In this episode of SciTech Now, define the odds with robotics; computers trained to identify wild animals and becoming an astronaut.
SciTech Now Episode 528
Coming up, defying the odds with robotics.
Even after being as far down as he was, to come back and come out with a victory on top of that.
Hopefully that lesson he learned there sticks with him through life.
We did it!
Computers trained to identify wild animals.
You know, if we have about 10,000 images of a species, we're going to get pretty good image recognition.
Becoming an astronaut.
We have everyone's favorite, I think, learning how to use the space suit.
It's all ahead.
Funding for this program is made possible by...
I'm Hari Sreenivasan.
Welcome to our weekly program bringing you the latest breakthroughs in science, technology and innovation.
Let's get started.
Students from Bassett Street Elementary School in Los Angeles, California, defy the odds by making it to the World Championship of Robotics.
Filmmaker Vincent Precht followed their journey in his film 'Never Give Up: The Journey of Bassett Robotics.'
So how did you get involved with the robotics team?
Just... It was just a fun activity that I would want to do after school.
I like it because it's, like, racing games and playing on the console, like a PS4.
My mom was retiring from being a teacher, and she's 73 years old now, and the only reason why she retired was because technology was slowly taking over, and she couldn't keep... Not slowly, quickly taking over.
She couldn't keep up with the technology, and so I saw a part of her start to disappear because she had such a passion for teaching.
An important part about being a teacher is to keep up with the times, and it's because I knew nothing about robotics that it became a little bit interesting to me.
This is our living room, where we play all our games and... [ Indistinct conversations ]
Let's do snipers, dude!
This is me and Moredecai's bunk bed.
It's my older brother's on the top, and then this is my younger brother Cameron's, and this is Michael's bed.
When you see us, we have four boys, so we're looking for scholarships, and, you know, on top of them being... doing well academically, Devon also plays basketball.
He is also really into the dance that they have at Bassett, did the Science Olympiad last year.
He did the Best Foot Forward, and so he's always involved in activities.
I was doing it the wrong way.
What is the regular way?
Devon didn't start speaking until almost 4 and 5.
He had a hearing problem.
He still kind of does.
We had to deal with it pretty late, but that also ended up delaying his speech, so for Devon, he's had a lot of uphill battles.
So are you a very technical person?
Not at all.
I mean, I could not set the DVR.
I'm the type of person that, in my car, I'm embarrassed to say that when the time changes, it stays that hour until the next year because, you know, by the time I remember to pull over, turn off the car and try to figure out how to do it, it's easier for me just to say in my mind, 'Oh, it's an hour behind,' so no.
I'm not technological at all.
So we would...
I made half of it.
You made half of it.
Can I see it?
Hailey, stay focused.
One of the things that we needed to start with was making sure that everybody understood how serious this was.
It's not a babysitting club.
It's not where you can just come and do your homework.
You know, I wanted the students to feel like there was an objective and that every practice had a purpose.
What do you want to practice on?
What do you think that you need to get better at?
We could try skills again.
Skills again, only this time switch off?
Ten, nine, eight...
Nice, get on the bridge.
...seven, six, five, four, three, two, one, 30 seconds.
Then I started to do some research on building a robot, and how do you develop one, and, you know, what part goes here and what part goes there, and that's when the Morenos came into play.
I knew the Morenos, Mr. Moreno, because his son was in my class last year, so I knew that they were really into it, and so I sought him out, and he became my mentor.
So there's a high goal and a low goal.
The low goal is on the floor.
It's below the high goal, and those are worth three points, and then the high goal is five points, and then when you knock the balls over, one point.
If two robots are balanced on the bridge, it's 25 points.
It was really a picture-perfect scenario.
It was like the underdog story almost.
You know, they already had won their section or regional here to go to state, which I was already impressed by, and, you know, I was very proud of them for, but when they go to state, they're playing against the best of the best.
You know, and some of the kids have probably been doing it longer, may be more experienced, maybe even more support, right?
So I wasn't sure what to expect, but, you know, I told Devon, 'No matter what, you try your hardest, and win or lose, I'm proud of you.
You did a great job.
You're one of the best in the state, period, you know, and you have an opportunity.'
So they get there, and the first few drives were great.
They probably had one of the highest-scoring drives in the very beginning, really setting the tone for the rest of the day, and there was competition after competition.
It was nonstop for 6 hours, 7 hours straight of them practicing with their robot, going, playing with the other teams, going to do a presentation upstairs that they had to do in front of the judges explaining some of their, you know, experience with it, so there's more than just the driving.
They did it all within a few hours, to the point where they started getting tired.
You know, the kids are hungry, and all of a sudden, they start playing with other teams that were not doing as well as they were, so their score started going down because you don't just have your own personal score.
You also score with the other teams as a team event, so they went from being in third place of the competition all the way back down to eighth or ninth, and at that point, it was pretty rough.
You know, Devon had a hard time.
Devon gives everything into it.
He wears his emotion on his sleeve, right?
So he was frustrated.
You could see it taking over him to the point where, you know, he went to his mother and started crying just a little bit, for a second, you know, because he was so emotional about it.
I thought it was over and we didn't do so well, and I just didn't like the way it went for us.
The whole team got up because we were sitting in chairs.
We told Devon, which is the captain, that we're not going to lose because we know that we can do this, and we won the other competition, so we might win this one.
I remember saying to him, 'Devon, when they asked who the captain was, everyone in your team pointed to you, right?'
And he says, 'Yes, Dad.'
I was like, 'Do you think you're acting like a captain right now?'
He's like, 'No.'
I was like, 'All right.
Let's do it, then.
Get up and show your team that you can come back from this.'
And it was like a cartoon.
It was really very magical to see it.
Just, like, I almost like pixie dust was just coming off of them, and they were able to get it together.
Six, five, four, three, two, one.
The scorekeeper is ready, so we're going to start, point and smash number eight in three, two, one, go!
♪♪ Point and smash number eight, we have O.H. Hawks from Orchard Hills Elementary in Irvine, playing with Bassett Robotics from Bassett Street Elementary School in Van Nuys.
Here we go.
[ Speaks indistinctly ]
Well done, excellent!
Now going back for a few more.
♪♪ [ Cheers and applause ] With 12 seconds left, and see if they can get both robots up there, and yes!
[ Cheers and applause ]
It was amazing.
One of my sisters brought some confetti cannons, and not just one.
She brought, like, five of them.
And all we see after when we hear, 'Bassett Robotics, they won third place,' all you see is a bunch of confetti flying all over the place while we won.
I felt really good and just surprised too because I didn't know if we were going to win or not, and it's just really happy to.
Filmmaker Vincent Precht joins me now.
First of all, why make this film?
I was a teacher at... I am a teacher at the school, and I saw this wonderful kind of energy happening at my school because of robotics, so I just tagged along.
Well, these are not the kids that you automatically assume.
They didn't come from households where science and technology is practiced on a daily basis and role models in their lives.
They don't have the extra computing power sitting around in their basements.
Give me an idea of how this interest got spurred in the first place.
I think that one of the children that I follow in the film, his father was a basketball player.
He plays basketball.
He has four brothers, and they live in a very small apartment in Van Nuys, and the mother said to me -- it's in the film -- she says, 'You know, we're looking for scholarships,' and so these children just are so thirsty to become involved in anything, in any kind of opportunity to, you know, to further themselves and to lift themselves out of where they are.
What did you see when you were watching these students as they evolve through the competition?
What was happening to the kids that is beyond the competition and beyond the robots?
Most of it was beyond the robots.
They, you know, they... Going to Kentucky, which is where they participated in the championship, it was a completely new experience.
Many of them had never been on an airplane.
Some...I doubt many of them had been to a nice hotel, where they stayed, so it was just kind of learning how to deal with the outside world outside Van Nuys.
And so, you know, a lot of times you hear about that mind expansion that happens when they literally are shown something beyond their neighborhood.
It's kind of a whole new planet almost.
When they came back from that competition, what are the changes that you and the other teachers started to see?
Well, I mean, the changes in them, it was a process.
It happened as they started to get involved in the robotics and continued on, and mostly, these kids were just very driven, and it was sort of this door that had been opened to them so that they could, you know, show their talents.
Is there a change that you see in the school?
I mean, do these programs now have more, I don't know, street cred, cool factor, whatever you want to call it?
I think so.
I mean, getting back to the boy who I was talking about, I asked him in one of the interviews, you know, 'What do you want to do when you grow up?'
And, well, of course, he wants to be a basketball player.
His dad was a basketball player in college, and he said, 'But then I want to do robotics, and I want to help people with the robotics,' so as far as the school goes, it just enlivened the entire school.
I've been there now since 2001, and it, for years, it has been about testing and English language and math and very important subjects, but to the expense of science and art, so this has been just, like, a renaissance at our school, where we now have technology labs, robotics labs.
I have a TV lab where I do, like, a news magazine with the kids
What are the things that robotics are teaching the kids in the sense of just critical thinking or reasoning beyond what they're going to apply on the competition field so to speak?
How do they problem solve now?
It's on the fly because of the competition.
It has really required them to think on their feet and to jump in if something goes wrong, to solve the problem, to fix the robot.
You know, as the robotic competition goes on, the robot wears out, so now we have to replace parts and that kind of thing.
Filmmaker and teacher Vincent Precht, thanks for joining us.
Thank you very much.
Mikey Tabak is an ecologist and scientist.
His research has led to the development of computer systems that are able to identify wild animals in camera-trap photography.
He joins me via Google Hangout to discuss how the system is being used to understand animals and wildlife.
So first, explain, you know, trap photography in the wild for people who might not know.
So one thing we do in ecology is we go out, and we try and observe animals in their natural ecosystems, and this is often challenging because, you know, we have to have people out there to observe the animals, and when you have people out there observing, they create noises.
They're visually stimulating to the animals, and they have smells that might actually repel the animals away from those, away from the scientists that are in the field.
So we set up camera traps, which are motion activated, so that when an animal travels past the camera, it automatically triggers the camera to take a picture, but when we have these cameras out in the field, they can...They get triggered by any animal that goes by.
Sometimes those animals aren't of interest.
They also could get triggered by wind or grass or leaves blowing by, so we end up with a whole lot of images that we need to analyze in order to use them for ecological studies.
So here you are back at the lab.
You've taken out the card from the camera.
You're putting it into the system, and suddenly you have 1,000 pictures, only three of which are the animal of interest.
So it used to be that somebody has to sit there and go through the other 997 to find the right ones.
Yes, and that was the big challenge that we were having.
When I was working US Department of Agriculture, we were getting millions of images from across our field sites, and so we had volunteers.
We had all kinds of people looking at these images, and it was taking an incredible amount of human time looking at computer images, at images on the computers, and so we needed to figure out a way to make that system more efficient and to try and alleviate some of that stress that we were putting on our interns and volunteers.
So how did you do it?
Well, so we got involved with some computer scientists at the University of Wyoming, who were working on a similar project, and we developed these models, the machine learning approach called convolutional neural networks.
These artificial neural networks are designed to mimic the behavior of a mammalian brain in order to learn how to do complex tasks, and so in this case, we have a whole bunch of images that were already identified.
We had over 3 million images where we knew the animal that was in the image, and we used these images to train the model to automatically recognize the animal species that's in future images.
So a fox in the wild does not look like a fox that's in a biology textbook or a perfectly cropped photo from a nature documentary, right?
So how many different pictures of foxes do you have to have to be able to train an algorithm well enough to say, 'Okay, this is probably a fox'?
Well, that's a good question, and it's going to depend on the species that you're looking at, the background of the species.
So, you know, you don't just see a picture of a fox.
You see that fox in a forest.
You see that fox running through leaves and grass, so it's really going to depend on the animal.
In our case, we had close to 2 million images just of cattle, and so our model was really good at recognize cattle, but we found, you know, if we have about 10,000 images of a species, we're going to get pretty good image recognition for that animal, and, you know, 10,000 is a whole lot, but we also found in some cases we only had 2,000 images for a species, and the model did pretty well at recognizing those species too.
One of the things that needs to be considered in the future of these types of studies is we want to have images of those species from different environments, so in our case, we were lucky to have data sets from five or six different field sites, and so by using those different field sites, we can train the model to have sort of a diverse search image for each animal, for each species.
So it could be that fox in the snow.
It could be that fox in the sun.
It could be that fox on a rock.
It could be that fox next to a tree.
Right, I get it.
...the underlying kind of system that you've built, how can that work in other places, or how is it working in other places now?
So it's also being used in applications for all kinds of vision... for computer-vision technology, so one of the people who worked with us on this project, or actually two of the people who have worked with us on this project are currently working on developing self-driving vehicles, and so they're using the same type of technology to recognize certain things that a vehicle would experience on the road in order to have the proper reaction, you know, while a vehicle is driving itself on the road.
