Open-BCI has had their Ganglion board available for years, and it is also open-sourced hardware and software. It's more expensive than the PiEEG, but this probably isn't as new as the headline suggests.
As discussed in a thread a few weeks ago, the hardware often isn't the challenging or expensive part of building a BCI system.
Electrodes and a softgoods system to reliably hold them in place comfortably is a big challenge.
The off the shelf pre-build components are expensive, uncomfortable, and/or unreliable (from our experience).
We're building a sleep wearable EEG headband which monitors sleep state and uses auditory stimulation to increase deep sleep in realtime (https://soundmind.co).
We started by working with the Open-BCI board. It was a good starting point, but once we tried to get it on our heads and sleep with it (not lying next to us on the bed), we needed to move to our own custom hardware.
That was not a big challenge. A month after that decision, we had a custom board.
But electrodes, and keeping electrodes in place, that's where the challenge is. Off the shelf stuff is either uncomfortable and expensive, or not of high enough quality to use in EEG.
We've had to not only design our own electrodes, but also the system to hold them in place comfortably while sleeping.
Many other BCI devices don't have the "sleeping" requirement, but it is still a PITA to get things going on an average users head.
Emotiv (also a local Sydney, Australia company) has some great devices which can be easily used - but I don't think they give direct access to the data.
By all means, people should play with the BCI tech, but just know that if you think you're going to buy a board, and use it in an environment where the person is not very still, you are going to run into lots of issues.
Sounds like you're solving real problems and that your solutions could end up being useful for more than just your target niche of sleep monitoring. Any chance your product will be hackable once it finally hits the market? I'd love it if I could buy an EEG with significant thought put into the design and ease of use and then mess around with the raw data myself!
Our target is not sleep monitoring at all. We are improving the efficiency of deep sleep as a initial use, and the physical and mental health benefits that come from that.
We're not focused on "hackability" atm, but if you look at Emotiv, they have a software platform, as do OpenBCI. Some people have tapped into Muse (I believe).
It seems the EEG units that have offered open APIs have mostly removed them, I'm not sure why.
What sort of "messing around with the data" would you like to do?
We have looked at opening our data to users, but seeing as we are classified as a medical device, there are considerations that there.
I'm an ML researcher and I'm very interested in doing some self-experimentation with training personalized BCI models. My thought is that if all I am the only person using the models, it doesn't matter if the models overfit on my specific brain—in fact, that would be desirable! One application in particular I've thought a lot about is training a model to embed my brain activity in realtime and then using those embeddings to guide a diffusion model as it's generating an image—basically using the EEG to give the diffusion model live feedback during the denoising process. I know EEG isn't super spatially fine-grained, but I'm suspicious that I might be able to get some nifty results. Sort of like RLHF tuning, but on steroids!
Yeah the article makes this one kick starter sound more novel than it is. Not that connections with a Raspberry PI isn't useful, but OpenBCI has been around for almost a decade now.
FWIW I have personally found OpenBCI's hardware to be incredibly useful. An OpenBCI board is basically a set of electrodes with a really good amplifier (and all the software figured out). There are a ton of interesting projects you can kick off with one of these beyond just looking at Brainwaves.
I'm the CTO at Seer Medical, we make a wearable EEG device that's designed to be worn for ~10 days, including during sleep.
Totally agree that the electronics is not the difficult part (battery management, and power consumption aside). Our most loved features are the industrial design and materials science around our electrodes and ways of attaching them to the scalp.
We've designed and built our own electrodes too. Though we don't have to do multi-day, and we're looking for a specific signal during sleep, so probably not as challenging as what you guys are dealing with.
However, as soon as they went out of business the supply of their replacement sensors dried up, that was it.
It must be interesting what you're doing with sleep. I heard a joe rogan episode where someone used a system like that and it seems like he went right to sleep.
We don't do anything with sleep latency (time to fall asleep). There is nothing I have seen in research that suggests it is possible to reduce time to sleep with auditory stimulation.
That doesn't stop other people from trying to sell into that space.
