a living room covered in toys and charging cords that would tangle most robots, though...

[mehuln]

Will add lot of it... before it can navigate precisely, it has to know what's in front of it. Lidar is a one-laser, it cannot see what's above or beyond that one pointer, so it's bumps and tangles with wires/tassles etc. all the time.

Imagine if you could only see just one-inch ehigh of what's in front of you -- not below or above -- then you'd un into tables or chairs all the time. Single pixel lidars are sort of like that.

[matthew-wegner]

I understand. IR security cameras can see the beam. Most have a kinect-like dot pattern that isn't purely horizontal, but that's neither here nor there.

I still don't think it's worth much attention. "It's better at avoiding tangles because it sees the world in full 3D".

[mehuln]

Appreciate your feedback.