SBIR Topics

2024-2 SBIR/STTR Topics

Transcription provided by Huntsville AI Transcribe

 Anyway, let me share and we will go through the stuff. Let’s see that.

All right. So we are at our favorite site for SBIRs for DoD, SBIRs and STTRs.

There are others.

There’s NASA. There’s we’ve had the Department of Transportation actually has SBIR topics. This is just the ones that I’m usually looking at because of my day job working for Cohesion Force as part of it. And so it kind of collides with what we do here at Huntsville AI. So one of the things, and this is something that Keith and I were talking about before we jumped into this. The filters that they give you here where you can go through the different priorities.

One of the priorities is somewhere in here, I think is trusted autonomy, maybe.

Yeah. Yeah. If you filter by this and then you close. It’s well, hey, here’s 41.

They’ve actually that’s better.

Whoa. Do they just drop these at random times? Not usually.

Usually you can see the reason we did it now this week is because usually what they do is around the middle of April, where you’ll see the open date on a lot of these be 515 or 516.

That means the pre-release was on April 15th. So they’ll do a pre-release April 15th. They give you 30 days to actually contact the author of the topic.

And we can we could draw through that some lay for you familiar with SBIRs or the process or anything like that.

Not no, not terribly.


So I know Keith, you’ve been on the other side of it. Yeah. A little bit. So for the Department of Defense, these SBIRs, which is Small Business Innovation Research or STTRs, which I don’t know what that stands for.

It’s a basically you have to work for those. You have to work with some type of a research institute or a nonprofit lab or something. Usually it’s a college or university where they want, you know, it’s kind of trying to get them involved as well. So about three times a year, actually, it’s actually three times a year. They come up with these topics where they’re interested in getting small businesses to go help do some work. They are a mix of very exploratory kind of work where you may have an idea that you don’t even know if it’s going to work or not. You can you can propose it and say, hey, we need a little bit of money to go figure out is this even feasible. And so some of the phase ones you’ll see are feasibility studies sometimes. So three times a year, they’ll drop them price range runs somewhere between what 120,000 to, you know, 200,000 for a phase one, depending on which branch you’re working with.

If you do and typically the timeframe on a phase one is about six months.

So it’s limited time.

They just want you to get something moving, feasibility study or prototype or something. Then they evaluate it. And if you do well enough, you can actually make it into a phase two phase twos are normally a little over a million dollars around a two year timeframe and actually to build out something useful.

And then at the end of a phase two, sometimes if you’re really, really close or you need to pivot a little bit, you can actually get a second phase two occasionally.

That happens some.

And then usually that what they want you to do is focus on a phase three is after that and that’s not really doesn’t really have a money limit on it.

At that point is where they have associated.

That’s probably what Keith was talking about, where you have a phase, an SBIR that is now getting integrated into some program of record.

And so at that point, you actually are part of a larger program.

And if you if you do it right and you wind up with a product where you’ve also got some type of a license fee or something else included in instead of not just your time, but something to use or maintain the product. A lot of times that’s much more sustainable and you can, you know, that has longer, longer term run. They are part of the proposal for an SBIR. They want you to be able to commercialize whatever they’re paying you to do because they don’t want to pay maintenance on it forever. So part of the proposal has to be, OK, this is great. How are we going to make money on this after we build it? You know, so that’s other stuff they’re looking at. So I’m actually going to clear the filter because this had a lot more stuff than what I had found initially. And you can see pre-release or open if we look at just the pre-release stuff, it’ll show, you know, here are the ones that are, they open in May or they open in July, you know, things like that.

So a lot of these are exciting. Still more than I was seeing, but I was probably just looking at Army. I don’t think I looked at the Air Force and the other branches. Yeah, you get some weird stuff. Sometimes you get one that’s looking at AI for something and you’re like, wait, I didn’t even know that was a use case. You know?

So, DLA, Logistics, all this kind of stuff has AI all over it, but it’s not the kind of AI that I’m used to doing, you know?

So if I just look for artificial.

And that first one that keeps coming up there, I don’t know if you’ve looked at it yet, but it skips phase one.

So yes, these are fun.