We're focused on improving the efficiency of deep sleep specifically, and the many physical and mental health benefits that come from that. A portion of that research is posted on our website https://soundmind.co/research
Our boards and electrodes are not publicly available. We are still in development. Our product is also a product, we are not selling our boards or electrodes independently.
This is a different technology that uses optical methods, but the ninja FNIRS project is very exciting and open source. They seem to have a solid and modular hardware design. I know someone who uses a commercial NIRS instrument in a clinical research setting and they said this open source system looks far more capable.
This is super fascinating. I've been itching to get into BCI hacking recently, I've got some ideas regarding deep learning integrations and it makes me happy to know that if I could just get my act together, there is consumer-available hardware I could use to start messing around with!
As far as I know, surface EEG is almost useful in terms of controlling anything. It can even be quite hard not to make the software cheat and work with the much stronger and easier to control signals from muscles.
Stuff like that is useful for neurofeedback, however. I haven't tried the newer models of the Mindwave headset, but the older ones were easy to interface with over bluetooth. That's a place I might start rather than building a custom electronics board.
Fascinating, I wasn't aware EEG was so susceptible to signals from outside the brain. I knew it didn't have the highest SNR, but I guess it does make a lot of sense that a big reason for that would be it contamination from electrical impulses from the rest of the body.
I would even say EMG is more useful as a HCI channel, except in the most extreme situations like lock-in syndrome or late-stage muscular dystrophy.
EMG is way more easy to control consciously and as Humans we probably have hundreds of degrees of freedom in our body for some kind of EMG, not all of it conveniently accessible of course. Controlling what consumer grade EEG devices can pick up is a lot harder. Brainwaves are "feedbackable", but that usually means changes in mental state like awareness, concentration, relaxation and so on.
Muscle movements, think talking, chewing, or blinking, generate voltages orders of magnitude higher on the scalp than sources from neural activity inside the skull.
I was looking at brain wave (EEG) sensors on AliExpress recently. But I couldn’t tell if spending hundreds plus euros on those sensors would be worth it or not.. so for now I did not order any of them
Also I am a little bit afraid that some of those EEG devices on AliExpress could accidentally fry my brain so there’s that as well..
You might consider trying a muse 2. I bought one used on eBay for $125. There is a very good app mind monitor you can see what's going on quickly and get captures and python options like muse-lsl for doing your own capture and analysis.
Inducing enough magnetic flux to create dangerous eddy currents in the brain is not a problem you can get without a custom power plant and extensive cooling for the superconductors (and even then). You need a field strength around 15 teslas.
HN never fails to surprise me with comments like these that can dole out exact numbers for very niche and specific hypotheticals. How on earth do you know all of that?
I'm not the person you're responding to, but my "strong magnets aren't lethal to people" knowledge is based (in part) on a Veritasium YouTube video[0] where they visit the "World's Strongest Magnet" and do shenanigans.
Learn as much as you can from Dreem - there’s just something about the sleep space that’s a tough nut to crack.
Even with the shocking restful night of sleep with mouth tape (a visceral need fulfilled), I’ll still forget nights and rip it off. That’s much smaller and “more convenient” than a headset.
Yup, we looked closely at Dreem and Philips SmartSleep which both implemented the same core science.
As we see it, they made 3 big mistakes
1) awful ridiculous looking product nobody wanted to be seen in
2) way to expensive to want to try
3) not picking a target market - hey, everyone needs sleep!
Sensing signals inside of the brain are much less exciting to me compare to writing signals via people's eye balls and ears or reading signals from people's various muscles.
Measuring how "focused" or "stressed" someone is is too much of a gimmick to me.
You can do that from EEG. You usually get an API which exposes a number of wave "channels". Depending on the person, if you for example make your hand into a fist, you might get an isolated spike on one of the channels. When developing custom controls based on these signals it's difficult to find actions that will make a channel or combination of channels spike consistently, but you may end up with like, "press tongue to top of mouth" for yes, "make right hand into a fist" for no kinds of things.
