I'm just getting started with this, working from the library to minimize distractions.
Read the NIH SBIR site
- Ruben Gur
- Flaura Winston
- Godfrey Pealrson
- Meg Grabb at NIMH
- Wendell Gibby
think about the proposal topics
- MRI driving system
- MRI game controller
(Wifi at the library seems limited to the broswer. I can't check my mail with an IMAP client and I can't connect to the cdi vpn. This is probably ok. But - WOW - wifi tethering through my phone makes it all work fine!)
There's a helpful JIT checklist. I'm going to start by checking on the 5
registrations that are needed:
- era Commons - req'd to do business with NIH
- SAM (System for Award Management - req'd to do business with Fed gov)
- DUNS: 83-660-0924
- SBA - req'd to do business with the SBA: Proof of registration
- grants.gov? not sure
- dates, deadlines
- tools (pdf editors, drawing formats, etc.)
- ASSIST - testing this now Instructions and use the General HTML section here: How to apply
- form entry - this is the old pdf attachment approach, which can definitely work
- 09 Jan 2017
Today I wrote a Facilties and Other Resources first draft and attached it to the draft application using ASSIST, just to see how that works. I think these sections can be submitted as non-form pdfs: that's what the instructions imply. I also filled in more of the first two pages and got a few of those details straight.
- 16 Jan 2017
Most of the boilerplate stuff is either done or has to wait based on the project specifics, so it's getting to be time to define the project more.
The driving system has a few weaknesses:
- 2 interfaces, 2 bundles: should be 1
- steering wheel has no fixed zero point
- steering wheel mechanics: should be smoother, quieter, etc.
- turn signals?
- output in degrees and percent?
- no lost counts, no speed limits
I can address these with proposed solutions, but probably should add at least one significant innovation to give it punch. The one that occurs to me is a more hardware based shaft encoder detector or, more specifically, a detection system that is independent of the timing in the controller.
Today I wrote skeletal Abstract, Narrative, Biblio, and Equipment sections for Other Project Information page. The abstract and narrative are key sections to refine. There are good examples online, and reading online papers helps with this too. I'll come back to these sections. The bibliography has only one entry now, which is a paper I was reading that turned out to have Godfrey as 2nd author
. Will probably add recent pubs from our website just as qualifying material.
- 23 Jan 2017
I didn't skip last week but I also didn't get enough done to write anything down. Mostly I read Godfrey's paper. I was sick and couldn't quite muscle through it.
This week I'd like to get into the meatier part of the writing, and I think the challenges are to (a) explain the motivation and medical/public health value, and (b) to flesh out the development and testing steps so they sound like they deserve funding. I know one of my applications was turned down when a reviewer said it had all been done before, which I don't think was true but probably didn't make the case clearly enough. That was for an optical connector with integrated electronics, which I think should have had some curb appeal to a research audience.
The technical goals are important. The shaft encoder design would be preferable for the pedals, too, but we had a problem there with timing and loss of position with fast movements. A better design would have some kind of absolute shaft encoder, and shouldn't add an onerous calibration step or require the users to jump through hoops at the start. Could we use an extra channel for gain calibration? Should we use a laser diode and glass fiber? How about setting the gain as we do but using a hardware shaft encoder chip instead of doing that in firmware? What is "the right" way to do shaft encoding in our situation?
For the grant we can propose a new interface and a new approach. We have been working at the ragged edge for quite a while. We could really use about 10x more signal, or 10x more gain, preferably in a single-supply circuit. Glass fiber could help. What would really help is an op amp with lower offsets, faster response, and better behavior at high gain.
But focusing on the subject's end: how can we avoid calibration steps? There are two levels of this: insertion loss compensation, and position zero detection. Insertion loss compensation we can handle with a spare channel (for a quadrature shaft encoder) or nothing (?) for a gray scale or cross polarizer approach. Ideally the subject and the researcher should not have to do anything except plug things in. With the pedals: those can have a zero position detected when they are released, which seems like a minor burden. This is not as much about hassles and convenience as it it about reliability and simplicity of behavior. The goal is to get close to the mechanical archetypes: the wheel makes the car for straight ahead when it points to top-dead-center, or very close to that. It should be sufficiently close that the subject can't tell the difference. And with the pedals what matters is that the pedals respond with even a very light touch, and hit a very close full scale value every time. We get 1.25x errors now because the overall gain also affects the signal range for full pedal presses.
Is there a gray scale+qse approach that would help? Or a synchro/resolver gray scale approach, where two gray scales tell us the angle unequivocally? We could use two polarizers, one for sine and for for cosine signals. But there would be offset errors and gain errors and they'd need to be corrected for by calibrations, so we'd be back there again.