So sometimes I don’t mind saying this on record because we are, I’m recording this, occasionally you will find direct to phase two. What that means is you already have to have something that works or that is not, this isn’t a feasibility study. There’s a concept or a process or something. They know it works. A lot of times these are things that the commercial world has already proven out. Now they just want to do it with their data set because a lot of times their data sets for phase two are classified and they can’t, you know, commercial just doesn’t, sorry, we can’t give you that. It’s not on AWS. You know? Other times I found that they’ve got something that’s geared towards a, sometimes you’ll find these and then you do some word, some Google searches on some phrasing and stuff and you figure out there’s somebody that already has something close to it and they’re trying to figure out how to buy it. But. Gotcha.

I wondered if that wasn’t. Sometimes that happens. There’s one in here we’ll cover if we get that far that appears to be directly related to something somebody already has.

Is in almost using the name of the specification.

It’s the one with cameras.

We’ll get there in a second. So this one, AI enabled, this one must have popped in.

This wasn’t even on my list from yesterday.

Actually, let’s, this is just shows, oh, never mind.

Hold on.

Let me clear some of these because this was throwing me. And then let me actually, we’ll come back for, we’ll come back for artificial.

Let me go learning first.

To pick up all the machine learning ones.

I think there are 20.


Some of these anytime you see X in front of it, this has been somewhat of a recent last year or two approach where this is almost a pitch competition when you see X in front of it. So for this one, the one above was basically the finalist for a previous one. I’m not going to really jump into that one.

For this one, what they’re looking for is they’ve got some, they’ve got some kind of a problem that they’re trying to solve.

In this case, they’re trying to find ways to check for vulnerabilities against models that have already been built. So trying to find some kind of evaluation method or some way to, I mean, this is where you’re, maybe if you find something and you run a specific piece of data and it goes nuts on it. You know, you’re looking for things that are possibly outside of a norm.

But for this one, they actually want you to put together some kind of a pitch.

And so, again, a competition.

The cool thing about this, this is also a direct phase to you.

So if you have something that’s related to this, the best thing about the X kind of, I think they based on the X prize or something that DARPA was doing.

For these, you basically go and do a pitch and they make the awards, I think, at the competition.

So like we did a, we put in a proposal.

Short take for the government.

Pretty much.

We put, and it’s some interesting stuff.

We thought about doing it as cohesion force just to get the experience behind, you know, under a belt. So we know what kind of what’s involved and what’s, you know, what it’s like.

But for us to travel and do all the stuff would be a bit much. But again, if you have something already that’s a technology readiness level, I think they mentioned somewhere in here, maybe not this one, but it needs to be at least definitely past the feasible, you know, and hopefully in use by somebody.

But for, yeah, there we go. They’re trying to get from a phase two to a, you know, TRL six or seven, which TRL is technology readiness level. There’s a bunch.

There’s at least a DOD chart.

And then there’s a separate NASA chart.

And they mostly agree with what a TRL level six versus one, you know, one is kind of a, I had a thought in an elevator on the way here.

That’s level one, you know, level seven is basically I have a user base and a, and an existing, you know, support system and, you know, and all that.

So there’s some range you can imagine.

The six seven is pretty far along.


Six seven is pretty far along.

But of course this one, they want you to enter it like a four, you know, for a phase two.

So we put in the first round that comes out like January timeframe.

You submit, I think it was February.

We submitted a proposal for one of those and we’ll find it out probably next month if we want or not. You know, so that’s, that’s kind of how long it takes to get them to get all the stuff evaluated. And then if you do win, then you have to have the further conversation about, about the pricing and they have to make sure that all of your systems are up and up, up to speed and not just, I mean, accounting system has to be the right kind of accounting system.

You have to check your, you know, you have to record your time a certain, you know, there’s there’s all kinds of government stuff that start to play in.

If you are a small business that is not used to going through that kind of thing, they do have a ton of workshops available and they are, they are happy to help you get figured out, get set up and get your account working and, you know, all that kind of stuff. So I mean, I’ll give them to them on that. They are extremely helpful and they, they even do kind of introductory sessions about a couple of times a year that you can log into and go see what it’s like and, you know, actually ask the experts.

So anyway, this one is an XPRIZE one.

And one of the main differences is this, this looks like 2024 South by Southwest Conference in March.

It’s the meta-directive phase too.

You know, it’s just kind of interesting, which they’re trying. But this one again is wrapped around how do you test and find things that might be vulnerabilities in models.

There’s kind of a thread on that through some of these other ones too.

They should do an SBIR on how to fix the user experience on this site.

Yeah. So, yeah, how many scroll bars do you need?

So this one is basically the next round that comes out in October.

Pretty much the same thing, but this one’s actually, I think, related to, they want to do like a risk management framework around AI models.

So what is the risk of a particular AI model?

What is, you know, what are the opportunities for it to give you bad results?

What are the impact of the bad results if they occur?

You know, that whole kind of a thing.

It’s kind of an interesting concept. But again, that pops out in… I can see one if I want it.

Yeah, I can’t do. Not my cup of tea, really, but an interesting… I hope somebody does it. That’s one of those things I’d love for people to do that. It’s just not what wakes me up in the morning. This one was weird. I don’t know that this is AI related at all.

For this one, what they’re looking for is a way to test materials at an extremely high temperature in a way that provides a high throughput.

Because you can imagine if you got to heat something up that high and then get it, you know, cooled back down, you can’t do a lot of them at a time.

Or, you know, again, this, I don’t even know why the reading through it. I don’t know why it was even machine learning related.

Sometimes you’ll see things get tagged to try to help them get through or something. I’m not quite sure.

This one looked super duper fun.

But again, it’s an STTR, so you’d have to find a research institute. I got that.

Yeah, you do.

So some of these might actually be more interesting. So for this one, they actually have some things out there that listen for radio frequency communication. What they’re trying to do is detect when there is a swarm of drones.

Because you can imagine all of these things are they’ve got there are ways to shoot down drones, probably directed energy, other kinds of things that you could do. They’re just really hard to see because they’re usually made out of plastic, you know, or some kind of composite.

They don’t have a lot of metal things that reflect on you. You can’t really see them well visually, especially if they’re, you know, depending on the color and depending on the weather and and is it nighttime type of things. But they do have to communicate. So they have to have some kind of an RF transmission involved. Plus you could also a lot. Most of them use, you know, electric motors that have to put off some kind of a, you know, something. I’ll say. Get her a plot of my wife about dinner. All right. So this one’s actually figuring out how to build some type of an AI model that can detect.

This is when I was actually need to talk to talk to Phil about. They want it to actually work fast enough and with low enough power to actually run it on the edge. So they want this thing connected directly to the piece that’s actually picking up these signals.

But to do that, you’re not running an LLM.

You’re not running, you know, that kind of thing. And if you build a neural network, there are actually some computer chips called near neuromorphic like an architecture.

Think of it instead of your risk based computer chip that just, you know, runs through instructions.

Think of taking a neural network and putting it directly into a chip. Using a specific architecture.

Of course, you don’t do training on them, but you can imagine how much faster that is if it’s all direct hardware.

So that’s that’s what this one’s for.

And I wish you could see this to have a dog slash co-worker climbing in my lap.

This is not fun. But anyway, there’s that. One of these.

This one is, I don’t know a lot about this.

This was one that sounded like it was written towards a particular, I don’t know, company or group that had already done a lot of work. Because I’m not familiar with even the term cognitive radar. And it’s through here some. So apparently there’s an OSWA and there was a phase one that made it feasible or proved it was feasible. And this is an SBIR phase two that wants basically to pick that up and extend it.

The only thing I could think of that would have been done here is if possibly the small company that built the phase one is no longer a small company.

Or some other conflict of interesting appeared or something. And, you know, who knows. But this is this is one of the more interesting ones that looks like it’s written directly to the output of a previous.

I mean, even say it, you know, good news is if they are, it looks like they’re already happy with the approach.

And if your company is able to pick up the approach and have some way to extend it.

It’s, you know, almost no brainer.

One of the other things I mentioned in a second when we get to top when we get to. Okay, so we mentioned earlier about these come out for a pre-release then for 30 days they kind of sit here. And they have technical points of contact that you can actually contact and ask questions about about the, you know, the topic.

Usually it’s good.

If you’ve got something and you’re not sure if they would want to approve it or not, you know, if it’s kind of edgy or something. It’s generally a good idea to set up a conversation because if they are, if they know that there’s something you’re doing that would automatically be a no, they’ll tell you usually, you know, and save you a lot of effort in building up a proposal and stuff and then waiting four months on a, you know, you know, so that that’s actually nice as well. Some of these will have, you know, authors that are really responsive and get back to you.

Some of the branches make you do everything through the SITIS, which is kind of what this where these questions and answers come in. Hang on.

Let me get off. Okay. My coworker was not helpful. So for this one, you can click here and you can actually, well, somewhere in here, maybe it’s there we go questions and answers. Again, user interface.

Some of them, some of the authors will make you submit all of your questions through this, this particular site, which is kind of neat because you can see all the questions and stuff.

The problem is if you have something proprietary or sensitive, like you’re using a SITIS, you can see all the questions and stuff. You can see all the questions and stuff that you’re proposing. You don’t want everybody else to know about it. This is not going to work for you. So, yeah, let’s be aware of that. Others don’t do some branches. Don’t do this particularly. You’ll see a, if you ask a question, you’ll get directed to one, think of it as a contracts person for that particular agency. And what they’re trying to do is make sure that all of the questions are answered in the same manner. So, you don’t really have to worry as much about the public part of that, but it does put another gate between you and the person actually answering your question. In other words, you don’t talk directly to the person. You submit your question to whoever and then they get it and they make sure that, you know, I think they’re making sure that there aren’t, you know, most of the answers you get will be couched in a, well, we can’t give you guidance on which way to go and we can’t tell you that we can’t direct or we can’t do this. But then they’ll like, well, but we have never seen that particular technology approved for use on our system before, you know, which is a pretty good indication that, well, I don’t know if I want to be the first, you know, depending on what it is.

So sometimes they’ll let you read between the lines.

This one, again, STTR for the Air Force.

This one is crazy. They go through an objective for purpose benefit, you know, Air Force Research Lab, which is nice to get included with. They did a lot of stuff, ITAR, then they get to the description and then they finally tell you phase one, you know. I mean, there’s a lot here. So it looks like they’re trying to find ways.

This one would be more of much, much more of a research project, which is a lot more academic type thing. They’re looking at different ways to do. I think they’ve got a double, double great double descent algorithm somewhere in here.

Yeah, some kind of a training phenomenon, double descent.

And so this would be interesting if you were really on the academic studying how neural networks work, all that kind of stuff. One of the problems is that one of the references is reference. But, you know, the other thing to realize is a lot of the people that write these topics also have a day job.

So, you know, and sometimes you’ll get into it and those like, well, yeah, I had to write that quick to get it in because the deadline to get the topic. You know, sometimes it’s actually really, really good to talk to them. Because sometimes it’s like, is that really what you want? Oh, no, did I put that in there? I’m the ones with that are SDRR and you team, you know, one of the things we talked about earlier was some of these things.

You got a hard bunch of people to get it done.

If you get it, right, can can you rely on help from the university to do that? Yes, you actually expect the university to provide at least 50%. I don’t know if it’s 50, 51.

There’s there’s a fairly high percentage that the university has to do the work.

What is that percentage limited? Can we give 90% to them?

I believe you could.

Yes. I don’t think it’s limited.

The only problem is from a like from a cohesion force standpoint, it costs a certain amount of money just in overhead to, you know, to manage the contract to manage, you know, so we need at least enough of a base to cover the extra.

But no, there’s no and especially a phase two, there’s enough, there’s enough meat to actually worry about it. The problem with the phase one for 100,000 something dollars, you know, there’s not a lot of time to split. So, but again, the phase one, sometimes you can you can get by with having, you know, a good bit of work done by research assistant and a small piece, you know, advisory done and helping with the paper, you know, from a from a professor that kind of thing.

Yeah, this one was actually really cool as well. This is actually interesting. Yes.

And it’s not something that cohesion force would likely that’s one of the things I’m actually a CTO at cohesion force.

And so if there are things here that I know we’re going to go after, I’m not going to tell you we’re going to go after it, but I’m not, you know, there’s, I’m typically wouldn’t look at chasing these as Huntsville AI, just because I don’t have a clearance under Huntsville AI, you know, thing like it would, it would cause more problems than anything. But some of these if others are interested and I know cohesion force is not, you know, we don’t do anything logistics, you know, I this would this would be a whole another thing that we, you know, I would probably if if I didn’t need to get involved or try to go after one of these, I would probably just go ask, hey, can I get cleared, you know, conflict of interest clearance for this and be like, okay, or be now that’s too close. Okay, fine. That’s, I’ll step out. But this one was actually pretty neat. They want a some way to to predict things that are about to be obsolete. This includes. Okay, here’s a here’s a good bit of they actually describe in their TRL what they’re looking for.

So they want a proof of concept at a level three.

They want a demonstration in the environment that it’s supposed to be in, you know, things like that.

They’re actually looking at the other thing that some of these care about a lot more than others is the cyber constraints, where this is likely to be running in a classified area at some point.

So a lot of them want you to jump through the right cyber hoops and the right information assurance hoops a little earlier on to make sure that there’s not a surprise later, where all of a sudden you get to it and it’s like, oh wait, that model came from Ali Baba, you know, or whatever, and it’s, you know, you can’t use that one.

Things like that that you got to watch out for sometimes. I was trying to think aware this this one actually had.

There we go.

They’re looking at getting data from like the technical manuals of parts the maintenance logs, market research reports, and any material related to product development and lifespan. So think of this as possibly a multimodal type thing that’s looking at not only the catalog of all of the parts but the the additional material associated with them, and then trying to predict when it may be obsolete or unavailable even. That’s something I’m not sure they put in here that I think this is specifically for obsolescence. But if you were also able to detect something like hey there was just an earthquake in this particular area and I happen to know that most of the chips that Nvidia makes are made in that area. I don’t know if it’s in video.

That was recent. Yeah, you know, maybe built a non AI system for for tracking weather events and catastrophes things like that. Yeah, it was. I think some of the obsolescence people work. Yeah. So that one’s definitely an interesting one.

This one I have no idea what this is.

Try to look at it.

Apparently there’s been a lot of studies that have been done on this one. I think that’s one of the things that we’ve been doing. This one I have no idea what this is. Try to look at it.

Apparently there’s been a lot of studies in a from a nuclear perspective over the last, I guess since the 50s or whenever it was where they actually started doing this.

And what it is is there’s a lot of data that’s out there and they’re trying to figure out ways to make use of this data. So they’re looking for some type of AI approach. Maybe this may be more just straight machine learning, like primary component analysis, things like that maybe things that might be interesting.

Again, this one was definitely much more of a research paper type thing. And I’m usually more on a application level. So I’m going to skip this unless y’all are super interested in it. Yeah, this one also super duper curious. But there is somebody that is in the AI group that does a lot with AI and optics. So I may throw this over their direction and say, Hey, this says I have no idea what to do with this, but it looks cool. And somebody should do it. This is basically I felt that way about. Oh yeah, I mean, they’re cool. I’d love to read about it.

But I have no business working in this one.

This is basically using AI to try to find the kind of or predict the anomalies and maybe maybe apply some kind of a filter to modify the image coming in based on the anomalies that we’ve discovered in some kind of a lens coding or something. Again, really interesting stuff. I have no business operating. I’m not an optical engineer or a physicist. So anyway, that one’s cool.

So if you got any nice folks over there that do that with, you know, or you age that I don’t know if they’d want to do that or not. This one was interesting in certain ways. One that it uses the term digital twin and I can’t find more than two people that agree on what digital twin means. Yeah.

They should have called it digital twin.


What does that mean?

Last I heard that was a new concept called a digital mirror and I can’t explain that one either.

Yeah, that’s been going around quite a bit. Yeah. I don’t know.

That’s a new buzzword.

Yeah. I like can somebody write this down and what it means? I’m not assumed by cognitive. They don’t mean because it’s a digital, like you said, everybody’s definition of digital twins different. You know, some of them are just where like the Microsoft demo they released this week or last week.

We didn’t take a still image and animate it like you’re talking and with your voice and right. But when it says cognitive, I assume that it wants to have the knowledge of the person, not the not the looks, voice. Yeah, I think this is where they want a digital twin. I think in the way they use it is kind of like a digital representation of something.

Yeah. You know, with the right physical, it knows it’s physical parameters.

It knows what kind of it knows how to interact with how it interoperates with the environment.

So what’s funny is I would I would create a supposition that a cognitive digital twin is actually an autonomous agent from the early 2000s is what I would say if you go look up what autonomous agent is.

It’s a thing that has an interaction with an environment and it has some some knowledge base within or context is that they would call it here.

It’s kind of funny. I bet I could replace stuff from my grad school days. Looking through the summary, I think you’re right. And it’s a decision support system that says on the first line.


And I know decision support systems, but I still don’t quite know what this means.


Oh, wait, wait, I missed a scroll another scroll bar somewhere.


Hold on.

I can’t believe this. Anyway.

So that’s there’s that.

Keep moving. Not sure how far I am through the 20.

Autonomous counter counter measures.

So this is an SBIR.

So this is this is similar to the one that we saw before where it was wanting to do one of them was a risk management framework written around a models.

