February 20, 2024 • 24 mins
This time, on the inside track, Andy Saba joins us to discuss his role in leading a mostly undergraduate student team for the IAC. Andy Saba is a master’s student of engineering at the Carnegie Mellon University Robotics Institute and has firmly rooted himself in the world of autonomous race cars. Ryan and Paul get the inside scoop into how Andy and his team went from a few lines of code to racing an autonomous car at 150+ miles per hour.
This is a two-part series.
This is a two-part series.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
I was an undergrad getting close to finishing, and my
advisor came to me said like, hey, you guys should
do this like Indy Autonomous Challenge thing. You know, it's
this really cool university challenge. And you guys, being our
robotics club, you mays have done like all these international
competitions before. And I said to him, I said no,
because this is insane. It's insane to think about, you know,
(00:22):
like a race car, like going at really high speeds
like autonomously. That's that's just insane to me. He convinced us, Like,
what's the worst that could happen? We bought the car,
you know, we like raised all this money, made this happen,
crashed into the wall at IMS at sixty nine miles
per hour, and then like, you know, we're passing Hawaii
at one hundred and fifty seven miles per hour. You
(00:42):
know that.
Speaker 2 (00:47):
The world around this is changing faster than ever before.
Ideas once only imagined in science fiction are becoming a reality.
Throughout the course of our amazing twenty three episode season,
we'll speak to some of the greatest minds in robotics
and artificial intelligence to discuss the groundbreaking work that's fueling
it all. I'm your host Ryan Marine joined me and
my co host Paul Mitchell, the president of the Indie
(01:09):
Autonomous Challenge, and see why we call this the inside Track.
This week on the show, Andy Saba tells us about
the amazing journey of a group of students who started
out in a small robotics club and ended up building
a folding autonomous race car for the Indie Autonomous Challenge.
Andy is a leader and a mentor for the Midpit
(01:29):
autonomous racing team from the University of Pittsburgh at Massachusetts
Institute of Technology. He shares the story of leading this
group of primarily undergraduate students all the way from a
few lines of code to a full racing technology stack.
We'll hear about the challenges they faced building their team
and why bringing undergrads into high level robotics is so essential.
(01:53):
This is part one of our conversation with Andy Saba. Andy,
great to see you, Wonderful to have you with us
on the program. What have you been doing since we
saw you last out at Las Vegas Motor Speedway.
Speaker 1 (02:09):
Yeah, so I've been working hard, taking classes, getting prepared
for Manza.
Speaker 2 (02:16):
Busy time for sure, and with so much excitement about
the new challenge of going left and right for the
first time with the ind the autonomous challenge, Right.
Speaker 1 (02:25):
Yeah, this is this is the first time we're going
left and right, so the first time we're going clockwise.
It's a lot of firs. So it's exciting and thankfully
the car handled it just fine.
Speaker 3 (02:38):
So andy when when you guys found out that we
were planning to transition to road courses versus just continuing
to push the oval format, which I think we could
have done right Ice could just say, Hey, we're good
at running on ovals, let's do it more, let's add
a third car, let's go faster. Fans certainly enjoy it.
(03:00):
What was your reaction from from like a team standpoint,
and like, how how does this challenge differ road courses
from ovals in terms of how you prepare and plan
for it.
Speaker 1 (03:13):
Yeah, definitely, If, for example, the i C would have
just said okay, let's do an oval with three cars,
the challenge becomes more of like, okay, detecting and handling
you know, more than one other car at a time
on a track, So that is that is a very
different challenge than saying, Okay, the car now needs to
(03:34):
be able to slow for corners, make right turns, make
left turns, and make aggressive turns. These are these are
very different challenges. So for us, UH it was it
was an exciting moment because we focused a lot on
the UH in on our development, on really getting really
(03:54):
well at at the multigent interaction with you know, two
cars at a time, and you know, there's a lot
of approaches that we take our stack that takes advantage
of the fact that we're on an oval. Right, there's
a there's a lot of simplifications. You make, a lot
of assumptions you make and with a with a road course,
you can't make those assumptions anymore. Everything everything starts to
(04:16):
to your assumptions start to fall apart and you need
you need more advanced techniques. So for us, you know,
it was definitely a oh, oh crap moment. You know,
we have we have until June to to make this happen.
But at the same time, it was an exciting challenge
to really you know, push push our existing stack even
(04:36):
further and seeing, you know, how how can we how
can we tackle something a little bit different. So we've
we've done oval with two cars, right, right, there's oval
with more than two cars, and then there's you know,
road courses with two cars, and then road courses with
more than two cars, right, And I think like the
first step of even just doing a road course is
(04:58):
a challenge on its own, and then hopefully the multi
agent part of it should be we should be a
little bit more comfortable with that now that we've we've
done you know, really high speeds on ovals with multi agent.
Speaker 2 (05:12):
What percentage of your existing code is still valid going
making that switch from what we saw on the oval
in Vegas to your first time on track at mindset,
How much did you have to create specifically for that
new challenge?
Speaker 1 (05:26):
Actually very little. That's something that like is has been
kind of a little bit of a surprise, but also
kind of not so. The thing that we were most
worried about is our controller. So the thing that the
part of the stack that says, Okay, I want to
follow this race line. Let's say I have a race
(05:47):
line that I want to follow, or if I you know,
have another car and I want to drive around them. Right,
So the part of the stack that says, okay, here's
where I'm at in the world, and here's where I
want to go, how do I steer, and how do
I control my gas? Right, So that is that is
a very core piece of the stack. It is the
thing that is ultimately controlling the car. Everything else is
(06:10):
providing information to that piece to ultimately make that final decision.
And initially, I mean we were a little concerned that okay,
you know, our existing controller might not be able to
handle to handle a road course. But with some tweaks
and some and some changes, uh, we it actually does
a reasonably okay job. It's you know, it's not perfect.
(06:33):
There's some edge cases where like it definitely could do better,
and it required a lot of like tuning to get
to get it to a good spot. We we are
we are working on our next generation stack that will
hopefully allow us to really push for the same performance
that we see on ovals on a road course. That's
(06:55):
that's like, that's what we're striving for. But we're trying
to we're trying to shoot for Okay, can we go
one hundred fifty, one hundred and sixty and seventy and
eighty on a road course? Right? So far, we've done
you know, like eighty, but we still got a long
way to.
Speaker 3 (07:09):
Go and thinking about your participation and in the autonomous challenge,
and you know, you're you're very much a student. You
you started out as a an undergraduate student that was
part of the team, you're now working on an advanced
graduate degree. Kind of what is it about the IC
(07:34):
and the participation in this particular collaborative project that is
different from your other colleagues and student experiences and sort
of what does that meant to you as you as
you develop and progress through your through your studies.
Speaker 1 (07:52):
Yeah, so, I mean for me, it's it's kind of twofold.
One is that it's just really cool having a race car.
You know, it's really cool like writing code for a
race car watching it go really fast, and these are
like full size things, right, and you know that is
that is honestly the coolest thing about the the i C. Right,
(08:13):
Like Ice could have said, Okay, let's do you know
electric race cars, or let's do like go cars, right, No,
let's like push the state of the yard with full
size human human sized race cars. And I think that
is like the thing that is just so unique about
this challenge and nothing like it exists, And so for me,
that was that was also the most exciting and terrifying
(08:36):
part because when I when I first started the project,
like like like you said, Paul, I was, I was
an undergrad getting close to finishing, and my advisor came
to me, my undergrad advisor, and said, like, hey, you
guys should do this indie autonomous challenge thing. You know,
it's a it's this really cool university challenge. You guys
(08:57):
being our robotics club. You guys have done like all
these Internet of competitions before. And I said to him,
I said no, because this is insane. It's insane to
think about, you know, like a race car like going
at really high speeds like autonomously. That's that's just insane
to me. But he, you know, he convinced us like
what's what's what's the worst that could happen? We bought
(09:20):
the car, you know, we like raised all this money,
made this happen. You know, uh, crashed into the wall
at Ims at sixty nine miles per hour and then
like you know, we were passing Hawaii at sixty nine
meters per second. You know, it's like one hundred and
fifty seven miles per hour. You know that's that's like,
that's like where we've come and the journey, and and
(09:42):
for me, like the the thing that you know, the
initial thing is like, hey, this is really cool, is
really exciting. But the thing that kind of has like
kept it kept it going for me is that a
race car is like the most exciting thing that you
can is is this really cool thing that you can
get students it's excited about. And for our team in particular,
(10:03):
a lot of our members are undergrad It's the majority
of our members are undergrads coming in with no experience,
and they go to University Pittsburgh, which is I mean,
it's a good public university, but it's not known for robotics,
and a lot of these a lot of these students
they didn't know they wanted to do robotics. They didn't
know that they liked you know, AI and machine learning
and controls and all these these interesting interdisciplinary fields. And
(10:28):
they came on our team and they got really excited
because it's such a cool problem, and then it was
really easy for them to just kind of get sucked in.
Speaker 2 (10:37):
And Paul, we've talked about this before. The race car
is a vehicle, and that's true in the most literal sense,
but it's also a vehicle to promote ingenuity. It's a
vehicle to attract the kind of interest that Andy is
talking about. And if it hadn't been something so visceral,
it's something that had real stakes attached to it, would
(10:57):
the level of interest have been there for you and
for the rest of the folks on your team. It
sounds like perhaps that wouldn't have been the case.
Speaker 1 (11:04):
No, I definitely don't think so. I definitely don't think so.
I mean, I've seen it time and time again where
you know, people join our team and you know they
they have like some level of interest, but like classes
are more important, and you know they don't they don't
pull the all nighters working on the car. And then
one of the unique things about our team met pit
RW is that we we allow everyone. If you want
(11:26):
to come field testing, if you want to go out
to Indiana, you know, go to Indianapolis and like test
the car and Lucas Oil I mean, and gets stung
by a thousand bees, You're you're more than welcome to.
But the second that people see the car, it's like
night and day in terms of like in terms of
their involvement and their motivation level, you know, you'll have
students who you know have been just around, you know,
(11:47):
coming to meetings. They'll like, you know, have like a
few lines of code here and there. The second they
see the car and they see how real it is
of you know, their code running on on this this
amazing platform, you can't you can't get them to stop
doing the all nighters.
Speaker 3 (12:05):
So one thing that we've been talking about through this
podcast series is the role that AI plays in the
in the Autonomous Challenge, but even more broadly in mobility
and even more broadly than that in society. I just
heard you talking about kind of the pride that you
(12:27):
and the other team members take in your code coming
to life in the form of these race cars and
the AI drivers that are behind the proverbial wheel. Can
you talk a little bit about kind of the literally
the personality of that code and the way in which
your AI driver for midpit r W is different from
(12:52):
say the AI driver of the team from Technical University
of Munich or or you guys were up against the
AI tech racing team of of why and Berkeley and
some other universities. Kind Of what is the actual personalities
Because I think people assume that everyone is trying to
(13:13):
write essentially the same code to solve for the same
exact problem, and therefore perfection would be all the cars
operating exactly the same, passing each other the same way,
and that clearly is not the case. Just like motorsports,
you have different personalities and different risk tolerances of the drivers.
How does that how does that come out in in
ural's code? From from your perspective.
Speaker 1 (13:36):
It's interesting. I've never really thought about the personality of
So we named our car Betty, So I'm just gonna
refer to her as Betty, but after after after the
late Betty White, and definitely I mean compared to the
first year, she's a little bit more conservative, you know, like, yes,
we you know, we were always willing to like push
(13:58):
to really high speeds and uh and and whatnot. But Betty,
you know, even though she has a really good understanding
of the of the cars around her. I will say,
probably our biggest like our biggest strength is is our
ability to to see the other cars. I think that
is one of our like secret sauces that that works
(14:19):
really well. Betty tends to be a little more a
little more conservative, give a little more space where needed.
If you notice, like during during the event, you know,
we always we always take a very very inside line
just just to give the other gether agents more space
because ultimately, I mean, in this competition, survival tends to
(14:39):
tends to be the deciding factor, right, and in our
case it was the amount of fuel. But you know,
it's a matter of you know, just just making it
to the higher speeds. She she tends to, she tends to,
you know, sit back and watch and uh and and
and make her moves, make her moves when when she
feels that it's it's right. But hopefully, I mean, we're
(15:02):
working on giving her some some some new some new
spice that will hopefully allow her to be a little
bit more aggressive.
Speaker 2 (15:12):
Well to that end, is Betty's personality on a road
course different than it is on a noval or is
it too soon to say?
Speaker 1 (15:19):
It might be a little bit too soon to say,
but she definitely, she definitely took to Monza pretty well.
That was actually one of the surprising things.
Speaker 3 (15:26):
Yeah, when you think about the role that that AI
plays in your studies and the kind of discussions that
I'm sure have to be happening on campus with you know,
chat GPT and all these other kind of learning based
AI tools coming about, and yet here you are applying AI,
(15:49):
machine learning, kind of advanced robotics to a use case
of a race car and mobility kind of How much
is the disc ession that you guys have about the
way you're contributing, what you're contributing and what it means
to the future of mobility or do you even have
time to think about that? Do you just simply think
about the next competition, the next run just kind of
(16:14):
trying to understand, you know, the mindset of a team member.
Do you have time to to step back and think
about the bigger picture and how you're contributing what that
will mean, you know, throughout your career ten twenty years
from now.
Speaker 1 (16:28):
Yeah, and we'll say we're we're kind of at the
stage with a lot of the technology where the problems
aren't you know, do you have the best language model?
Speaker 2 (16:41):
Right?
Speaker 1 (16:41):
Are you using GPT four or GPT three? Right? Our
problems are more more tangible than that, right. Our problem
is is that, Okay, does the trees get in the
way and you lose GPS? You know, and like like
how do you how do you make a robuff system
around that. So one of the I guess, like this
is a more philosophical answer. One of the things that
(17:01):
I really had to learn a lot as a as
a researchers, as a when I started my masters was
understanding like where where's the right places to like focus
your attention because there is so much right now right
like right now, there's just an explosion of advance fits,
and it's really easy to just kind of like your
head to just spin like, Okay, you wait, should we
(17:24):
be using this? Should we doing that? Should we doing that?
And that that definitely was like early on in our team,
we definitely had a little bit of that, that head
spinning and you know, the things that we were focusing on.
But what's what's really good about a competition and what's
like really grounding about it is that there is a deadline,
(17:45):
there is a set requirement that you have to do,
and that is a really good motivation for really grounding
the approach and really focusing on what are what are
the problems that we're actually trying to solve. So in
you know, there's a lot of fantastic tools that are
being developed right now right and I think of chat
(18:05):
TVT as a tool I think of, you know, you
know these generitive models like stable diffusion and and all
these you know, all these really great advancements, they're just tools, right.
We haven't quite found the problems though, for mobility, for
autonomous racing that we can apply those tools to. And
(18:26):
I think the way you advance the field is you
can't think of you can't think of things in terms of, Okay,
how do I use this tool? It's what are the problems?
And is this tool the right tool to solve that problem?
Speaker 2 (18:40):
And if not, presumably then how do you craft that
tool right to continue to drive this forward?
Speaker 1 (18:46):
Yeah? That is a very like real question, right, And
that that is that is fundamentally what what advanced research
is all about. It's all about like taking your problem,
simplifying it down to the essence of like what are
the problems? Right? So for Manza, the the problem statement
that we came to is how can you go into
(19:09):
a corner at very high speeds with another agent in
front of you and come out of that corner either
in front of the other agent or in a position
to successfully complete an over an overtake. That is that
is like, that is the problem, right, That we're we're
we're trying to address because it's it's there's multiple levels
(19:31):
to it. Right, you have you have to be able
to handle corners. You have to be able to right corners,
left corners, tight corners, you have to be able to
handle Okay, how is this What is this agent going
to do? Right on an oval? It's a lot simpler.
Right on an oval, you say, okay, they're probably going
to follow the race line right for a corner. There's
a lot more decisions about what they can do. And
(19:52):
so there's a ton of like sub problems to this
like overarching problem. But for us, it's it's really grounding.
It's it's really useful to have a grounding problem to
focus on rather than just focus on all of the
all of the Really, at this point, it's kind of noise, right.
It's there's going to be always new things, but until
(20:15):
you have a problem to apply it to, it's it's
it's not useful to you.
Speaker 3 (20:20):
It's just a shiny new object that may be a
distraction more than it is a part of your solution.
When when you think about the pioneering work that you
and the other teams have been doing in this field,
of high speed automation right in excess of one hundred
(20:41):
miles an hour multi agent, and you think back to probably,
I guess a couple of generations before you. Now we're
talking to Sebastian Thrun who who won the darp and
Grand Challenge, and there was about low speed kind of
level five navigation in a desert where the trains are
(21:04):
completely different, and I think the winning team was averaging
speeds of nineteen miles an hour. And from that we
saw a lot of work in this kind of urban
suburban low speed robo taxi as the focus of a
lot of the institutions, including you know where you are
now at Carnegie Mellon's done a lot of work in that.
Do you feel that you guys are working on a
(21:27):
new field. Do you feel like a pioneer where instead
of you know, further perfecting the low speed urban suburban
robotaxi use case, you're tackling this use case of extreme
high speed automation. And where do you think that goes
in terms of the connection back to inevitably, back to mobility,
(21:50):
back to passenger safety on highways.
Speaker 1 (21:53):
Yeah, I don't know. I never really thought of ourselves
as pioneers, I think because so much of what we
rely on on, you know, is is just learning from
from what has come before us. If we're pioneers, that's
just because you know, we're like standing on the shoulders
of the true pioneers. But I will say I think,
(22:15):
I think like kind of where where you know, the
high speed autonomy kind of fits in is I think
it's a really good vessel to develop solutions to problems
that are maybe not relevant in the ninety percent case
for passengers, but more in that ten percent. Right, So
in the context of self driving vehicles, there's always this
(22:39):
this talk about the last ten percent, Right, how do
you get a system to work ninety nine point nine
to infinity percent of the time? And part of that
is you need to be able to handle you know,
very fast moving objects. You need to be able to
handle snow. You need to be able to handle ice. Right,
(23:00):
you know, a race car driving at the limit, right,
you know, at one hundred and eighty one hundred and
ninety two hundred miles per hour, that is not much
different than a passenger vehicle driving on black ice. The
way that the handling of the tires, so the things
that we are developing, I think are very applicable to
(23:20):
that last like ten percent. I don't necessarily think of
it as a as a totally separate thing because so
much of it is is the same, right, we deal
with a lot of the same problems, maybe like actually
we deal with the same problems, but kind of too extreme.
You know, a lot of these self driving car companies
they develop their own custom hardware, They develop their own cameras,
(23:42):
their own light ours, so they're able to have you know,
dedicated compute. You know, they don't have to rely on
you know, commercial off the off the shelf parts put
together into a platform. They they also have you know,
significant more engineers resources, and they also have a little
(24:02):
bit of a larger time budget than we do. Right,
you know, we have to make decisions. We measure late
and see in the order of you know, microseconds. They
you know, can maybe get away with the millisecond, right
at least they they will say that they're targeting the
microsecond because obviously, you know, faster reaction time is always better.
(24:25):
But for us, you know, it's it's not a matter
of like a safety case, it's a matter of doesn't
work at all.
Speaker 2 (24:36):
That was part one of our conversation with team leader
and mentor Andy Saba. Andy's passion for motorsport and robotics
is something everyone can aspire to. Join us next time
for the rest of this incredible conversation, where we talk
about Andy's philosophy of teaching students to explore the thought
process behind decisions and not focus solely on a deadline
I was an undergrad getting close to finishing, and my
advisor came to me said like, hey, you guys should
do this like Indy Autonomous Challenge thing. You know, it's
this really cool university challenge. And you guys, being our
robotics club, you mays have done like all these international
competitions before. And I said to him, I said no,
because this is insane. It's insane to think about, you know,
(00:22):
like a race car, like going at really high speeds
like autonomously. That's that's just insane to me. He convinced us, Like,
what's the worst that could happen? We bought the car,
you know, we like raised all this money, made this happen,
crashed into the wall at IMS at sixty nine miles
per hour, and then like, you know, we're passing Hawaii
at one hundred and fifty seven miles per hour. You
(00:42):
know that.
Speaker 2 (00:47):
The world around this is changing faster than ever before.
Ideas once only imagined in science fiction are becoming a reality.
Throughout the course of our amazing twenty three episode season,
we'll speak to some of the greatest minds in robotics
and artificial intelligence to discuss the groundbreaking work that's fueling
it all. I'm your host Ryan Marine joined me and
my co host Paul Mitchell, the president of the Indie
(01:09):
Autonomous Challenge, and see why we call this the inside Track.
This week on the show, Andy Saba tells us about
the amazing journey of a group of students who started
out in a small robotics club and ended up building
a folding autonomous race car for the Indie Autonomous Challenge.