The shaft encoder seems like the more modern approach, but we had a speed problem and also a resolution problem. We fixed the resolution issue using the analog approach, which is clever. But if we used a hardware quadrature detector/counter we might be able to fix the speed issue, too. How about dual quadrature shaft encoders, in which the relative phase of the two channels tells us the angular position? This would be detectable with an image-type sensor, or with the two-fiber sensors we use now as long as there is motion. We'd need to be able to detect 360 phase differences to get 1° accuracy. That would be like breaking our existing 4x or so analog 'boost' into 128 or 256. What if every strip is 1°, and every N strips we have a band? We could detect an error if the count we detected was not equal to N, and then we could reset/realign to the new band. For this to work you just need to know what the max speed is and then space the error detecting bands far enough apart that they can't be missed. I guess that means they would also need a wide width, and it would be the transition we'd watch for?
A wide parallel encoder is also an option: N channels to break the wheel into 2^N slots. We would need something like 10 channels, I guess. Would be ugly.
None of that is needed if we never lose counts. Once it works absolutely reliably, then we only need to establish a good zero once at power-up. So: why do we lose counts? Is it the op-amps, or the sampling period? Can it be fixed if we reduce USB transmissions? It would help to debug this if I could detect errors as they happened on our system now.
Should probably make the ADC reading interrupt-driven, so the sampling is predictable, and then calculate the limits on speed. Wild guess on speed: sample once per millisecond. On the foot pedals, have 2° resolution. So, 500µs/°? Jump 30 degrees in 15 ms? If the pedals were heavier and had a slight drag, we could probably limit it.
The more I think about this, the more I think we should:
- make both (steering & pedals) sensors quadrature shaft encoders,
- find improved op amps to decrease offset and increase speed,
- test for lost counts with some kind of robot,
- make sure that counts can't be lost under normal use conditions
All of which is not really the subject of this grant.
A better op amp would also mean the polarizers would work well. But they'd still have the 1.25x problem. The polarizers we're using are of the 'polaroid' family, made by stretching the material along one axis during manufacture, which aligns the molecules. They have an inherent Transmission of close to 38%, or 1/2.6x. This is ~8 dB or 4 gains steps, for each film
. According to Malus' law the best we can do is 50% transmission, I think, which would be 6 dB or 3 gain steps. The tradeoff here is against the reflective shaft encoders which are also lossy. I think we're winding up with similar gains. Both of these would probably work significantly better with a 932 than with the 904. So really, we should be able to address this with a better analog front end.
OK: the above writing helped me get going on the Specific Aims and the Research Plan. Both seem kind of weak right now, but I think I can flesh them out so they're better. The kind of work I'm proposing is the kind of thing we do normally in our work, but I think the trick is to enlarge each step and make it a bit more serious and well thought-out. And to recognize that the driving stuff took a year or more to develop even in the way we did it. So more development is worth paying for I think.
- 06 Feb 2017
Backing up for a moment: what we need to do is explain
- why this hardware is needed: how it could save lives, why the research is valuable, and
- that we're making significant innovations that needs investment
I need to push Godfrey and Flaura on the first point.
For the second I think the answer must lie in careful thought about the hardware: What does it need to do and what does it not do well enough now?
These points were already made above, mostly. Generally at the end of today I am thinking the push should be on the shaft encoder and its details: it should have a zero mark and also a loopback reflective channel for gain setting independently of the position. But even this will require passing the zero point at least once after powerup, which is bad. The only alternative is to have a true absolute encoder I think?
- 20 Feb 2017
The application deadline is now about 1 month away. I'm seeing some good writing here already, and a lot of good material in the Research Plan section, but that one needs to be tightened up and organized.
- 06 Mar 2017
I'm not sure what agency to send this application to now, and that impacts on how the grant is written. I've been focusing on the alcohol angle but the program officer there (Meg Ryan) was not encouraging because their focus is narrowly on "alcohol prevention". I spoke with Flaura Winston, Elizabeth Walshe, and Dan Romer on Tuesday and that was very stimulating. I'll meet them again on Monday with Tim (?) at CHOP. And I picked Godfrey's brain yesterday about the agencies and the angles, though I think the main upshot was that he gave me two names of people he thought would be helpful at NIDA. I also wrote 'cold' emails to DOT (they won't have a solicitation until the fall, and they are targeted ones), and NINDS. NINDS makes sense to me, because lots of our researchers work with stroke subjects.
So at this late stage I need to hustle to find out where this can go, and do a re-write based on that. Not that it is actually done yet anyway.
- 09 Mar 2017
Yesterday I met at CHOP with Flaura, Tim Roberts, Bill Geats, Liz Walshe, and Dan Romer. Tim was very enthusiastic and offered to write a letter of support and send his biosketch. Later I emailed Ruben to catch him up and he is also on board. He's offered to "rubberize" a letter that I draft.