The other one was a way to test for to see if something had been maybe poy data poisoned or had some kind of vulnerability in it. I think this one is actually, I think this is on the offensive side of that, which I’m not quite sure. Or it could be the same thing. Yeah, it’d be the same. It could be the same thing. Just a different agency that didn’t know the other agency was doing something. We see that occasionally. Right.

So, yeah, Wake and Starcraft 2 and Dota really jumped out at me in that description.


So they’ve done some searching on Google. So in other words, they they’ve this is the other part. Apparently the DoD has rapidly adapted these technologies for planning combat missions, missile defense and so forth. Like, please elaborate. Yeah.

You know, I’d love to know about this.

I did see the Air Force did their first autonomous jet flight. Yeah, dogfight. Yeah. Yeah. That against a human versus an AI that actually I saw that in Top Gun.

So, yeah, yeah, yeah.

So anyway, this is kind of this might be a little interesting. Again, this isn’t anything right up my alley. So I can’t really speak too much to how you would do this. But basically, that’s similar to the other. They’re looking to figure out what the attack services are.

You know, can you modify the data that’s used for training?

Can you modify messages in transit? Can you, you know, what, what is it that you could do to perturbed these?

Let’s see.


I saw something the other day where the Tesla auto drive and autopilot thing, if it sees a stop sign, it stops. So if you’re trying to cross the street somewhere and a primary primarily you got a lot of Tesla’s there.

So this is a video of some guy was like wearing a t-shirt with a stop sign on it.

And he would open his jacket stand on the sidewalk and the cars would stop. And he’d cross the street. Like, wow.

I get it.

Yeah, you got to, you got to lean more towards the stop than the fact that it’s a t-shirt. Which I am happy it stopped. Open source academic publication. But this is the longest title I think we have.

But let’s see.

This one’s on technical writing.

Peer review of technical papers. I think this one, that’s what this one is.

This one actually seems reasonable. Yeah, they put thought into this, which is actually nice. And again, this, these have been getting better over the last five or six years. Definitely. I need to go back and do a redo back from like 2019. I think it was the first time we did one. No, it was 2018. We did the first, the first SBIR walkthrough. And there were like maybe eight topics. And I think five of them were blatantly impossible. And I think two were automatically, were already available if they would have searched, you know, and leaving one that was actually useful. But over time, these have been getting better and better and better. So this one is looking for, what’s this interesting to do? Whoever this is, is a writer.

How does one distribute well reason, but ultimately doomed research that tells a compelling and cautionary tale?

You know, that is somebody that writes a lot.

Or they use check to GPT to write it for them.

Oh, I wonder if they use my GPT.

I was thinking.

My GPT two thing.

I did publish that. Anyway, this is great though.

Commercial innovations take advantage. So you could see some things doing some amount of, you know, this, this, this has a LLM kind of written over it.

Some way to automate some, some of the process.

It would be interesting to build something that could look at comments and say, hey, the tone of this is harsh.

It would be constructive if you phrased it in this way. You know, things like that.

Do that with Amazon reviews, things like that from some of the e-commerce apps I built. Yeah. You can get the sentiment of the reviewers and things like that. Right. Yeah. You could do something about a, you know, a little icon at the bottom from like a, like a little devil icon to an angel icon that, you know, or a frowny face up to a smiley face and tell your, I need that for my own emails. What is the level of saltiness?

This particular reply.

Um, yeah, I need, I needed that too. Back in the day. I needed every, I have, I have people I use as filters. Um, this one was the, this one was the one I initially remembered as being the one that was probably based towards a particular company or something that already had something.

That’s the camera one.

Yeah. Yeah. Uh, especially, apparently there’s a theme called event based cameras.

And apparently there are off the shelf event based sensors and a standardized machine vision interface.

Um, and software they want it to work with. Uh, so that’s interesting.

Um, so again, I read this is almost a, somebody probably has a product that’s super close to doing this already.

Um, but I’m not, I can’t, I don’t know that for sure.

Right. Uh, phase one is basically develop a concept, do a design. Uh, phase two is build a prototype. Um, and this would be good if you had folks that were solid in, uh, possibly, uh, designing hardware.

Uh, you could pry. I mean, this, this is the kind of stuff that AdTran does day in and day out.

You know.

Um, but if you had something that, you know, if you had a group that did, uh, embedded chip design, not just to the software, but the actual, you know, architecture of the chip.

Uh, that would be useful.

I don’t have that.

