Hubs are dying, tell me it's not so


That looks like Scottsdale, way to represent :wink:

Home automation utopia. Automated when and where you want it, but only when and where you want it.

I've automated as much as I feel like I can at the moment...the part where I struggle is what do I do with areas that I wanted automated sometimes, but not others? Some of that is possible with modes...automated during the day, but not at night. But there are lights I sometimes want to turn on during the day, and sometimes not. Voice assistant takes longer than walking over and pressing the switch.

Also struggle with things that I want automated, but my wife doesn't.


My names not Joe..............................


All Right. Well we are all on the first categoria so no surprise. Lol

It's a long road but HE and we all we will get there. To the place where average Joe gets what it wants.

It's not a matter if... but when and how will get there :wink:


I find the three layer model a helpful way of looking at smart home systems:
First, you have remote control - anything from apps to voice control to handheld remotes
Second, you add automation - rules based typically, manually configured or predefined by manufacturers.
Third, and the bit I'm most excited by, the system has a level of autonomy. Think Tony Stark's Jarvis.

At the moment we're still spending most of our time on automation, trying to agree common standards and connectors etc. Eventually the number of standards will reduce. At that point I think that the hubs that continue to be a success will be those that have managed a transition to include autonomy. Think hub + machine learning + semantic web (adds meaning to data, not just structure) + agent based computing (autonomous programs that try to achieve goals on behalf of their user). Now, that's going to be fun!

I bet Google already have some AI autonomous smart hub like things in the pipeline but it's not quite ready yet, and to be honest, neither is Joe.


I agree, but we need to keep the processing out of the cloud.



Bottom line, for the "average" user to get into home automation - and of course, I mean more automation than voice control - it will need to be simple, out-the-box, and seamless. People don't want "control". They don't want "options". They want it to work with a minimum of effort.

ST, HE, etc., can't offer that. Nowhere near it. Not now, and probably not ever. If it is attainable at all, then it will be through AI, and only the big boys can leverage that. Partially because of the tech, but mostly because half of "AI" really isn't technology, it's having massive piles of data. (That is to say, it's not just processing of transactions, but processing of historical data behind the scenes to pick out the trends and correlations.)

Now, that's not to say HE's prospective customers will decrease, rather it will be a much smaller proportion of the total number of home automation users... Even then, I suspect the likes of HE will need to pivot to being an add-on to the Gen 3 Google Home Controller or Gen 4 Echo Pro Plus or whatever, for the power users who do want control and options.


I agree that AI is the way forward. But as someone in the Data Center business I can tell you Edge processing is the Holy Grail, and it's not far off. Even 5G wont make data free to transmit, and nobody wants to give up their privacy. Intel's next processor architecture will have AI optimizations baked in. There are USB sticks that can make anything programmable with a USB port into an AI engine. There's no reason an in-home hub can't be trained on a families habits over the course of weeks. No cloud, no Google, no Amazon, no Microsoft, no government peeking at your data.

It's possible.


Right now the biggest challenge to home automation is well, automation. In my opinion, nobody, and I mean nobody, has figured out how to do automation. Whether you use Smartthings, Hubitat, Wink, HomeAssistant, etc. you pretty much have to think like a programmer to build any kind of real automation. That kind of thinking is way above the level of the average Joe-customer. I think that Iris, for all of its faults and limitations, probably comes the closest. It has the easiest out of box experience, a very limited, predefined set of automation rules written in natural language make it easy for newcomers to understand. But even so, it's not really automation it's home control.

What doesn't exist yet, but will in the very near future is a true "smart home" hub, whether it is a stand alone device, or part of another. Here an example of what I'm thinking..

A customer buys a motion sensor, opens the box, holds a QR up to a camera on the hub (or via bluetooth NFC). At that point the system connects the device and asks the user what room he's putting the device in. The customer installs the device and confirms (by voice) the device is installed. The smart home system responds, informing the customer there's already a light switch in the room, and asks if he would like the lights turned on when anyone walks in the room. With a simple yes or no response, the customer has already began automating the room.

How does this happen today? The customer logs into a mobile app or web site, clicks a button to begin pairing a device, opens the box, pulls a plastic tab then waits for a few minutes, hoping the device actually is recognized. Then the user installs the device and returns to the mobile app/website, looks for the appropriate "rule engine", not understanding what they are, or which one is most appropriate for the task. The user then follows a very complex set of nested menus and options, scrolls though list of devices to find the newly connected device, and finally, after several minutes of frustration, completes the task.

For many of us, especially power users, this kind of detailed programming is second nature and not all that difficult. But for most consumers, it's a non-starter.

What the "mass-market" hub of the future looks like? That's anybody guess. But I bet the letters A and G will be huge players in that future.


This, but more. It would be able to take motion sensors, and determine our work/out-of-house schedule... or vice versa, using a known schedule work schedule combined with a motion sensor, to send security alerts, without ever having to define a "security system". Basically, with just a few inputs plus sensor data, it could extrapolate 90% of what any of ever define into HE.

That power can't be put in a hub. It has to come from the cloud. The ruleset could be run locally, but the ruleset can't be defined locally. Privacy minded people might balk, but most people wouldn't find clicking the "opt-out" button is worth it. Frankly, most people don't click "opt-out" when it makes no functional difference and presented in an obvious way.