There's actually a lot of enthusiasm for this and, really, for any collaboration that I can come up with that involves hardware going to the researchers. I think the challenge is to well characterize the merits of doing driving in MRI and MEG settings. It is a 3-way thing: Ruben and Godfrey represent the MRI interest, Tim the MEG, and Flaura represents driving research that seeks to be validated at a neurophysiological level by MEG and MRI. My gut says that we can get funded if I convey that well. As a first step I'm going to draft the letter Ruben requested.
Today is a snow day so I get a surprise free day to work on this!
From Develop Your Budget
: "The modular budget format is not accepted for SBIR and STTR grant applications. SBIR and STTR applicants must complete and submit budget requests using the SF424 Research and Related (R&R) Budget component."
- 14 Mar 2017
I want to set a deliverable target for today:
- design goals for the device, emphasizing ergonomics, structure, helpful features (not too much about tech)
- Phase II plan
- NIMH-targeted abstract
Think about what is really needed with this device and make it clear.
Ideally, in Phase I we would demonstrate that driving in a virtual reality setting is a highly effective enabling tool when used with MRI and MEG. My supposition is that the power of this tool derives from it being a familiar daily activity for many users. How can this be tested? I think we can assume we'd get letters from the consultants raving about the usefulness of the results. But is there anything that can further prove this? Sales would be one thing. Publications. But are there things we could measure that show it is very effective? This is a separate problem from the one researchers confront, at least in part. It is more mundane and can be overlooked when things work, but it's question like: Did the device work? Was it easy to set up? Did it introduce any electronic interference? Does it adapt easily to different software? Is the timing accuracy good? Can we record pedal activity and steering accurately? Is it comfortable for the user? Does it introduce other artifacts, such as head motion or vibrations? A good list of these questions would help clarify the design challenges and also set up the bar for a Phase II award.
In Phase II we would move towards commercialization -- how?
- 17 Mar 2017
I met Ruben this afternoon and talked for a bit with him. He mentioned "RDOC" which is a concept that Ruben thought might be helpful with an SBIR application. He mentioned both Social Cognition and Working Memory as examples of RDOCs, which we might be able to successfully refer to instead of specific mental illnesses or deficiencies since NIMH is interested in pursuing these as better indicators of risk factors, in combination, than looking at specific illnesses. Ruben gave cardiology as an example of where this comes from: the genetic indicators for heart disease are weak, but not for things like cholesterol and hypertension. Those things, when combined in an individual, suggest high risk factor for heart disease. So, he says, NIMH is looking at RDOCs as a way into this for psychiatric health.
- 09 May 2017
It looks like the grant got an impact score of 54: so it was scored, meaning it was in the top half of the applications, but this number is high compared to the NIMH "payline", which is around 10-20. So I should get ready to re-submit it.
First question: what is the next deadline? Looks like it's Sept 5
, which is very do-able.
- 04 Jul 2017
Over a month has run through my fingers and now I have to rush to:
- argue/explain the commercial potential
- argue/explain the innovation
- add human subject testing (read about co-investigator vs consultant)
- add the inclusion form
- add a photo or two from MEG
I'll do some work on this at the beach, too.
Here is the summary statement: Summary_Statement_1R43MH115672-01.pdf
- 10 Aug 2017
Now a few weeks have gone by trying to find a new home for this application because Meg Grabb has told me they don't want it. The trouble is it was written for NIMH and I don't think it can be rewritten fast for another agency but, more importantly, I see it as a tool for mental health research generally anyway, so I plan to send it back to them. Tim's advice was to send it in but ask that it not go to NIMH, and let them deal with which agency takes it up. In a way this may be the same: NIMH isn't going to take it, so it will probably wind up in a pile of scored applications without a home. Maybe someone then will take it. I think there should be text about its application to aging, drug and alcohol, and NIND-ish stuff, too.
- 22 Aug 2017
The NIA program officer has given me a green light, which will at least keep it from being rejected by Meg. But the testing that Tim has suggested will involve ASD kids so I don't know how to present that in a light that will be favorable to NIA. NIDA has also given tentative "we might be interested" approval.
What we're proposing is a tool with lots of good uses, and I think the reviewers see that. But once it's scored the institutes have to see their mission being addressed with some specificity. I don't see how I can do that for NIA if the testing we suggest is with ASD subjects, but maybe Tim will have a suggestion.
I've emailed Ruben to ask if he can line up some older adult subjects.