Um, but it was, it was still neat to see that they were looking at it. Uh, this one sounds really cool, but I don’t have the, I don’t have the expertise from an RF perspective. Uh, this is, but think of it as a, in the RF side of the house where you’re trying to transmit data. Uh, you’ve always got conflicts between, you can’t all talk on the same frequency at the same time. So you do skip frequency stuff. You’ve got wide band radio. You got a lot of stuff going on. What this is trying to do is have a context aware as in, here’s what the content of the data kind of is. And here’s what’s going on kind of wider than just what I’m trying to send and then possibly modifying the priority of which messages need to get where first based on knowledge of the context.

I mean, it’s a, it’s a really, really interesting problem.

Um, I just don’t know how to solve it. You know, um, so yeah, again, this one phase one is a concept.

You know, face, you don’t really have to build anything until phase two. Uh, and even then you’re just validating a prototype.

You’re not actually building up too much of a, basically you build enough to show that it’s valid. Um, so that one was really cool.

You see how much we don’t have too many more signal queuing. This one seemed to be somewhat closely related. Uh, I don’t know what queuing really means here.

Um, and a multifunction electromagnetic spectrum monitoring system is what I definitely don’t have enough awareness of this domain.

So I’m going to skip this one unless somebody really wants to jump in.

Um, this one was pretty cool. And again, an STTR, uh, so aircraft, they have to know where they are. Oh, that’s kind of a given.

Uh, and they have to have redundant systems that can tell them where they are.

Again, it’s flight, it’s flight, you know, a flight safety things. Uh, so what they’re trying to do here is find other sensors that might be available that you could infer where you, you know, or, or augment or have. Uh, help know where you are, things like that.

Uh, so basically fusing set different pieces of data into something that helps you estimate your where you are, what your direction is, you know, things like that.

So again, this one was pretty interesting. Um, if you knew somebody that was possibly a combination of machine learning and aeronautics might, might be a good group to go after that, which again, neither of those is me. No, but we have a lot of people in town that. Yeah, I know. There’s a lot of people who do this stuff. Um, this one was kind of neat. Again, an STTR.

Uh, this one is looking across a bunch of different data sets and looking at what, like the, what is the impact of image quality on the actual models ability to do discrimination, you know, things like that.

Uh, is there any way that you can come up with, I guess, an improvement to whatever they call image quality equations?

Uh, you know, things like that.

And then, uh, so phase one, if you’re looking at complexity, image quality, what is the capacity for the model?

Uh, again, tons and tons of theory.

Um, I’m not, and this is, they want you to publish after you do a lot. So this has come up with the research and come up with a plan for research, do the research and publish your paper. Pretty much. Um, hmm.


Anyway, that, I don’t know.

Uh, I would go nuts doing research for that much time.

Um, here’s another STTR.

A space domain.

This one would be cool just to get to do stuff for space. Yeah. And then, uh, if they get the space force back over here in Huntsville. This is definitely one that I don’t know enough about the domain to know almost what they’re talking about, which is that’s another interesting thing from an SBIR perspective. We’re coming into the, so the SBIRs are typically written as in they have a problem they’re trying to solve for their particular system or approach or whatever, or domain.

And then we’re coming at it orthogonally from an artificial intelligence perspective where we know AI, we can do classification, we can do, you know, I mean, there’s, there’s a lot that you can do from an AI perspective.

And you’re looking for the intersection.

And sometimes the domain part of it is so in depth that it would take, you know, I mean, it would take a lot just to get to the point where I understand that domain well enough to be useful.

And I see a lot of, a lot with genomics as well. When we get into some of that, it’s almost a different language.

Finding guaranteed RL control for, this one actually would be a good application to some stuff that came out last year.

So you got reinforcement learning.

And they’re trying to use reinforcement learning to try to go closely behavior nonlinear controls.

So some of this actually apply we were, I’m trying to remember what year it was, we were doing the dirips thing.

And then Amazon had the deep racer challenge that we actually replaced ninth globally in that one, and actually got some credits from 8 of us to, you know, that coming that that funded our entire website and a lot of different things we were playing around with for like a year and a half. Anyway, what what the deep racer challenge was was using our approach to develop a race car model thing using a virtual environment, and then they take that model, and they plug it into a physical real small race car on a real track that’s mapped exactly like the track that you’ve been training on. And they check to see basically they were trying to see how well does the virtual translate to the real. And they’re, oh my gosh that we place ninth but our submission sucked. When you got into the extra I mean the I’ve still got the video somewhere of what this thing is driving around and going all over the place. When it did, I mean we could make laps around the track.