Completely agree. A smart home needs to be able to learn users activities and personal habits. My favorite is bathroom motion lighting. The one complaint I get is that when the DH is getting dressed, she frequently travels between the bedroom and bathroom. I have the lights turn off after 1 minute, but she still complains about walking into a dark bathroom while she's still "using" it. The problem is not quite as pronounced with HE since the reaction times are quicker, but it still annoys her. A "smart home" system should be able to adapt to those period of intermittent motion, see that there's opposing motion in the adjacent room, and be able to adapt to the knowledge.

As much as we all love the local aspects of HE, a true smart home system, at least in my utopian view, will need a blend of cloud intelligence as well as local execution.


Of course it won't meet our utopian view. Never does. Even at best, there will be things that AI just can't figure out from in-house data. My hope is that there will be room to fill in the gaps. My fear is that letters like G and most especially like A won't allow something like HE to connect with an access level needed to override the automated automations.


Why would you have lights shut off after 1 minute of no motion?? Of course you are going to have MANY false on/offs at that level. Hell, I hold still at the sink longer than 1 minute...

Set it to 15 minutes. The $0.00001 difference in electricity won't hurt anything, and will work much closer to 100% of the time than 1 minute will.

But whatever. Your house, your system.


Yeah I'm with @JasonJoel.
I'm a control freak and LEDs are cheap to run.
One major improvement I can think of in the future home is that as I create rules I get warned about conflicts thanks to AI simulation that could know, he probably can't see that this is going to have an undesirable result.


Based on work I’ve seen my son do I think we are underestimating what machine learning will produce.

He’s done deep dives on photography and used what he learned to do real estate photography. After a few months he realized that real estate agents won’t pay for high quality photography, so he moved on to the next thing he want to dive into.

That next deep dive was into AI. He started finding libraries people have created to do all sorts of things. An example was one where somebody trained on recognizing garbage using a camera. You can drop a banana peel through a hole and that library can tell it was a banana peel or potato peelings or just about any imaginable piece of garbage the system knows what each piece is.

After seeing that he wondered if he could take a normal high-quality daytime real estate picture and let AI turn that pic into a twilight shot. He remembered just what a pain it was to setup, wait, and then get the perfect twilight shot.

Using something called GANs he fed one cloud system 1500 twilight shots and let that system see things like shadowing at sunset, the sky darkening while the sun falls lower in the sky, porch lights having turned on at sunset, and so on. Then with only 1500 pictures of learning it was interesting what his little cloud experiment produced. While it would not pass for high quality photography it sure did give you a notion for where we are headed.

So, you can already take a $20.00 Wyze camera and record 24/7 to an SD memory card. That video has a crude, but useful motion sensing thing included in the video, so you can quickly find motion in that big video. Local facial recognition is as easy as recognizing what type of garbage is falling into the garbage can in real time. Once the AI trains in the cloud the result can run on small computing pieces like a phone or Raspberry Pi.

It will be trivial to know who walked through a room and begin to learn patterns. Then it will be just as easy to come up with rules. Unlike where computers can create pictures or write music without human help, or drive cars in traffic, AI for home automation will have access to hard, not fuzzy data as light switches are turned on or temperatures is recorded in real time. When you consider how close we are to connecting facial recognition, smart home sensor data, and other types of existing system metadata, it’s possible that home automation could wind up being one of the first real AI driven items sold widely in the marketplace.

HE’s rule machine might soon be a real rule making machine instead of something that makes me remember just how much I hated 7th grade algebra.


@rudy that iris software on a wyze camera outside your door would be awesome.


The power can absolutely be in the hub. These hubs don't have to be rPies forever. Todays desktops have more than enough compute to do this today. Prices are falling as power rises. Another few years and a cheap landfill android phone will have enough power to do what we need.


I use LEDs mostly in lamps with Hue Bulbs but even those suffer from metamerism. So for the rooms we use most Its still 100% halogen. I use LEDs outdoors, stairwells, basement and attic plus scattered Hue bulbs in lamps. For those areas I use a longer dwell time as the cost is neglible. But 700 watts of halogen is not cheap to run. At least not in New York.


It will be a very long time before a home computer can process terabytes of data on a routine basis. It is absurd to think that anyone would invest in one, and to have that degree of redundant processing would be about as ethical as Bitcoin mining. To form the predictive connections will require a data center. Anything less than the complete sensor data from millions of people, and it wouldn't be accurate enough to bother with.

edit: And of course it would require distributing that data to literally everyone. Aside from privacy issues, no corporation would have incentive to collect the data only to make it public, so it would have to wait until after the Great Revolution.


You can say it's absurd, but that doesn't make it so. If you'd look past your preconceptions, and look at the strides being made by semiconductor companies in the field of AI you'd know I'm not as far off as you think. Also we're talking about bathroom schedules not autonomous cars. The work load isn't nearly as large as you've implied. Time will tell. Maybe your bathroom workloads bigger than mine, I dunno.


Let's see... If electricity is $0.25/kwh (which it isn't that high in NY, but it makes math easier).

You turn them off at 1 minute, I suggested 15. That is a difference of 14 minutes.

700w * $0.25/kwh = $0.175/hr. $0.175/hr * 1hr/60m * 14m = $0.048.

So for the bathroom, my comment stands.

But again, it's your house and your life. If a nickel is worth annoying the spouse - go for it. :slight_smile: