A conversation with Michelle Eshow, Software Lead for Air Traffic Management at NASA’s Ames Research Center in Silicon Valley.

Transcript

Matthew C. Buffington (Host):You are listening to the NASA in Silicon Valley podcast, episode 21. Today’s guest is Michelle Eshow, Software Lead for Air Traffic Management in the Aviation Systems Division. As we like to say, NASA is with you when you fly. That couldn’t be more true during the crazy holiday air traffic season. We discuss with Michelle the fascinating software tools that NASA is working on to better understand air traffic management and how researchers can help improve flying for everyone. Apart from this conversation, you should also check out NASA.gov/Ames and our Facebook and Twitter feeds for info on Future Flight Central, a two-story, 360 degrees, full scale Air Traffic Control/Air Traffic Management simulation facility at Ames that enables researchers and others to test and practice in a simulated environment. We’ll have some images and video up before the end of the week. But before this introduction gets too long, here is our conversation with Michelle Eshow.

[Music]

Host: I always like to start it off with, tell us a little bit about yourself, tell us about how you joined NASA, how you got to Silicon Valley.

Michelle Eshow: Okay, yeah. I grew up in Nevada. I was born in Las Vegas, and my dad was a bandleader and trombone player in Las Vegas, and my mom was in real estate, so I don’t really have a NASA pedigree or heritage or even engineering heritage, but in high school I was really into science fiction and fantasy, and we even had a Tolkien Society Club at our high school, which was intensely geeky. And this was back in the days before the web or internet.

Host: Before the movies.

Michelle Eshow: Well, before even — yeah, before the movies. I think there was a bad cartoon at the time, but nothing good. And so…

Host:I remember that one. Are you drilling into the Silmarillion, is that how far down it goes?

Michelle Eshow: I did read all the books several times, and we would have meetings and discuss it, and it was intensely geeky. But that same English teacher who ran that club also taught a science fiction and fantasy class that I got a whole credit for in English, which was awesome. And I didn’t really know what aerospace engineering was, I just knew it had to do with spaceflight and sounded really cool. Then I was looking around at colleges, and University of Colorado turns out has a lot of astronaut graduates, and even back then it was one of the top as far as astronaut graduates. I’m like, it must be good. If it’s good enough for astronauts, it must be good enough for me to go learn aerospace engineering. And so, I ended up going from Nevada over to University of Colorado.

Host:So, not too far, but still in the west.

Michelle Eshow: No, not too far, definitely the western, yeah. And then in my senior year, NASA was doing on-campus interviews, and I talked to someone from Marshall and someone from Ames, and ended up getting an offer from here. And I have relatives in the area and wanted to stay in the west, so it was just a natural fit to come out here. And the ironic thing is, I really haven’t worked in the space side. I’ve worked in aeronautics almost the whole time. I had three years in space where I was on an assignment over there, but then came back. But my first 10 years or so was in rotorcraft research, and I spent a lot of time in the back of a CH47 tandem helicopter, which was a research helicopter with a fly-by-wire control system on it, which is ironic because it’s such a big, lumbering aircraft. But, that’s when I learned that I have really bad motion sickness and probably would not have made a good astronaut anyway, because I had to take medicine every time we flew.

Host:You can only take so much Dramamine.

Michelle Eshow: Yeah, you can only take so much. I don’t think you can take that every four hours in space. So anyway, did that for a while, and then got asked to come over to the air traffic control side, because I have a lot of software experience, and what we do now is a lot of software focus.

Host:It’s a lot of the stuff that I think most people don’t necessarily think NASA right away, even though NASA’s origin is in aeronautics, NACA, which is what Ames used to be before NASA was founded. For people in the area, they see the big wind tunnels, or maybe they know about the vertical motion simulator or something like that, but moving into air traffic or some of the work they’re working on drones and stuff, some people wouldn’t be aware we’re doing that kind of research.

Michelle Eshow: Right. I think some time in the ’80s, we had a scientist in our group that was really visionary in the air traffic management world, and wanted to take some of the ideas that were made for auto pilots and things like that, and see if they could work on the ground and make things work better. And he was able to make contact with FAA and they’ve come to rely on us as their research arm. It’s one of our I guess you would say missions now, is to help FAA deploy more advanced technologies. We’re really good at the algorithm side, the automation side, and the FAA of course has to deal with the reality of deployment to many different facilities, and making it work in all conditions and that kind of thing.

Host:And so when you came straight out of school over to Ames as an aerospace, aeronautics engineer, and you talked about some of the work you’ve done with air traffic control. Did you have to also have a background in computer science, because you’re dealing with these complex algorithms and stuff, or is it two sides of…

Michelle Eshow: Computer science is one of the foundations of the work we do in air traffic management. I went back to Stanford and got a master’s degree with a focus on control theory, which is also a big part of air traffic management. So, it’s a blending of a lot of different disciplines. I happened to be really interested in writing software, so even in the rotorcraft side I got into that.

Host:Okay, because everybody is.

Michelle Eshow: Yeah, it’s just a lot of fun. I think it’s very rewarding to write something and see if work. And code that I’ve written has been handed to the FAA and is now running in all the air traffic facilities in the US, so I can point to lines of code I know I wrote 20 years ago that are still out there, so I think that’s really exciting. I’m not a computer scientist by any means. I did get to spend a semester at MIT as a sort of sabbatical, which was an awesome thing NASA did for me. Back in 2000, I got to spend a semester at the Laboratory for Computer Science back at MIT, and worked with a professor there who is more involved with high-dependability systems, which was a great learning experience for me, to see how they model the dependability and predict the dependability of complex software.

Host:When you came onboard at NASA, when you first started looking at some of the air traffic control, was there already a group that existed that was already working on that, and you joined in, or were you building it almost from scratch?

Michelle Eshow: When I joined, there was already something well underway, probably eight or 10 years. They were getting close to doing their first field test at Dallas-Ft. Worth Airport. So, the first time they were going to have controllers use the tools in a real environment with real traffic flying. And so I came in, and a couple months later they promoted me to be the software manager of that whole effort to get the tools out, because I guess I’m good at wrangling software engineers and keeping them focused and keeping the researchers from giving us more than we can handle to implement. The researchers have the ideas and the software team implements them, and then we take it through a simulation in the lab and simulation with more complex lab with real controllers, and finally testing in the real world. That was already well underway when I came on, but I was lucky enough to get to watch it all the way through the field testing and hand it off to the FAA, and then five or six years of tech transfer work with the FAA to make sure they understood what we gave them, because it was a million lines of software.

Host:It’s not like you write it all, and you’re like, here you go, guys, take it from there.

Michelle Eshow: You can’t just give them a USB drive and say, have fun. But the software was very — we had a lot of people exchanging. We had people come out here for a year at a time and learn it. And it’s been the foundation for a lot of really good stuff for the FAA, for their whole next-gen program, I’d say, we’ve had a big contribution to that.

Host:I know recently you’d been working on a project, or a program called Sherlock. I figure anything with NASA has to be a fancy acronym. They don’t just — or is that one of the unique ones? Tell us a little bit about that.

Michelle Eshow: It isn’t an acronym, no. It’s just a standalone name that we thought was cool after searching around for a while for a cool name for our data warehouse. It’s more than a data warehouse, it’s a data analytics platform. So we thought Sherlock, he’s really good at inference, he’s good at data analysis in his mind, and so we thought that was a cool and catchy name. Rather than coming up with a tortured acronym, it’s just called Sherlock.

Host:A tortured acronym — and so, what exactly is that. What exactly is that program, or how does that function?

Michelle Eshow: It fits in with our air traffic management research. As part of that research, we get live data feeds probably more than anyone except the FAA as far as maybe 50 or 60 different air traffic control facilities in the US feed us their real-time track data, and where the flights are. And you can go on the web and see that, but we get the real data at the native update rate of 12 seconds or 5 seconds or 1 second.

Host:Much faster than your phone is like, this is where your flight is.

Michelle Eshow: Right. And we get a lot of information about each flight — what its intention is, where it wants to go, how late is it.

Host:Is that just standardized? All the flights are required to put this data out there?

Michelle Eshow: Yes, in the US aerospace system, everyone has to have a flight plan. If you’re going to be talking to air traffic control, you need a flight plan. You need certain information to be published. Your flight plan is the planned route of flight you’re taking.

Host:I’m guessing, is that an API or something that’s available for anybody to grab that info, or is it they have a standardized system and you’re just tapping into that database?

Michelle Eshow: There are many APIs, so unfortunately, the different types of flight data have different formats that we have to blend together. So, they do have standards, but there’s more than one standard, I guess is what I would say. And so, we recognize that we’re getting all this great data, and wouldn’t it be great to put it in a format and a form that people could actually use it to look at trends over the years, or to mine the data for interesting occurrences, like are there things that happened before an incident that could be what they call a precursor to a problem? Can we look at fuel burn, can we look at how much delay there is, those kind of things. And so, because we get all the air traffic data, we also get a lot of different types of weather data. But in the past, before we started Sherlock, it was sort of just spread around our file system and not really very well-catalogued. But now, we have a very rigorous data recording, and then we have a data recording process, and then we move the data into a database, a traditional database, and also we move the data into a big data system called Hadoop, and into a huge rack of computers where people can bring their algorithms to the data. Rather than having to bring the data to their computer, they can bring their analysis ideas to the data and run them in this big rack and get massive amounts of analysis done in a short amount of time.

Host:And so, that’s basically what Sherlock is doing, looking at this deluge of information from all these different sources, and it’s making inferences and figuring out how to make sense of it all?

Michelle Eshow: Right, and to provide that platform for researchers to come in and use it. So, rather than they have to figure out all the formats of the data, we’ve put it in a consistent format. For any one flight, you can look at all the data about that flight, from when it took off to when it landed, rather than having to look at 35 different facilities in might have talked to. And so, we provide more of the infrastructure and platform for researchers to come in and do their analysis. We don’t typically do the analysis ourselves, but we give them a good platform. We have about 40 terabytes of data so far, and it goes back to 2008, so we have a lot of good trend data people can look at.

Host:This is almost kind of in the sense where NASA is gathering all the data together, making it an easily-digestible format so then researchers can actually pull it together and figure out what they’re trying to do.

Michelle Eshow: Exactly right. So, it’s been used for everything from creating good simulation scenarios. If someone wants to test a new algorithm with real-life traffic data, they can use it.

Host:You have years of it, so test it out.

Michelle Eshow: Years of it, so you can like, okay, if all the flights start here, and we turn on our algorithms, what’s going to happen? They can also use it to look at, like I said, what happens when there are storm cells? How close to a storm cell will an aircraft fly? And that tells you how much delay they might have to take depending on how big a storm, because the biggest source of delay in the air traffic system is weather events. Those are worst in the summer. And they calculate that there’s over 10 million minutes of delay per year due to weather events when you’re traveling by air.

Every minute we can save per flight adds up to many minutes of delay savings for passengers and for the airlines too, because every minute of delay costs them something like $70 or something like that. So, it all adds up. And so, we look for ways that we can route around weather efficiently, and so the warehouse is really good for studying those kind of events as well.

Host:We’re recording this in early December. Hopefully, we’ll have it out coming up around that time where it is the holiday crunch. Have you found anything interesting, or have any great hot tips for folks traveling the holiday season? Has Sherlock looked at any of that data to find out the different trends, because there’s the average day of flights going around, but there is that crunch at certain airports.

Michelle Eshow: Right, so the airlines certainly plan more flights during these busy times, and the airlines can plan the flights. That doesn’t mean the airspace has room for them. So, someone’s going to wait. And they put the — of course the FAA puts every staff they can into it, everybody’s on their best game. But if you add a lot of extra flights with a bad weather event, it’s going to become very — it just sort of, the delay builds up slowly and then suddenly builds up hugely in a nonlinear sort of way. I would recommend, for myself, I don’t fly to the East Coast this time of year.

Host:As a researcher.

Michelle Eshow: As a researcher, stay away from the New York airports this time of year, because if you combine the bad weather with the fact that they’re always overloaded anyway, just a little disturbance can make it…

Host:Can really shuffle it up.

Michelle Eshow: Can really start to snowball, as it were, across the whole country. Yeah, we are working on technologies to help with that, to help with better planning and to help the airlines make the best decision. They’re the ones deciding which planes to delay and which to let go depending on where they’re going and depending on what the situation is. So, we want to develop tools for them to help them make better decisions there.

Host:And that’s the cool thing about Sherlock and gathering all this data, because they’ve bene running this thing, and this is just a way you can sit back and see what they’ve done since 2008, and be like, we’ve looked at this, we’ve run these algorithms, so maybe you want to switch it up and try and do it this way, be a little bit more efficient.

Michelle Eshow: Yeah, and definitely we can identify the problems, and then we have other programs that are actually looking at solutions. I’m looking at a project right now for Charlotte Airport. Charlotte Airport has unique problems in that they don’t have a lot of room on the surface for planes to move around. There’s areas where there’s only one-way taxiing.

