For several years I worked in spectroscopy. It's REALLY tough to take a sensor out of a controlled lab and into the uncontrolled real world and get clear, unambiguous results like this video shows. Usually the absorption signal you're trying to measure is tiny compared to changes due to background fluctuations, etc. Things tend to only work well for special types of materials that have very distinct absorption properties. There are massive unknowns here and I would never fund them unless they opened up about what spectroscopic technique they're using.
I wrote this elsewhere in this comment section but it is directly relevant to your sentence.
It is probably very similar to how Ubisoft made Just Dance using the Wii controller. The Wii controller was really horrible in terms of what you could read from it (it had no clue what your body position was), but if you stored what measurements you got from people dancing properly you could score people against that. Basically there was enough degrees of freedom to identify decent dancing to a degree and if you hid what was actually going on people though it sort of worked. Although the Wii-based Just Dance is horrible compared to the XBox version that used the Kinetic.
The key is to hide the intermediate results and profile a lot of things using the same crap sensor and then do fuzzy matching between the bad results. This works if there is enough discernible degrees of freedom.
Hello. I am interested to read more about the potential sources of variation in measurement with an NIR Spectrometer.
Specifically the effect of environmental conditions and of nearby system interactions, such as ambient radiation, ambient temperature, temperature of instrument or workpiece, contamination on the surface of the workpiece, complexity of workpiece composition, etc. Could you help me with some reading suggestions?
Thanks!
Some thoughts from an organic chemist who has a decent amount of experience using IR spectroscopy.
- IR spectroscopy is one of several methods used together to determine the structure of a molecule. It's certainly not definitive by itself.
- Some functional groups in molecules will give very specific signals (1700 cm^-1 for carbonyl functional groups, C=O), but for a lot of other functional groups, you get very weak signals that can sometimes be hard to distinguish
- IR works pretty good for pure compounds, but when you start to measure mixtures, the overlapping signals can get very difficult to isolate and identify
- As rjdagost mentioned, taking a sensor out of the lab will lead to all sorts of issues. Any stray light will likely cause any readings to look like garbage
I could see this technology being useful to identify different types of plastic. You have a limited universe of possible materials along with some pretty specific functional groups that are either there or not.
To use it as a "molecule sensor for all", seems a big problem to try and tackle.
And how the hell does it determine the ripeness of a avocado through the skin?
If your only exposure is analytical lab work (where by definition the work is difficult), you may not be familiar with the breadth of field utility of near-IR.
[edit: removed LMGTFY link - sorry, that was rude. But please don't pull the expert card outside your area of expertise: NIR has been used to assess fruit ripeness for decades]
The optical and internal quality data were then merged and a PLS regression analysis was conducted using the NSAS software package (NSAS, 1990).
My comment wasn't that you couldn't test the ripeness of fruit using near-IR, it was that you could reliably doing it using a handheld consumer product.
There is a BIG difference between running a lab analysis using near-IR and making a consumer friendly product that can accurately produce the same data.
One of the proposed use cases of the SCiO is pharmaceutical identification. Consumer Physics needs to be EXTREMELY careful with marketing this use. I would actually discourage this type of application if I were them.
I was interested to see them use a Bayer Aspirin in their demo. Aspirin is -- or was (my pharmaceutical formulation training is nearly three decades out of date) almost unique in being one of the few drugs that has the perfect powder flow, binding, and dispersal characteristics to pressure-form into tablets using a tablet punch machine without any other binders, lubricants, bursting agents, or other ingredients. Back when I was studying the topic, an aspirin tablet was generally just that -- powdered acetyl salicylate compressed into a tablet shape.
If this thing can correctly identify an enteric-coated or other sustained-release pharmaceutical formulation other than by checking a muddy absorption curve against a database of known samples I'll eat my hat.
Also: the app they used purported to identify tablets by their manufacturer brand? Ahem: WTF? This does indeed look like it's just throwing a spectrum at a database of samples and looking for the closest match. Which is hardly going to weed out some of the better counterfeits on the [black] market ...
Not to mention that it supposedly can tell you how many grams of sugar are in something, or what "percentage of water" is in an apple? No, there's no way you could do that with spectroscopy. Say you have an apricot, you measure it with the device, and it indeed says it's an apricot. Apricots are not all the same, the "CandyCot" apricot has about 2x to 3x the sugar content of other apricots. There's no way they could check the DNA of the plant and determine specifically what kind of apricot it was, unless perhaps you were to isolate the dna and change the setting on the device to 'DNA' and you had a database of the characteristics of different Apricot DNA.
This is inaccurate; water content is one of the easiest things to measure with near-infrared spectroscopy, and NIR has enough penetrance to be useful even for fruits with fairly tough skins (eg avocados, see last link below)
There are multiple commercial manufacturers selling NIR sensors for fruit quality control:
SCiO is based-on near-IR spectroscopy, instead of typical IR spectroscopy. Even though they both share the same "IR" in the name. The way how these technique work can be significantly different.
SCiO do not use typical IR spectroscopy. For near-IR typically, chemometrics (or let's say PCA) has to be used to get any type useful information from the result.
