Welcome to NVIDIA's AI podcast. We've talked about autonomous cars or self-driving cars before, but today we're going to talk about self-driving cars that are put together with little more than some smarts $1,500, and I don't want to call it a pile of junk, but our guest can tell us all about that.
Today we have with us Jim Burke, the founder of the Power Racing Series and its chief what?
Propellerhead. Jim, welcome. Hi, how's it going?
Thanks for having me. I know the Power Racing series now.
This is something that we should describe from the beginning, but this is where...
You get souped up Barbie Jeeps and whatever electric car of your choice, and you race them around a track.
Is that correct? That's right. Since about 2010, we've been working with local hackerspaces and community makerspaces to build and modify $500 electric goat carts, the human-driven variety.
In the last year or so, we've been slowly migrating to the autonomous realm because as we evolved, we wanted to play with those new technologies.
What made you think that that was even possible?
Because there's been, again, there's these huge companies. building driverless cars and they're having a time of it.
There was years ago, there was the DARPA challenge where universities and others spent hundreds of thousands of dollars to mostly crash their cars in the desert.
So how is it that we've come so far that you could with $1500 or less, put these things together.
Well, it's something that's definitely only been possible in the last few years, mostly because the kind of sensors that you would need to do that have become incrementally cheaper or rather the sensors that are cheaper are getting good enough that you could build a small vehicle with them.
We've seen some pretty great experiments and results with small RC cars being able to, their depth sensors, the lidar that they use, and those are a lot more cheaper systems that you know just a few years ago were tens of thousands dollars you're getting around to the you know a couple grand and then there's even more sensors being put on the market that are a little bit slower.
There are just hundreds of dollars. So it seemed like a good time to kind of jump into it and start experimenting with it.
So describe one of these cards for us and describe kind of the parameters if somebody wants to get in on this, as it were.
What do I start with and then what are the rules as you've laid them out for people to do this?
Well, it's an incredibly exciting time for Power Racing series because We've spent many seasons developing our human piloted league, you know, just the power racing series.
And year after year, we get more and more teams sign up.
And every year at the end, we have a summit where the teams that participated contribute to the rules.
It's kind of What I would probably say is a benevolent dictatorship that I found.
Right. I listen to all the teams. I take their input and then we kind of come to a consensus on what direction we want to take things.
And after several implementations of this that we just wrapped up our eighth season, we've developed a formula that's become incredibly exciting to watch. watch.
Very close racing, very entertaining, most importantly.
And when I saw a lot of the contests for autonomous vehicles or autonomous robotics and would watch them.
I came across the same thing that always would get me interested in that There's something really neat here that I think that a general audience would be entranced by.
But it's just not being communicated very effectively.
There's a lot of like student based contests out there. where the rules are extremely esoteric from an outsider's perspective.
It takes a long time to explain how it works, why it works, happening, who's winning.
If you can't explain that in less than a few seconds, you're not going to be able to captivate an audience that would otherwise not be interested in learning this sort of thing.
So this is... I just started to interrupt, but this is equal parts...
Science technology kind of rolling up your sleeves, but also entertainment.
Well, yeah I mean one of the things that I've talked to with a lot of our high school teachers teams that signed up in Power Racing Series.
A lot of them do a lot of robotics contests and competitions throughout the U.S.
The vast majority of our younger teams that are in high school are teams that pursue FIRST Robotics in that.
And I would always ask them, hey, why do you...
You know, we're not first robotics at all.
We... you're building a car out of junk.
And they're like, yeah, but we're building a car from complete scratch.
We've never done that before. I want to have a league where you're given that much potential responsibility to build something from the ground up because you learn so much from doing that. and the only reason that someone's going to want to sign up to it is if it's fun if it's not fun no one's going to want to take the insurmountable task or seemingly insurmountable task of building a go-kart understanding its steering geometry understanding the you know how to you know best manage the power versus the weight versus the what kind of Battery technology you want to use, like how can you keep that under budget?
And then the car has to be able to perform.
It has to manage corners. In our league, you have to build the car to be ridiculous and entertaining looking.
You know, there's themed cars. There's a Back to the Future car once.
There's a car that was a horse. And those are rewarded points from the crowd.
We actually have an Arduino controlled voting board that is handed out throughout the crowd.
They vote live. So you get style points as well as speed points, I guess, as it were.
Yeah, and they're almost... Almost evenly weighted.
I'd say that the style points are just about 10 to 15 percent less valuable than the actual race finishing points.
And we want to do that because. building a cart from scratch is very difficult.
And to incentivize teams that are new they can claw back some points by building something ridiculous while they develop their car.
That keeps them in the game and also it makes it so that it's social engineering, not just good old engineering.
Right. Was it a natural step then? Like, are these cars that people are building Without drivers, are they similar to what they would've run in the Power Racing series with pilots, but now, okay, We're going to do it and we're going to add some machine learning system to do the driving.
Yeah, that's basically how it started. However, the first few editions of these vehicles were more on the lines of, hey, here's a few teams that are, have done really well in the power racing series in the past.
Here's a few teams that have proven themselves time and time again to build really good cars and we want to build an autonomous league.
Can you build the first few for us? And we'll test them out at a few events.
And, um, So the first couple that we've built have been pretty specialized as of last year, but now, Now that we had the ABC Spark Fund contest this past September in Colorado, their autonomous vehicle challenge.
One of the requirements is that you should be able to have a human sit inside of your machine as it drives around.
And hopefully that human doesn't become a crash test dummy.
So have you have you sat in one of these cars and been that human?
I haven't during an actual contest, but I have sat in it bravely to see. just what I'm getting people into.
And it's a very weird feeling to realize that, you know, nobody's technically driving this thing and I'm in it and I know the person who sat there and coded it and they're laughing.
So that's probably not the best indication of this turn coming up right now.
Wait, describe that for us. Are these cars like...
Google's version where there's no steering wheel because why would you need one?
You're not doing the driving. I mean, are they stripped down in a different way than we see... the likes of Uber and others doing because they have different standards and things that they need to do to get to their version of a driverless car?
Yeah, it's a little bit of both. We've kind of had no real listing of whether or not there should be proper steering input by the driver or pilot, depending on what setting you have.
However, We have all decided that in the next coming seasons that the vehicle should be able to be switched from autonomous to drivable.
And we've already had one car built. in colorado that does display this though i have to say it's autonomous uh setting at Well, what what what happens?
Were you heading off into the mountains when you were?
Yeah, it has. It has a proclivity to hay bales that I don't think the programmers intended it to have.
I see. I see. It locks in and heads right for them.
How fast do these cars go and what kind of a track are they navigating?
Is it the same one as the Power Racing Series with drivers navigates?
So they're kind of around the size of a go-kart track.
So picture a go-kart track and your friends going there and It's basically that kind of race course set up with either hay bales, a really complicated system. them of tires or even just go-kart barriers if we can get them.
And The autonomous vehicles, I will say, are not very fast yet.
I was going to say, because in theory, they should be faster.
There's less weight on board, right? And there's nobody to be like, no, no, no, don't do that.
I would have to say that most of it's on the air of the caution of the teams learning how to make their cars successfully complete laps at speed.
A lot of the teams have told me that they can go faster, but they're scared of it so far, mostly mostly because they know that they have to prove it working very effectively lap after lap after lap.
And we've simply haven't had enough contests where we've been achieving that.
We've had tests on the side where we've had pretty good success with it.
New York last year, we actually had Actually, New York this year was pretty successful too.
We had cars that were just sitting around completing lap after lap. um you know about a few laps at a time and without any input but it does take a lot of work and you know because we're operating on such a tight budget you're not relying on systems that normally would solve these problems.
And it kind of introduces a unique challenge to it, mostly in regards to how do you make junk work?
We're kind of working our way down that pipeline. and finding cheaper and cheaper sensors to experiment with and having various types of success.
And it's kind of built on different strategies.
We've had teams that... actually our most successful teams, I'll be the first to admit, was just from dead reckoning.
Our first ABC challenge event probably went about as well as DARPA's first challenge.
Which is to say these cars drove off into the wilderness.
Yeah. And then or they just went five feet and fell over.
Yeah. I mean, it's and that's kind of what.
What it takes, though, I mean, that's part of what makes it work is that you have to fail spectacularly before you get anywhere.
But I mean, right now we have, you know, a few vehicles, you know, we have, you know, 270 degree depth sensors that, you know, are.
They map out the obstacles manually with a controlled lap, and then they try to repeat the course.
So it's, you know, it's basically slam mapping.
I see, because there's a, I hesitate to call it a toy because it's so cool, but a toy car system called Anki Drive.
The way Anki works is that the cars are on a track that's basically mapped and they have a camera so they know where they are on the track.
And I'm just wondering what approach... these cars take to getting around that track.
Two of our best vehicles are mostly using SLAM.
That seems to be one of the most effective, well-rounded ways of getting around the track.
But we also have one of our teams, one of our very successful power racing teams, the FUBAR Labs over in New Jersey.
They built, basically, they have a webcam that has a genetic algorithm running on it, and the car keeps attempting to learn which turns it can and cannot take now.
Granted, that one is an extremely slow vehicle and also he has to kind of fix the steering servo on it because it is completely molasses, but Once he gets that car better developed and built on a stronger and faster chassis.
That car can be incredibly fascinating because he built it with a bunch of spare junk sitting around.
Most of the vehicles are built... fairly cheaply.
And the minute they get something to work, they try to find cheaper parts to slap on it to see how much they can go in costs with getting the same results.
Take us under the hood a little bit. Are these parts scavenged again from like Barbie Jeeps and, you know, laptop batteries?
And, you know, how how does it come together?
A lot of the chassis are built from the ground up.
They're usually welded frames. I've seen a couple teams desperately use like 80 20, which is not recommended because it's just not the right kind of strength and weight. balance that you're looking for not to mention the cost right so they so they're custom frames um and then how do they come to life right so a lot of them have on you know they'll have a laptop mounted where either the driver is or the driver is going to be.
And sometimes a potential passenger slash driver will sit in there with a laptop.
Kind of hunched over because they are children's power wheels, by the way.
Getting inside of them is a tricky thing without a pile of computers and wires hanging from everywhere.
And yeah, so basically one example is like MIT's Gigatron.
That one is actually using a bunch of depth sensors a bunch of lidar sensors hooked up with a you know, GPS and an accelerometer and that's all plugged into a Jetson and that's using SLAM as well to, you know, find its way around the course.
Right, so it's mapping the course as it goes and then hopefully getting it right.
Right. So we have a qualifying lap where it can be human guided through the course.
And then when the flag drops, you're supposed to be able to just hands off it and go from there.
So what happens when you're supposed to hands off it and it goes from there?
And what's happened over the time that you guys have been doing this as it's progressed and gotten bigger? in theory, better.
In theory better is a good way to look at it because the conditions what's really interesting is a lot of people have success with this stuff indoors because you have control of lighting and in controlled environments.
But because we're doing this with an outdoor track, that suddenly becomes a lot more challenging.
A really, really sunny day can wreak havoc on certain And as a result, you can get some almost comedic results.
We've had a lot of luck with MIT's Gigatron and Hack Pittsburgh's Slamberghini in New York last year, and actually this year as well.
We were able to get both of them running and completing laps on their own.
And we came across the interesting fact that none of them were really programmed to sense each other.
And whenever they approached one another it was kind of like watching dogs sniff themselves.
And kind of waltz their way around each other.
So we were able to have overtakes. It was just very laborious. and slow overtakes.
I would say no more faster than a brisk walk.
Oh, that's interesting because they didn't see each other as kind of, competing agents on a track, but as obstacles.
Yeah, so they would start fighting each other by moving around each other, and the other one would try to move around the other.
And occasionally, one car really enjoyed just... like pushing the other one out of the way while the other one would like fight wheel spin to get back around.
And I have to say, like, you know, when everyone's hands off of that, even that slope pace was really exciting to us, even though I know that a crowd, well, a crowd did form anyway because they were just excited at the novelty of two autonomous vehicles hitting each other.
From an entertainment perspective, you know, we have a long ways to go.
Are the brains all similar or are they all custom?
For the Autonomous League, we've had two systems that are using SLAM.
One of them is using a Jetson. The other one is using... and just a pile of onboard computers.
And another one was using a pile of laptops, you know, hooked up to a webcam running, you know, genetic algorithm.
So like different types of sensors. We even had someone using like echolocation combined with a LiDAR system.
That one didn't work nearly as well as it could have, but it was also really funny to watch it go around the track and just fail at making one particular turn.
It would get like four turns done and then just like completely 180 and then into a wall.
Oh, it's interesting. And I'm sure they're like, why that turn?
Huh? Yeah. And they spent the whole weekend trying to solve it.
They eventually did. And then they found a new turn to. crash into and then they never really got past that because just simply ran out of time but you know that's part of you know the process really What's the larger lesson of this race series and how do you see it? going from a race series to people's careers or maybe even their approaches in the careers that they already have.
That's really the big goal right there is the idea that You want to have an educational program that shows people something that's immediately valuable to the public and immediately something that they could say graduate and people are already offering them jobs and that's really kind of one of the driving forces behind even pursuing this in the first place I mean Granted, a few years ago when we did this, we wanted to do it out of the novelty of how it's kind of fun to build these all electric under $500 essentially goat carts that are. you know, Power Wheels in name and they were built from Power Wheels, but they've been stripped in such a race setting modification madness that they are basically these like tricked out full speed go-karts that are within the constraints of a power wheel body and after doing that enough years and realizing how the crowd reacted and how we didn't have to reach out to anybody to get teams to show up anywhere we went.
We would just get groups of people who would just you know, Google us and find out what we did.
And they see a couple of videos and they were already on board and it kind of dawned on us like, especially when we would have high school students join without ever stepping foot in a high school. that there's something really interesting here.
We have students come into our league and They've never done any of this before.
A lot of them want to become engineers, electrical engineers, mechanical engineers, and they've never built a go-kart. and they're in high school.
So like when you step back and think of it from that perspective, all of a sudden it seems incredibly logical that they'd want to build you know, a power wheel racer because it is a low cost, low stakes way of learning something completely new and not feeling like a complete idiot when it fails because the entire league's built on screwing up.
You're making cars from garbage and that results in people building custom controllers from scratch. and, you know, converting a car alternators to, you know, three phase electric motors and, and people, uh, engineers who then show their students the proper and correct way to disassemble and build battery packs out of a discarded Nissan Leaf Cell.
Like, that's really powerful stuff. I was really impressed in Detroit for the past couple of years because, um, Just by chance, one of our teams from MIT is right next door to a kid who's 15 years old and built their car out of the frames made of wood.
They built it in their garage with two of their friends over the summer.
And they're learning from this wonderful group of makers who are people who at the top of their game.
What's incredible is how much you can do and how, like you say, the cost of sensors, the cost of jets and boards, the cost of these things is coming down to the point that we can try all these things and be pretty sophisticated about it, but not so sophisticated that we aren't having fun at the same time.
Right, and the best thing is like, that price is also super important because it's a lot easier to make a $500 mistake than a $5,000 mistake.
I mean... And even 500 is kind of a higher approachable number.
I make no mistake around that. At the same time, when you realize what you get out of that amount of money is so amazing.
And the amount of joy that I see from these teams building their car And their car is like trashed by the end of the weekend.
The tie rods are snapped off. The wheels are all wobbly.
The tires are gone. the controller is almost not entirely on fire and they're just sitting there with a grin on their face and they just trashed about $500 worth of stuff.
They're happier than some teams that, you know, won the race where they're like well we we should develop this better but we're we're excited we're really excited the one kid's like my car's destroyed but i can't wait to do this again That's the kind of making that people need.
Yeah, and that's racing too. If you still got tires lifted at the end of the race, you weren't running fast enough.
Yeah. You know, or a controller if it's not melted down by the end.
Yeah. And how much fun is that to do that for virtually nothing with the car you built from scratch? scratch that's the cheapest racing series you can have aside from just rc cars we'll talk again in a year or two years, how do you think that this whole autonomous series will have changed and what will it look and feel like?
Well, our big goal is the goal. I keep telling the new team signing up is like I want a one thousand dollar autonomous car that can make laps around 10 to 15 miles an hour.
I want them to be able to overtake each other That's the goal I believe we can do that in a few years time.
And a lot of the teams have been already showing me progress that indicates that we can definitely reach that goal.
It's just a matter of getting enough people on the problem and reaching that level of momentum.
It was the same thing even in 2010 when we were doing just the human work. piloted regular Power Racing series, I actually got a lot of emails from hackerspaces saying, I don't know if you can build this kind of cart for $500.
It seems pretty low to do that. I don't know what you can really achieve with that. um you know we had teams that said yeah we think we can and the first season or so was kind of uh It was kind of hilariously bad, honestly.
But after a few seasons, like I'd say roughly two seasons.
We started getting cars that were just on a completely different level.
And then the season after that one, it would go from two cars that are on the next level to 10 cars.
And the next season, the whole field is on this level.
And it was just such a rapid shift. the minute someone figures out an effective way to do it you'll have a couple teams copy it and then one of those teams that copied it found an even better way to do it and it just multiplies and multiplies and multiplies.
The iterations are exponential. The improvements are nearly exponential in that regard.
Well, and I expect that you're expecting that same kind of progression and that quick uptick in the autonomous series, too.
Yeah, maybe a little bit slower just because it's going to... I mean, you know, technology, you only can go as fast as that, but...
The fact that we see sensors with better scanning speeds, better like just better hardware getting cheaper and cheaper it's only a matter of time and we want to be sure that we're positioned for that because We know that you can make autonomous racing entertaining, which is something that people don't think, and they firmly don't believe it.
I think that autonomous racing opens the doors up for something that's completely different than what we've seen before.
I've been telling a lot of The teams that like eventually the goal is to basically have a racing series that looks like Speed Racer.
Well, you have your goals, racing teams out there.
If you don't have a racing team, get it together.
Jim Burke, founder of the Power Racing Series.
Thank you so much. Thank you. Thank you again.
The checkered flag to you, my friend. PostScript here.
If you are listening to this podcast on iTunes, Google Play Music. whatever your favorite podcast player is, do us a favor and leave us a review.
Thanks for listening.