So a stop sign in 100 different types of environments and lighting, different conditions and in front of and behind trees, et cetera, et cetera, so that the car could recognize it.
Right, and, you know, and other hazards on the road, other vehicles.
You know, a child runs out in the road and the vehicle, you know, is going to need to recognize those kind of things as well.
So what's the challenge here?
I mean, is it just about that acquiring the initial data set?
Is it about the initial amount of work that goes into determining what is and what isn't something?
So one of the big challenges is getting these data sets that have been observed and identified and have all the images identified.
We want to make sure that we're dealing with a good data set to start out with so that, you know, we're training the model on proper... on images that are correct.
We did have some instances that I checked where the model was wrong for a certain image, and I looked back at that image and found that the human observer was actually wrong, and the model did a better job of recognizing what was in that image than the human did, so that's one of the greatest challenges, I think, is developing these large image data sets, and that's really a common problem in the machine-learning computer-vision field, and so one thing we did with our paper is we shared all of our images through this new program that Microsoft is offering.
It's a free program that allows different users who have these types of data sets to share them publicly for other people who want to work on similar machine learning problems.
So I'm assuming now that there are a lot of happy interns at the USDA that don't have to go through thousands of images of cattle.
Yeah, so that's the idea is that we're saving, you know, so people don't have to spend their time now doing this kind of stuff, and they can spend time, you know, doing field work or doing, you know, actually learning more useful skills.
You know, the skill of identifying species on a computer isn't all that useful, and so we were... We're kind of trying to train our interns to do things like mathematical modeling instead.
You know, things that might be more useful in their careers.
Mikey Tabak, thanks so much for joining us.
Oh, thank you for having me.
Each year, tens of thousands of people apply for NASA's astronaut training program, but only a handful are chosen.
The candidates who are accepted begin a two-year, five-step training program at the Johnson Space Center in Houston, Texas.
Let's take a look at how these aspiring astronauts prepare for their journey into space.
This is building nine, where we have a whole one-to-one, life-size mock-up of the International Space Station.
Take me back to when you were first selected to be a part of NASA.
What did that feel like when you got the acceptance letter?
It's actually a phone call, and I was just so much in shock that I said something very eloquent like, 'Really?'
We're going to be welcoming the new class of astronauts.
They will embark on this 2-year astronaut candidate training period where they will be instructed and trained and then demonstrate proficiency in five main areas in order to graduate from being astronaut candidates and become full-fledged assignable astronauts.
We have everyone's favorite, I think, learning how to use the space suit.
We do that in the Neutral Buoyancy Lab.
You're weighing over 400 pounds when you're in the suit, so every moment, you know, every hand movement is like squeezing an exercise ball.
Then we have robotics training, so we learn how to fly the Canadarm, the big robotic arm that's on the outside of the space station, so we still use it on a routine basis to either reconfigure the space station, using it for space walks, using it to do different tasks or repairs on the outside of the space station.
The third area that we have is in the International Space Station systems.
When we're up there, you know, if a light bulb needs to be changed or the toilet breaks, and that does happen quite a lot, we can't call the plumber and electrician.
We have to have a really diverse set of training so that we do that, all of that ourselves.
So the fourth main area is Russian language.
It's an international crew up there.
Everybody has to be able to speak both Russian and English.
And then the fifth main area in which we have to become proficient in order to graduate is flight training.
So this is Ellington Air Field, and this is where we have our T-38s, and we do something called Spaceflight Readiness Training.
It is the most real-world thing that we do other than the actual flying in space, and so it also requires us to learn to prioritize, make decisions and communicate effectively, so it's a really dense type of training that we get to do.
I mean, yes, it seems very important, but it also seems really fun.
I mean, dangerous but fun.
Dangerous but fun, that's a good description of flying a T-38.
And that wraps it up for this time.
For more on science, technology and innovation, visit our website.
Check us out on Facebook and Instagram, and join the conversation on Twitter.
You can also subscribe to our YouTube channel.
Until then, I'm Hari Sreenivasan.
Thanks for watching.
Funding for this program is made possible by... ♪♪ ♪♪ ♪♪ ♪♪ ♪♪ ♪♪