There doesn't seem to be any mention of what exactly new technology is this. I would expect all the patents filed at this stage and this information be public - unless there is something fishy or they are going trade-secret route which would be very difficult.
They say the device can detect materials of 1% concentration. But what kind of materials? Spectroscopy is hard, precise measurements are even harder.
Also why are they running $200K campaign? It seems they already have significant staff and $200K would be peanuts as a goal.
> Also why are they running $200K campaign? It seems they already have significant staff and $200K would be peanuts as a goal.
A lot of people use Kickstarter as a marketing means rather than a fundraising means. At least one of their cofounders is a VC thus I expect they already have VC funds or at least very significant angel funds.
I managed to come upon a couple interesting pertinent articles. One about other, similar Kickstarter projects that turned out to be scammy[1]. And another article, written by the same journalist about scio[2].
Hopefully it wasn't a "Clever Hans" type of demonstration for the second one. IE, someone watching on camera and then manually entering data into the interface from a different room.
It's a common Kickstarter practice to set intentionally low goals. It makes the project seem more achievable to those pledging before the goal is met, it makes the project seem more successful mid-campaign as it exceeds the goal within a short time, and you can later brag to the press and potential customers about exceeding the target goal by hundreds of percents or more.
Look at all the successful projects of late, most of them adopted a certain Kickstarter formula in terms of goal, videos, page design, marketing language and so on. There's nothing casual about Kickstarter anymore, at least not in the big money campaign.
Also, if they fail to reach their goal they get nothing. Outside funding is basically a given. They're likely just using Kickstarter as an extremely lenient pre-order service.
I think he was referring to the technology, as in the hardware.
To do this stuff properly, you often need big expensive equipment, so the real question is what they are doing to achieve results using cheap hardware.
It is probably very similar to how Ubisoft made Just Dance using the Wii controller. The Wii controller was really horrible in terms of what you could read from it, but if you stored what measurements you got from people dancing properly you could score people against that. Basically there was enough degrees of freedom to identify decent dancing to a degree and if you hid what was actually going on people bought it. Although it Wii-based Just Dance is horrible compared to the XBox version that used the Kinetic.
Very cool idea, but I hate that it must have data connectivity to work. Not only that, they won't actually give users the raw scan results! You have to transmit them and then get the analyzed results only.
Reality is that many such services inevitably go under, and then your device is useless. Especially when the service is as compute intensive as this one sounds.
That is indeed a very big catch. Almost backed this on impulse before reading the full description.
Looks like you have to pay separately for each individual use case as well!
Hard to see how they will compete with the inevitable clones, unless they get really lucky with patents. The nickel-and-dime scheme is not only consumer unfriendly, but probably a strategic mistake.
Unfortunately this did not happen with most fitness trackers. There's no successful reverse engineering of Fitbit's, Misfit's or Jawebone's Bluetooth protocol yet. Although the user base and interest may be much higher than that of spectrometers.
I joined the work on reverse engineering Fitbit's protocol but the encryption isn't hacked yet.
I second that! It's an unnecessary date of expiry for hardware. You see that in every fitness armband and more and more apps. In two years ahead it could end up as a useless piece of electronics.
In the past we used cloud features for offline tracking in AR applications because an iPhone 3GS wasn't capable of real time pose reconstruction. But today it seems more like a lock-in and a way to collect data. A quad-core cpu on a phone should do most tasks offline.
No, I'm not saying it does. Single fixed function devices seem to have a longer Mean Time Between Replacement.
I would argue that you're comparing apples to oranges here. Say Fitbit (of which I'm a user) came out with a new improved version that's more accurate, lasts longer between charges and tracks more aspects of my health. Would I replace it - yes.
When the device is part of a service, when you look holistically at the product as a service then I believe concepts such as 'date of expiry for hardware', 'it could end up as a useless piece of electronics.' are moot.
When you upgrade to the lasted version of an installed software application, and the previous .exe can't read the new file format is it 'a useless piece of software?'. Technically yes, so what is different here?
I assume that using the SDK they provide lets you access the raw scan results. It would severely limit the possibilities for the apps they want developers to work on on top of their device.
>Dear backers,
Some of you have asked about access to the raw spectrum and some subscription fees related questions. We take your comments very seriously and are now processing that internally. Please stay tuned for our upcoming updates.
Thank you for supporting our project!
The ConsumerPhysics Team
I take that, along with the pitch page to indicate that they weren't intending to initially, at least.
This would make sense to me (despite how distasteful it is to kickstart a closed-data project). Back in school, I remember what what essentially an encyclopedia set that was just X-Ray diffraction peaks for various (all?!) materials and substances. There was also peak detection software, but the libraries were not as easy to trace and look up, so we resorted to the books often (at least for a second opinion).
That they may treat the sensor fingerprints as a Trade Secret of some sort makes sense, considering the current cost and value of such data.
I see a sort of structural flaw. They seem to be matching measurements against known materials and then reporting that. But where is the utility? Figuring out whether a pill is asprin or tylenol? I'd be interested in identifying characteristic of -new- materials unknown to the database. That's where the raw information is really crucial.