This is from some minor diy experimentation a few years ago, might be out of date.
Not well. Feel free to prove me wrong. Doing more localized measurement of your hands or your tongue will give you much more accurate readings of what it's actually doing and you can build a proper control scheme.
This is a really bad idea. There's a direct conductive path between the power supply and the person. Knowing users, the PSU will be a $2 Shenzhen special wall wart rather than a battery pack.
At least the other dev kits they compare themselves to have a prominent warning they should only be operated off battery.
Totally agree. When building our EEG system, we have to design it so the user can't have it on their head and plugged in at the same time (for charging).
However, it is fairly easy for them to update documentation to clearly say "battery only".
I think things are starting to improve, yes. If you look at rpilocator[1] now, you see a LOT more green in general than, say, a month ago. Now to be fair, a lot of it is for Pico boards and/or Pi 3 boards, but still, progress is progress. And this lines up with that they told us back in December[2] when the word was that the situation would start to improve sometime in Q2 2023 and that stocks (and prices) would reach normal levels sometime in the second half of 2023.
I suspect that once enough supply hits that the scalpers start getting stuck with stock they can't resell at a premium, and turn off their bots, things will improve very quickly. Of course that's just my hunch. YMMV.
Also, FWIW, there's no difficulty in getting a Pi 4 now. You only have to be willing to done one or the other of two things:
1. Pay an exorbitant price. If you're willing to buy from scalpers who scoop them up and resell them, you can buy an 8G Pi 4 right now on Amazon for the low, low price of $180.00.[3]
OR
2. Be very patient. There are authorized resellers selling at MSRP who are taking backorders and who will ship you a board as soon as they get it (relative to your position in the queue). I believe both Mouser and Newark are taking backorders for most Pi models (and one or the other, or both, have some Picos and other low end boards in stock).
The only real problem is if you want one RIGHT NOW and you want it as MSRP. And even then, if you watch rpilocator enough, you'll eventually catch a few in stock somewhere and be able to order one for immediate delivery. I understand that Adafruit receive stock fairly frequently (which unfortunately sells out very fast) but I've had luck getting Pi's from Elektor a couple of times in the past year.
Or one could drive to the nearest Microcenter store that has stock, if you happen to live within reasonable driving distance of one of their stores.
this has to be forecasting mismanagement at this point. The chip shortage has mostly come to an end, they're either severely underforecasting from incompetence or intentionally
I don’t even think it’s a forecasting issue at this point. It’s a lack of a plan B.
In this time they could have launched new products that make the most of available supply. They didn’t.
They could have recognised the hit that their b2b-first distribution model has had on their reputation and tried to make amends. They didn’t.
They could have coordinated with the alt-board suppliers to make “RPi approved” models for the Rock Pi, Orange Pi etc as a stop gap. They didn’t.
Hell, they could have bumped costs up a bit to spend on mitigating supply issues, reduce demand and burn scalpers. They didn’t.
So instead they just keep promising, just keep delaying, and we keep waiting. Even if their supply issues are through no fault of their own, it feels like they aren’t even trying.
The Raspberry foundation at this point is no diferent from TI's Beagleboard, a veneer of educational project who's core business is to cater for industrial clients.
Perhaps I'm being too cynical in lumping it with BeagleBoard but at this point to me the bbc micro:bit + MicroPython is effectively filling the educational void that Raspberry left behind.
I wonder how long it will be before something like this can interface with the brain with electrodes that go into the brain (like Neuralink) instead of on the surface?
Is that actually an issue with current technology? Other implants, like bone replacements or simple screws for orthosynthesis, have higher sterility requirements than a brain implant might have, given that the brain has an easier time fighting infections. And they wouldn't usually implant the electrodes inside the brain but rather on the surface but below the skull.
And it's not completely necessary to have a physical connection to the outside. Again, I'm not sure what the current generation of BCI implants are doing, but technically it should be possible to completely close the wound (which may "only" be a small hole drilled into the skull), then use wireless transmission of power and information. Maybe that's not convenient enough right now to get the amount of data required, but it's more than theoretically possible.
