Use this space to cheer the creator along, and talk to your fellow backers.
Have a question?
Just got first pictures of the production samples of the injection molded parts! These are first shots, so we're waiting for adjusted samples, but most parts look well 🙂 Samples will be shipped to us any day now! Once approved, it's full-on production of the injection molded parts. Stay tuned, next update is about assembly 😃 Thank you all for your patience!
@Charis, we're super excited to deliver them and hear everyone's feedback! :D As a quick update, we've started assembly this week. Still waiting on those injection molded samples (supposedly on March 27th), but we've started putting together the internals now 👍
I’m so excited for this even more now! I’ve discovered my two youngest loooove lasers so much. My Pug and one of the cats are obsessed. I initially backed thinking this would give the kiddos some additional fun while I’m at work but now I’m happy to know the two most energetic will be a little worn out. ;) looking forward to seeing this project finalized and getting to drive Missy Kitty and Frank the Pug batty.
@Terry, we're wary of the potential destruction, so we're implementing "restricted areas" in a near-future update. This will let you specify areas for the laser to avoid. It's a somewhat elaborate feature, but something that is actually covered by our patent and part of the overall feature set we want to implement. We just didn't market it for this campaign because we weren't sure if we'd have enough time to implement it. Same with Alexa & Google Home integration - these are coming :)
As for the TV, no it wasn't caused by the dog 😅 The family that let us film their dog had just moved into their new house so they haven't had time to set up yet.
So I was digging through the Android app (Can't wait for Felik!) when I saw the video of the dog chasing the laser. Just as I thought to myself how much damage my dog could do thrashing about the house I noticed a TV on the floor... That didn't have anything to do with being taunted by Felik, did it?
Anyway, the app look great so far and I'm excited to wear my cats out with Felik in hopes of them not needing to rally race down the stairs at 2am as their favorite source of excitement.
Fantastic news! Christmas is coming a little earlier this year...
@Dirk Wouters, yes but we haven't gotten around to writing an update. We've ordered the packaging (finished design, started production of tooling to create the boxes and inserts). Also, the two mobile apps for iOS and Android are going to be submitted to the app stores this week.
We're still waiting on the injection molded and shelf stand samples that we ordered back in January. The Chinese New Year really torpedoed production. Normally it would take about 1.5 months to get samples, but with the CNY in the middle it becomes 2.5 months. We were told to expect something before March 27th, and ideally both the packaging and part samples will arrive at the same time. After that it's a matter of pulling the trigger on final production of these ~500 units. Very close!
Is there any news? I'm curious as a little child. :-)
Wow thanks for the quick and precise answer! Its totally understandable and of course the main goal of this product to first build a solid base to provide the promised features and please our loved pets, great focus of you guys and leaving the doors open for the future! +1
The update looks great, keep it up!
@Stefan, the short answer is that for now the models are static.
The long answer is that there are a few reasons why we don't (yet) do learning on the device itself.
Partially it is due to the scarcity of computing resources - we have only ~40 milliseconds to process an image from the camera, perform motion detection, object tracking, look at the history to see what the object was doing, predict where it's likely headed, and execute a movement plan. It's a partially closed loop system that relies on a sliding window of past events to make decisions. 40 ms is a very short amount of time to do this for one or more objects on the hardware that we have available (1.2GHz quad-core), assuming that it also has to deal with image compression, real-time socket communication, running an embedded server, and more :)
The 2nd reason is that on-device learning (even if deferred to times when the device is idle) would require some user feedback to tell the model whether it identified an object correctly or not. It's not impossible, but for now I have a difficult time imagining how this would work. One thing that was tossed around was to use a TLD tracking algorithm to build a local representation of the different animals and then ask the owner to name them using the app. Akin to how Apple and Google do face recognition while the smartphone is charging, then asks the user to name the people. TLD, however, is an algorithm that runs at about 200-300 ms per frame (3-5 fps) on our hardware, so if we ever do this, we'll be doing this kind of closed-loop learning at times when the device is not being used, by reviewing locally-recorded play sessions and attempting to extract the sub-images that represent the most actively engaged objects to build custom classifiers for each "player". For this to work, we have to enable local on-device recording first. Which is a feature that's planned.
This can be applied in the future to better track the animals. But for now it's an enhancement that depends on one or two future features that have yet to be implemented. We did train the image classifier to recognize the laser (to avoid tracking it as a moving object), but we have to err on the side of caution and consider anything that moves (including something like a Roomba) to be a potential living thing that the laser should steer clear of. So the detection model is geared toward differentiating between something like a running TV or a blinking light vs. an animal or person, but it isn't robust enough to differentiate between different animals (certainly not without user feedback).
Interestingly enough, there is a security product called Lighthouse AI (https://www.light.house) that does classify objects in the video, it even can differentiate between different people. But from the looks of it it does use server-side learning that refines and loads the model back to the local hardware periodically, akin to the recently announced AWS Deep-Lens. We want to avoid sending any imagery to a server as it opens a pandora's box of security and privacy concerns.
Sorry for the long-winded answer :) It is interesting to ponder what you can do with a camera, real-time image classification, and object tracking. There can be a lot of cool applications. But in order to not balloon the scope of the project and actually deliver something, we have to (for now) constrain it to being a really fun game for pets first, and add other features later. As a side note, I'm personally working on the feedback / customer service form within both the mobile apps today, and one of the things that will be added is a feature request form :)
Thanks for the information! Another thing, I assume you use machine learning for the cat detection (hence AI), will the felik be able to better recognize the cats over time or is the model static? Like can/will the owner be able to teach it multiple cats and it will be able to distinguish the different cats? Maybe thats just me doing some dreaming for future products :D
@Olivier, most likely shipments will begin in March.
Injection molding tooling is in production, as well as packaging. We're delayed by the Chinese New Year, which takes up most of February. If not for that, we would be assembling and shipping in February barring any unusual delays. As of right now we expect to finish tooling some time at the end of February, following by a few weeks to dial things in and make the first production run. We're close!
I hope you are well.
Can you tell me when we will receive it ?
@Stefan, we are planning on exposing Alexa and Google Home functionality for those who want to enable it. The object tracking does have interesting potential, we've toyed around with some "out there" features where you may get an alert if a pet has jumped on your countertop or something of this nature. Will have to look into it, but at launch we expect the API to be mostly black box. We are documenting it just in case, it's mostly REST HTTP calls and protocol buffers over a web socket. You would require SSL pinning with our certificate to make usage secure. Anyway, food for thought for us :)
i was thinking about integrating the stream and maybe detections into a dashboard for homeautomation. or even just registering when a cat has played with felik
@Sven Möller, thank you for the kind words. We're finding at this stage that we need to have a physical manifestation (3D print, CNC prototype, etc) of the final data that will go into production before we can proceed with the next step. Right now we need all production prototypes on hand in order to proceed with finalizing the shelf stand. The shelf stand is pretty much the last piece to be DFM'd before plastic injection molding tooling can be made. The stand is actually metal, but it has some plastic bits. It was easy to prototype, but in production it needs some load bearing, center of gravity, and drop-resistance considerations that Shaun has been great at identifying as we go along. We definitely want Felik to last :)
@Angry Lychee LLC to be a few month late is nothing bad in this complex project... For me it only shows that you think, work and improve your poduct and don't source it as it was planned, and at the end the customer get a non working prototype!
Keep up your good work and don't mind the few month, I am sure everyone expect more a high quality product than crap you get in time...
Thanks for all
Thanks for the very descriptive update! Enjoy your weekend
@Curtis, if I had to honestly estimate when we will begin shipping rewards I have to say it's February - March timeframe. We are a few months delayed..
Yesterday we've placed an order for the custom electronics, finally, after 3 months of making 4 revisions. Each revision took about 3-4 weeks to arrive. I think that was our biggest bottleneck, which prevented DFM from being executed in parallel to software development.
Right now we're waiting on 3D printed, CNC, and water-jetted prototypes of the external enclosure. These are slated to arrive on November 24th, although I suspect they might be a few days delayed due to the holidays.
Once we lock down the external parts (assuming the prototypes are good to go on first try, which they should be), we have to prototype the shelf stand, which was modified for this. This can take a few weeks, mostly due to how long it takes to 3D print and ship parts. Depending on the shelf stand's design, though, we may be able to get started on tooling for the plastic parts without waiting for the stand prototypes.
If all goes well, by mid-next-month we will have all our production data locked in, and we'll be creating QA documentation and quoting tooling work. We have to account for the holidays disrupting things a bit, so likely we'll award a tooling job for the injection molding in January. That's at least a 5 week turnaround. Again, assuming all goes well we'll be producing injection-molded parts by February.
At that point we'll need a bit of time to assemble and test everything, before shipping it out. But at this stage all the other stuff, like packaging, user manuals, liability insurance, certification, etc. will be long done so we'll be shipping them out as they're being assembled.
There is a slightly more optimistic timeline where we may be shipping units earlier than that, but this is my most realistic estimate at this point.
- Yuri B.
Hello, can't wait for your product. Is there a guesstimation for a delivery date?
@Daniel, since the last update we've been mostly focused on software hardening while we wait for the latest prototype parts to arrive. Some have shipped already, while others are being built. Turnaround on prototypes is anywhere between 2 weeks and 1 month from order to delivery. Right now we've got 3D prints, CNC metal parts, and water-jet cut plates on order to test the external enclosure. We are waiting to test the final CAD data to ensure that all the parts fit and work well before proceeding with injection molding tooling manufacturing. The next update will be when we receive the production-data prototypes, likely around Nov 20th. Unfortunately this project is a few months behind schedule, but very much on track otherwise :)
We have 4 cats also. It's crazy! 3 of them are black! So whats the updates on this guys?
@Stefan, I guess that it depends on what features people are interested in implementing and whether or not that will compromise security. We do have a REST interface over TLS, as well as a protocol-buffer based secure socket interface. But this is primarily for changing settings (schedules, sensitivity), software updates, and real-time video & control. There is not a way to fundamentally change the behavior of the device from this interface (although there is a firmware update endpoint). What kind of features were you thinking of potentially using the API for?
Will there be an API for Devs?
@Terry, thank you! We're happy to provide the updates :) Taking a product to production is a fun and fascinating process. This project may be 1-2 months delayed, but it will definitely ship, and ship soon. We're freezing some internal parts and putting them into production next week. We obviously want to hit all our promised deadlines in terms of injection molding, but there's also a secondary incentive to get all the tooling and production paid for and done by the end of the year for tax purposes.
Speaking of four cats, between me and my girlfriend we also have four cats :) Felik definitely works with that many. Once we ship Felik and spin up regular production, our next product will do what you ask (minus the airsoft!) - that is, keep animals off furniture and countertops in a safe and effective manner. We hope to create a connected pet product ecosystem that solves these kind of problems - currently a lot of these solutions do not exist and we ourselves experience the same frustrations as other multiple-pet owners.
I just want to say I appreciate the updates and responses to comments. I look forward to a successful project after the recent run of failed projects I backed. I purchased this when we had two cats but since then my wife has purchased the crazy cat lady expansion kit and now we have four cats... I was thinking you could tweak the software swap the laser for airsoft and we can keep the cats off of my counters? Kidding, sort of.
@Jerome becher, for now there is no need to give us your address. We will send out the address surveys close to when we ship the rewards to make sure we have everyone's updated address :)
How can i change my address for the pledge?
@Luca Vergano, no worries :) We will send out an address survey closer to shipping date. For now we just need to know which country you live in so we can package an appropriate power adapter with you Felik. We purchased extra adapters for all regions just in case since we did anticipate that some of our backers may move. We will collect shipping address information in a few months as we get closer to sending out the rewards :)
Guys hi, I need to get in touch with you as I will most likely move abroad so my shipping address will change. How can I reach you? Best.
@Brandan, it's awesome to hear the excitement :D We're so close to finishing the core mobile apps for both Android and iOS now! The software hardening is nearly finished and we're hoping to begin including some extra features. Got a solid production update scheduled for tomorrow (Thursday :)
Is anyone else stupid excited? My cat just sits here and looks at a laser pointer unless I hide it behind something and THEN he will chase it a little bit. He's so lazy but I'd be happy if he just stares at Felik's enticement and the miaows contentedly. I'll get a kick out of it either way. If he doesn't like it, when people come over and ask why is he just sitting there instead of chasing it, I'll simply point out his poor work ethic. I adore you folks for making this!
Great update guys. Keep up the good work.
@Andrew Just, no worries! We haven't sent out the address surveys yet. So unless you are moving countries (in which case you would need a different power adapter) there is no need to update your address yet. We will send out the address surveys closer to the ship date to make sure we have the most recent address from everyone.
I am currently moving and was wondering if it would be possible to change my delivery address?
@Bill Heffner thank you! We'll do our best to add more video content into the updates. It's a bit of a process to record, but probably the best way to demo any new features.
@creator Thank you for the updates and the set-up video was a great first impression. I can't wait for more updates like that.
@Kean, thank you for the link! Your friend points out a few other things about the Zero that made us pass on it. Like the lack of soldered-on GPIO headers. We power the SBC off of our daughterboard. The daughterboard contains the power socket, a USB hub, and an AVR chip for controlling all the "hardware" (laser, servos, limit switches, buzzer, LED, buttons) over a serial connection. With the 'One' or 'Lite' we can just run 4 wires to the soldered-on GPIO pins and be done. With the Zero it requires soldering on a header. A small step, but when you have a bunch of such steps they add up during assembly. The H2+ CPU also heated up a fair bit more than the H3 under full load.
We actually wanted to run with the Raspberry Pi Zero initially. A year ago we tried to source it in bulk and that just wasn't possible. It's still not available in bulk today, a year later. We also explored other SOMs, but you can see the cost for yourself here: http://www.variscite.com/products/system-on-module-som
The Orange Pi lind just makes so much more sense.
The Orange Pi Zero is a good board, can run Felik's software, and the price is very good. But there were these little things here and there that we just didn't feel entirely comfortable about. We'll probably find another use for it for another future project, but are sticking with the 'One' for Felik as it has proven to be stable over the past year of testing.
We have ordered and paid for a batch of Orange Pi Ones directly from the manufacturer about 10 days ago, shortly after KS deposited the funds. At the time of ordering they didn't have enough in stock, so we're waiting a week or two for them to build more.
@Creator Thanks again for the detail, that all makes a lot of sense. I haven't tried the Orange Pi range, but have a good mate who has done some reviews of them https://www.mickmake.com/tag/orange-pi
@Kean, you guessed it! The Wi-Fi module is the EDUP EP-N8508GS model. We're using the Orange Pi One, and there are a number of reasons for it:
- The Orange Pi Zero's Wi-Fi was terrible for live video. Connection quality and bandwidth were both unacceptably low, it just was not usable. The Orange Pi Lite has a different Wi-Fi chip that is better and more usable, but...
- The Wi-Fi chips on the Zero and the Lite models have not been tested/certified for FCC, and would most likely fail the recent CE requirement for packet loss rate.
- The Armbian distro for the Zero's H2+ CPU is not as stable as the one for One's and Lite's H3 CPU. We've had strange quirks happen on the Zero, and decided to stick to the H3.