This seems like it could be pretty useful to, say, the blind. Also a great educational tool, a notion which isn't just limited to children (one of the primary drivers of change is consciousness raising -- e.g. a dieter monitoring their food intake may find this tool useful).
IR Spectroscopy is purely a pattern matching game. All it will ever do is compare a spectrum to a known one. The raw information would be useful to very few people outside of a lab environment.
That said, I will probably recommend that my PI order one for our lab.
Does anyone have an email address to contact them directly? I'm concerned this might be a scam and I want to ask them a few in depth questions and I'd rather not be forced to do that through kickstarter or twitter.
Knowing the people behind this, I don't think it's a scam. They have a good staff, so I think they are certainly trying to make this a reality. Are they "hustling their MVP" in startup speak? Maybe, I don't know, I never saw the product (nor did I ask to). Are these people scammers? Absolutely not, to the best of my knowledge.
You are not very intelligent.
1)Considering the CEO or CHO went to MIT Sloan Business school I would try drorsh@sloan.mit.edu
2) Since the company's website is www.consumerphysics.com I would try googling dror@consumerphysics.com and you will find this gem: http://www.bezalel.ac.il/alumni/workforgrads/?thread=9142
It is in Hebrew but according to google translate it has the Phone and email of the founder.
So I called him and talked to him. I said I think it is a scam. HE CAME TO MY HOUSE here in the US, I am not kidding!!!
He sat at my table with the gadget and I had a bunch of pills, fruit and just random stuff from the fridge. He let me use it while he was on the phone, no gimmicks, no sales pitches, no hovering, no excuses.
It works pretty well for a prototype. Did not detect the alcohol percentage in the wine. Had no problems with pills and fruit.
Interesting that the first comments are slating this for being cloud-based; in this particular case I think that's like slating Google for being cloud based. The raw data will make sense to very few people without the very large database of "known" things to match against. And that large database will be continually updated as people try new things.
Having said that, I'm not sure who the target market really is. It's basically a Cool Science Toy. Educational use? But in that case maybe it would be nice to show the raw data just to give an impression of how it works.
The problem is not with being based on external database - the problem is with that database being privately owned by startup, which is almost certain to disappear within a year or three, at which point the device you bought will be useless.
I'd personally be fine with cloud-based recognition if the database was open, free to copy and host locally.
That, and also not giving access to raw data pretty much kills 90% of possible uses for the devices, turning a cool piece of tech into another pretty much worthless gizmo.
Could be very very good for drug users. Being able to find out 'what pills are these' would dramatically change the nature of that marketplace, I think.
Taste and smell are already pretty good at this in the noisy background that we live in. I'd really like to see how this works out in practice and if they achieve any degree of accuracy. Labs are labs for a reason, it's to reduce the number of variables you have to contend with while trying to read experimental data from sensors.
I'm imagining it would have to be. A raw near-IR spectra isn't going to be much use unless you can try and match it up with a database of near-IR spectra of known materials.
If this can scan for carbs/sugars, this could be a real winner within the Diabetic community. I would love to be able to eat out and be able to scan my meal and know for sure how much insulin I should take.
Maybe this could even be used in conjunction with an insulin pump and some sort of sensor to determine the amount of food consumed and, BAM, artificial pancreas. (It's that easy! LOL)
Here are some links to background information, existing products, and applications of the underlying technology. They could be exaggerating the capabilities of their particular miniaturized sensor, but as far as I can tell they are not making up any new science. Spectroscopy is useful and widely used - but also occasionally oversold, so some skepticism is certainly warranted. The concrete applications they cite have been previously demonstrated (links below).
SCiO uses near-infrared spectroscopy (as clearly stated on the kickstarter page: "How does SCiO work").
There are products on the market that do what SCiO claims to do - in field conditions, but generally with a larger form-factor, and at much greater cost. For example:
My first reaction was : "That's amazing!!!". Indeed a very cool idea. A tiny spectrometer. I would be very much interesting in the details of the technology behind it and what kind of details it can find about the scanned objects.
Interesting that your comment was voted down! How many backers have ever done any spectroscopy? For that matter, how many of those commenting here have done any serious spectroscopy? I have to agree, this seems very misleading.
I did not down vote, but maybe it is because there is no extra information.
Two people have mentioned the possibility of a scam, but there is no additional information as to what indicates that it might be. I don't know enough about chemistry to make an educated guess. It would be nice to know why others think this may be the case.
I have some experience with fluorescence spectrocopy, to me it seems that if you do a proper control as the first step and then a sweep of wavelengths, look at the reflected and scattered light you can certainly get a lot of info. It doesn't seem like they do a wavelength scan though. But they do look at the spectrum (they say in the movie), do they just excite with one wavelength? Which one? The lack of any information on the inner workings is annoying.
A very big shame though that they don't say anything about the raw data. One could imaging many things (authentication? pollution measurements? ITTT kind of stuff) people could build for this.
You're right about that, I'm just putting it out there that they're not a scam in the sense of "a fake company that's built just to fleece people on Kickstarter without having anything real behind it", which is what some people mean. That's also the kind of "scam" I heard when people discussed other projects like this.