Page limits: see this
|| 30 lines
| Project Narrative
|| 3 sentances
| Introduction to resubmission
|| 1 page
| Specific Aims
|| 1 page
| Research Plan
|| 6 pages
- CHOP MEG photo or two: crop/gimp for research plan
- write back to NIA PO
- write back to NIDA PO
- 27 Aug 2017
What would be great -
- get a test result behaviorally,
- interpret it with new understanding based on MRI/MEG tests
For example: measure an old person's reaction time and somehow identify its signature as not damning for driving. Or show that reaction time following green-to-red light is one thing, but reaction time from a sideways obstacle (like a deer) is a much better indicator of general slowing. Or show that better executive function can override slowed reaction times.
An older person might naturally compensate for slower reaction time by following at a greater distance and/or driving slower. So the driving test should include some obstacles that pop up to test for that.
The driving simulation is a tough thing to beat, in that it is really the behavioral functioning that matters and this can be measured without MEG or MRI. How does the deeper measurement add value? Presumably, it can reveal the causes of deficiencies and the relative efforts in the brain of, say, executive vs motor function. And it might lead to suggested remedies like caffeine, brighter headlights, massage chairs -- who knows.
- 28 Aug 2017
New specific aims: Design the device to adapt rapidly between MRI, MEG, and office based VR lab settings so that data collected in the 3 settings are equivalent. A convergence of increasingly naturalistic behavioral testing in MRI/MEG setting and growing use of virtual testing in doctors offices and driver licensing centers is creating an opportunity for significant sales.
There is widening interest among MRI and MEG researchers to present more natural task to subjects in testing, coinciding with increased interest in the use of virtual reality and simulations for behavioral testing in office settings. This is a convergence of interest from research and practical/clinical domains: a device that can serve the two should be valuable for research and also widely marketable in the MRI/MEG market but also to places like doctor's offices and driver testing centers. We're proposing hardware that would serve these two markets with equivalent performance specifications and establish a standard for driving. The development proposed here seeks to address the needs of two converging markets: MRI/MEG researchers seeking increased naturalism in behavioral tasks, and office/VR testing labs such as doctor's offices and drivers license testing centers. The novelty of the proposed design lies in its ability to serve these two markets.
This application proposes development of a response device for interchangeable use in MRI, MEG, and office desktop VR settings. It will enable realistic virtual driving tasks to be performed by subjects, presenting a consistent experience in all three settings, producing identically specified output data in a universal standard format (conforming to the USB HID specifications). Such a response device is needed to address the trend in fMRI and MEG research away from abstract/reductionist behavioral task presentation to more naturalistic and ecologically valid ones. Such a device, with well controlled behavior validated through advanced neuroimaging applications, also offers significant value to desktop testing settings such as doctors offices and drivers license testing centers which are served at present by either gaming systems or very expensive proprietary simulators.
Aim 1: to demonstrate that the device works effectively for a wide range of adult subjects, covering ages 20 to 70 and distributed well by gender, height, and weight.
Aim 2: to show that the device works effectively in the 3 testing areas (MRI, MEG, and office/desktop VR testing).
Aim 3: to show time relatedness of acquired brain activity data to the subject's interaction with the driving response device.
- 29 Aug 2017
This is a general purpose subject response device of a new generation where virtual reality and simulations increasingly replace more traditional stimulus/response behahvioral testing paradigms.
- 30 Aug 2017
This tool is a kind of "glue technology" augmenting the power of simulation-based behavioral testing with the underlying spatial brain structure revealed by MRI and the spatial/temporal activity revealed by MEG. We propose innovations to make the device work interchangeably in the three target areas (MRI, MEG, and office/desktop VR testing labs) and, especially, to be acceptable as a realistic simulacrum of actual car controls to a wide range of test subjects including older adults. As an activity of daily life for many adults, driving could become a very powerful tool for passive detection and diagnosis of deficiencies or impairments arising from age, and when coupled with deeper understanding gained from MRI and MEG neuroimaging, might also enable aging drivers to drive, and remain independent, longer. In this proposal we do not make that claim or propose testing it, but instead we propose to test a tool that is needed for that study and to demonstrate that this tool is effective and feasible.
Response device needed for different reasons by all three: MRI, MEG, office VR. Making one that serves all very powerful. VR good for many reasons. Driving in particular good for a few reasons. Making it work well in 3 modes requires innovations in design and careful testing in 3 areas and across ages and genders. To demonstrate its working, it has to be convincing to the subjects but also technically effective. Demonstrate tech effectiveness using autism/MEG feasibility b/c there's an expected outcome there. Use age range to 70 to confirm it doesn't come across as just a game platform.
We propose a new kind of subject response device specifically designed to work equivalently in 3 application areas: MRI, MEG, and office/desktop VR labs. The MRI response device market is something of a niche in which the constraints of MRI compatibility have dictated most of the design choices, and
- 31 Aug 2017