In the end, since our code can run on any model of the Orange Pi, we decided to stick with the Orange Pi One. It's less expensive than the Orange Pi Lite, and more stable than the Zero. We got a decent factory discount by ordering direct in bulk, so it worked out :)
@Tom Knox, thank you! Our main goal is to stick to the timeline as closely as possible. I think most projects that end up not delivering happen to do so due to financial reasons. We don't have this problem, since we've done most development up front and also have more of our own funds on hand if we go over budget. The biggest worry we have is being delayed due to unforeseen circumstances (shipping, customs, tooling issues, etc).
As for the beta testing, we do plan to produce several production-quality units to distribute to a few testers. Our goal for that was end of August, but this beta test timeline may slip a bit since it looks like the lead time for the camera modules is 25 days and we're finally placing the order tomorrow. So while we have sample units to work with and develop with, we won't have enough to build more than a few test articles and we won't be able to give them away until we have the rest of the camera modules here. By that time (end of September) we may already be in full production mode..
Thanks for the great update - as someone who builds similar devices for a living I am loving the tech details.
Those look like EDUP nano WiFi modules. I went through a similar selection process on another project and at the time they came out at the top. I'm guesing you're not using the Orange Pi Zero as it has WiFi on board (but possibly not FCC tested/approved).
It's actually a pleasure to see a project making such swift progress. I have backed several projects, most with good intentions, but the majority of them get snagged up along the way, pushing their timeline months, even years out. I am avidly following and hoping you have a smooth turnout!
If you need a beta-tester I am willing to help. I would absolutely love to help code, but don't think I could give that much bandwidth at the moment, but beta testing I could do. Even to the point of dedicating a place here where I could place ancillary components/computers here to help capture real-time data from it for debugging. Just a thought!
In any case, thank you for the hard work, and being so transparent with us!
@Tom Knox, thank you! Yes, the software stack is certainly interesting. There is Java, Swift, and C code involved. When we started this device there wasn't a good way to add wireless connectivity for the app, so Felik was originally designed to be a standalone unit with just some hardware buttons and an internal timer for control. Only in the last 1.5 years single-board computers got inexpensive, available, and powerful enough to allow addition of wireless. It opened up a lot of new functionality, but added two extra codebases (Android/iOS) and security implications.
We designed the code to be modular in terms of how the different components interact. For example, it's pretty easy to create an interface for controlling the servos, and we have swapped three implementations over the last 2 years without changing the interface visible to other modules.
Started with using the SBC's GPIO by writing to a Linux device directly and using the PWM pins. After swapping a few controllers, we switched that module to one that sends serial commands to a control daughterboard PCB which houses an AVR controller so we don't have to figure out each SBC's GPIO over and over. Also have an implementation that prints servo commands to the console for testing the code in a desktop environment. Same with the camera, the AI layer, Wi-Fi, etc. The AI module can take input from either a live camera (two implementations for Linux and OSX), and it can also be fed pre-recorded video for training. Breaking it up into modules is definitely the only way to make the codebase manageable without it turning into spaghetti :)
Very exciting to see this happening! I am a seasoned developer (30+ years) on everything from enterprise-level systems to embedded systems, and have done a lot of programming for servos and such. It's a fairly simple concept, until you start adding the video recognition and the AI. I am impressed with your work so far and look forward to giving my cats an extra device to play with!
Thank you all so very much! We are so happy to know that there is excitement about Felik :) We'll be posting a quick update shortly! Thank you again for all your support and encouragement throughout this project. We can't wait to begin sharing our progress with you!
Congratulations and looking forward to this ingenious product!
gratulation on funding
now to the hard part ;)
Congrats on Funding!!!!!
@Kean: makes sense. Yes, the US lab tells us that as of June 13 the Radio Equipment Directive (RED) is in effect in the EU. This means some new tests are now required. In theory, if there is an issue we can simply throttle down the transmit power until it complies. We're working with fairly low TX power, so I don't think there will be an issue. Will find out around August. As for the FCC, you're 100% correct - Chinese facilities can no longer be accredited. Though that rule does not extend to Taiwan and HK, there are still officially accredited labs there. Well, we always have local facilities as a fallback, two are within driving distance of us