> given that the brain has an easier time fighting infections [than bones]
This doesn't match my understanding at all, but I'm not a medical professional. As far as I know, the brain relies heavily on isolation for protection. One function of the blood-brain barrier[0] is to prevent contamination of the central nervous system from pathogens and toxins. Bacterial meningitis is treatable with proper antibiotics, but still maintains a 10% mortality rate[1].
Additionally, simple mechanical implants (e.g. plates, screws, replacement joints) don't require interfacing with the body. They are often made of solid metals which are not bioreactive (e.g. titanium) or coated with bioresistant polymers[2].
Brain is an immune privileged organ so it is the opposite of your intuition. Get an infection into the brain and it is done.
> Immune-privileged sites include the central nervous system and brain, the eyes and the testes. Even foreign antigens accessing these tissues do not generally trigger immune responses.
It's immune-privileged. There's plenty of access for immune-cells. Some infections like with Gondoplasma are quite survivable. By comparison, even a few bacterial cells in the wrong place during certain bone/joint replacement surgeries causes major issues.
Any type of foreign body is inaccessible to the immune system, therefore prone to harbor sources of chronic infections, like bacterial biofilms.
"A significant proportion of medical implants become the focus of a device-related infection, difficult to eradicate because bacteria that cause these infections live in well-developed biofilms."
>>brain has an easier time fighting infections [vs bones]
While the brain has better blood supply than the hard outer parts of bones, which tends to accelerate healing, the brain also relies heavily on the blood-brain barrier to keep out many threats. Crossing the blood-brain barrier is a big deal, and we shouldn't assume automatically that it'll heal better.
>>it's not completely necessary to have a physical connection to the outside
INDEED! This is completely key, as having a continuous surface breaking to just under the skin is a huge infection problem that must be continually cleaned and monitored, and having one right into the brain is a truly scary high-risk proposition. So successfully encapsulating and sealing it behind the blood-brain barrier is essential.
The problem is that this means wireless communication at a meaningful data rate, through the meninges encasing the brain, the scull, and scalp. This means power consumption, power supply, and necessary power supply replacement operations - into the brain, again.
The biology side is not trivial, even as we advance the electronics side, but I'm very much looking forward to these hurdles being overcome!
Electrodes would usually be mounted on the surface of the brain, outside the blood-brain barrier of course. That's plenty close enough for most purposes.
That'd help, but it is really not the same as fine electrodes talking to just one or a pair or triplet of neurons.
Once we're out there, we're either using an array and a lot of signal processing (sort of reverse synthetic aperture radar processing) to identify detailed signals, or just using aggregate signals.
In which case, why not use more gathering points and processing power and just go outside the scalp, avoiding infection problems entirely?
As discussed in a thread a few weeks ago, the hardware often isn't the challenging or expensive part of building a BCI system.
Electrodes and a softgoods system to reliably hold them in place comfortably is a big challenge.
The off the shelf pre-build components are expensive, uncomfortable, and/or unreliable (from our experience).
We're building a sleep wearable EEG headband which monitors sleep state and uses auditory stimulation to increase deep sleep in realtime (https://soundmind.co).
We started by working with the Open-BCI board. It was a good starting point, but once we tried to get it on our heads and sleep with it (not lying next to us on the bed), we needed to move to our own custom hardware.
That was not a big challenge. A month after that decision, we had a custom board.
But electrodes, and keeping electrodes in place, that's where the challenge is. Off the shelf stuff is either uncomfortable and expensive, or not of high enough quality to use in EEG.
We've had to not only design our own electrodes, but also the system to hold them in place comfortably while sleeping.
Many other BCI devices don't have the "sleeping" requirement, but it is still a PITA to get things going on an average users head.
Emotiv (also a local Sydney, Australia company) has some great devices which can be easily used - but I don't think they give direct access to the data.
By all means, people should play with the BCI tech, but just know that if you think you're going to buy a board, and use it in an environment where the person is not very still, you are going to run into lots of issues.