Host:So, that’s not necessarily the runways, but this is areas to maneuver and get into your queue to take off.

Michelle Eshow: Right, exactly. There’s areas where if there’s a jumbo plane coming in, everyone else has to get out of the way, there isn’t room for anyone else to taxi. So, they get literally a traffic jam on the surface. And then, if you combine that with the delays of trying to go into the northeast corridor from there, trying to fly into Washington or New Yok, LaGuardia or Kennedy, it can become very delay-saturated in a short amount of time. So, we know what the problem is. Now we’re trying to develop schedulers and advisories for the airlines and for the controllers to make better, to make the best possible sequence of decisions so they have the minimal amount of delay. We can’t get rid of all of it, but we can certainly reduce it.

Host:I remember, it was a couple months ago on the podcast, we had an audio version of a story on Sherlock that we put up and posted out. So, folks who are listening, you can dig back into the feed and find that story that we put out. But I remember online, on NASA.gov, there was an image when that came out, and it was the United States, and it had all the yellow and red lines of all the different flights going through.

Michelle Eshow: That was just one day of traffic going into and out of Charlotte, because that’s our airport of interest right now, so that’s the data that we chose to show. That’s just one day of traffic, and it’s an amazing many thousands of flights that somehow traverse that area.

Host:Sherlock is up and running, so what are the next steps you’re working on, or the next phases that you’re excited and looking forward to?

Michelle Eshow: As far as Sherlock, I would say just making it easier for researchers to ask complex questions without having to write a lot of software. So, how do you pose a question you’re interested in solving? For example, tell me the maximum crosswinds that aircraft are willing to land at SFO, or give me the worst days for crosswinds at SFO, and tell me what the average space between the aircraft was, and things like that. It’s sort of like preprocessing all the flights, so they can ask those questions in an easy way and get the answer back and increase the productivity of the analysis, which sounds very dry, but if you get more analysis done, you advance the technologies all faster, because if you’re bound up trying to figure out what the data means, or what does this field in the data mean, if you’re bound up with that, you can’t answer any real questions. So, trying to make the data very consistent and dependable and easy to use.

Host:And I know one of the things they work at, at Ames, and we just came up on the 40th anniversary, was the aviation safety reporting system. Does that play into some of the stuff you’re working, or is that another source of data that gets fed into some of these?

Michelle Eshow: It’s not something we’ve looked at yet, but I think it would be a really interesting source of data. But definitely there are safety researchers here who look at those reports and try to find common threads, and I think they have been used to improve the safety. If there’s things that multiple actors or pilots or whoever complain about, then people take a look at it.

There’s definitely so much to be mined there. And then you add the onboard recordings from the flights, and what’s going on onboard the flight, and the weather radars and all that, and you get a very complete picture of what’s going on in the airspace. How can we use that data to predict what’s going to happen in the next few hours? Can we give passengers a warning that we think there’s going to be several hours of delay on the East Coast in the next few hours, keep an eye out? I think there’s a huge effort now to look at social media too, because the airlines are looking at social media. You see their customer service reps responding to Twitter complaints. They’re also getting notifications about things going on in the flights that are coming by Twitter first, like an unruly passenger or something like that. Someone will tweet about it or post a video before the airlines might even know.

So, working that into the system as well, and then also looking at, what is the traffic jam getting to the airport? Can we tell the airlines your passengers are going to be late, or your air crew’s going to be late because there’s a huge traffic jam? If the air crew’s willing to share their location data with us from their cellphone, we can predict what time they’re going to get to the airport. It’s all about information and being able to have situational awareness and make better decisions based on that.

Host:Especially going into the holiday season, as much as your research and work has helped anybody from spending a little less time on the tarmac, or less time circling an airport, I’m sure there will be people all throughout the country raising a glass to you and the research you’re doing to help people stay out of flight delays.

Michelle Eshow: Yeah, I think so. I think every time you fly, you save a couple minutes because of what we’ve done, if not more. And that all adds up over time. Over 700 million people fly in the US, or something like that, so it all adds up.

Host:Excellent. And then, speaking of Twitter and culling social media, we are on Twitter. We’re @NASAAMES, and also for the podcast, we’re using #NASASiliconValley. So, if anybody has questions for Michelle, or if they have complaints for holiday travel, I’m sure they can just ping you. Look into ways to hopefully do it more efficient the next time. But, thanks for coming on over.

Michelle Eshow: Thank you, Matt, that was great. Thank you very much.

[End]