You put it on a real track that’s exactly the same controls in the car the same the same inputs.

But what’s funny if you watch the video, you can see the track from the car’s perspective, and then you can see the wall that’s at the edge of the track.

But then above that you see the legs of the guy that’s sitting in the chair controlling the video. You know that’s not in the virtual world, you know. So there are some ways in RL that have come out a couple of maybe last year maybe year before for basically constraints that you put around it.

As in basically within this narrow segment yeah we use RL outside of the segment there are safety controls that like keep your car on the road, you know things like that. Like I don’t care if the stop sign is on a t-shirt we’re still going to stop. This could be a good application of you know there’s research that’s been done and presented at NeurIPS and in use in other places but likely not for this particular domain. So you could possibly take existing research and extend that which is something that again they whoever’s evaluating this might like it because it’s already in use commercially in places that actually help you. So in places that actually helps a lot with your commercialization plan. That kind of thing. But again I see design theorems and control structures.

Oh my gosh I don’t know.

That’s a lot of academics but again that’s probably why this is an STTR.

So we made it through 20 of them in under an hour. Let me look for artificial instead of learning and see if there’s anything interesting that pops up that we missed.

So there were 18.

And this was one that we didn’t cover. That was one that goes directly to phase two. Okay. So what you said. This one might actually if well for phase two maybe you’ve got this these are typically good if there’s somebody that has something commercially available in the let’s say you were doing in this case it’s acquisition control acquisition process. If you had some acquisition system you were selling small businesses or had somewhere commercially. But needed to in this case you’re going to have to get this thing to where it’s hardened it’s got to be cyber it’s got to you know all of the things to be able to run in the labs they need this. So that I mean that I would be interested in this if it was if it started with phase one but I don’t have an ARP to begin with.

Right. Yeah. So if you found an open source ARP that you could add a module to, or something, you know, maybe try to think.

Do use applications.

You may be able to girlfriend who’s a contract specialist for DoD and she said oh they desperately need this. Oh yeah. As a terrible as a user of the contracting departments. I would love if they had something like this. Yeah, so again leading to quicker awards less protest better contracts I’m all for that. Absolutely. This is doable.

I know. Should have been done years ago. You didn’t need that much AI to to improve. Right. Yeah. Let’s see except for a blah blah blah. Stamps sheets equivalent to a phase one price so in this case you just have to show some kind of a technical merit and feasibility equivalent to a phase one project. In other words, you got to you got to have some way to show that the idea that you want to build for a phase two is feasible. And if it’s if it’s already been deployed somewhere or if it already has a user base, or if it’s our you know I mean that’s where we’ve tried this before I don’t think we’ve been successful I think we’ve tried this before with open source products.

You know hey there’s a there’s something from eclipse that does this and we want to take it modify it slightly and apply it here.

Some of that based on how much you want to modify versus what the current capability is. You know, so that’s kind of what they’re getting at they they don’t want to fund the building of an entire ARP. You know, that probably costs a whole lot more than it would cost to add some AI or some modules or you know things like that but.

Yeah, this would be interesting. They may give you some hints. You might be able to follow natural sector is you that last line there.

Right, little bit of a point you know the direction.

Yeah, one of the other things you could do as a small business would be go to one of the big businesses that build ARP systems or ERP systems and see if they would license something to you know I mean you could some of these ERP systems provide public API is to build your own modules.

Something like that would probably be sufficient. You know I want to add a module to I can’t think of the main ERP system that a lot of people use.

But there you know you have to go find some but.

So yeah, that’s probably it sap is probably the one.

The other thing is if are they are they using an ERP system today. No, no, that’d be a good question for the I can I can find out this. Yeah, um. Yeah, well let’s that will close out the current. Well actually before I do it let me go see I know there was that one let me just do a quick check.

I don’t know if I want to get open topic for persistent experimentation sounds interesting.

Water jet vision Wow. So there are at least two.

That are artificial intelligence but not machine learning.

Yeah, we did that one. I don’t think we covered this one another direct face to though. It looks like the artificial side. I mean there’s probably a good five or six that are.

Separated because a hypersonic definitely didn’t pop on the.

On the other one.

Nor did whatever this is. But that’s really all the time we’ve got for this round if you got anything that pops in you’re interested in having further discussions you know drop a line let me know.