Andy is a leader and a mentor for the Midpit
(01:29):
autonomous racing team from the University of Pittsburgh at Massachusetts
Institute of Technology. He shares the story of leading this
group of primarily undergraduate students all the way from a
few lines of code to a full racing technology stack.
We'll hear about the challenges they faced building their team
and why bringing undergrads into high level robotics is so essential.
(01:53):
This is part one of our conversation with Andy Saba. Andy,
great to see you, Wonderful to have you with us
on the program. What have you been doing since we
saw you last out at Las Vegas Motor Speedway.
Speaker 1 (02:09):
Yeah, so I've been working hard, taking classes, getting prepared
for Manza.
Speaker 2 (02:16):
Busy time for sure, and with so much excitement about
the new challenge of going left and right for the
first time with the ind the autonomous challenge, Right.
Speaker 1 (02:25):
Yeah, this is this is the first time we're going
left and right, so the first time we're going clockwise.
It's a lot of firs. So it's exciting and thankfully
the car handled it just fine.
Speaker 3 (02:38):
So andy when when you guys found out that we
were planning to transition to road courses versus just continuing
to push the oval format, which I think we could
have done right Ice could just say, Hey, we're good
at running on ovals, let's do it more, let's add
a third car, let's go faster. Fans certainly enjoy it.
(03:00):
What was your reaction from from like a team standpoint,
and like, how how does this challenge differ road courses
from ovals in terms of how you prepare and plan
for it.
Speaker 1 (03:13):
Yeah, definitely, If, for example, the i C would have
just said okay, let's do an oval with three cars,
the challenge becomes more of like, okay, detecting and handling
you know, more than one other car at a time
on a track, So that is that is a very
different challenge than saying, Okay, the car now needs to
(03:34):
be able to slow for corners, make right turns, make
left turns, and make aggressive turns. These are these are
very different challenges. So for us, UH it was it
was an exciting moment because we focused a lot on
the UH in on our development, on really getting really
(03:54):
well at at the multigent interaction with you know, two
cars at a time, and you know, there's a lot
of approaches that we take our stack that takes advantage
of the fact that we're on an oval. Right, there's
a there's a lot of simplifications. You make, a lot
of assumptions you make and with a with a road course,
you can't make those assumptions anymore. Everything everything starts to
(04:16):
to your assumptions start to fall apart and you need
you need more advanced techniques. So for us, you know,
it was definitely a oh, oh crap moment. You know,
we have we have until June to to make this happen.
But at the same time, it was an exciting challenge
to really you know, push push our existing stack even
(04:36):
further and seeing, you know, how how can we how
can we tackle something a little bit different. So we've
we've done oval with two cars, right, right, there's oval
with more than two cars, and then there's you know,
road courses with two cars, and then road courses with
more than two cars, right, And I think like the
first step of even just doing a road course is
(04:58):
a challenge on its own, and then hopefully the multi
agent part of it should be we should be a
little bit more comfortable with that now that we've we've
done you know, really high speeds on ovals with multi agent.
Speaker 2 (05:12):
What percentage of your existing code is still valid going
making that switch from what we saw on the oval
in Vegas to your first time on track at mindset,
How much did you have to create specifically for that
new challenge?
Speaker 1 (05:26):
Actually very little. That's something that like is has been
kind of a little bit of a surprise, but also
kind of not so. The thing that we were most
worried about is our controller. So the thing that the
part of the stack that says, Okay, I want to
follow this race line. Let's say I have a race
(05:47):
line that I want to follow, or if I you know,
have another car and I want to drive around them. Right,
So the part of the stack that says, okay, here's
where I'm at in the world, and here's where I
want to go, how do I steer, and how do
I control my gas? Right, So that is that is
a very core piece of the stack. It is the
thing that is ultimately controlling the car. Everything else is
(06:10):
providing information to that piece to ultimately make that final decision.
And initially, I mean we were a little concerned that okay,
you know, our existing controller might not be able to
handle to handle a road course. But with some tweaks
and some and some changes, uh, we it actually does
a reasonably okay job. It's you know, it's not perfect.
(06:33):
There's some edge cases where like it definitely could do better,
and it required a lot of like tuning to get
to get it to a good spot. We we are
we are working on our next generation stack that will
hopefully allow us to really push for the same performance
that we see on ovals on a road course. That's
(06:55):
that's like, that's what we're striving for. But we're trying
to we're trying to shoot for Okay, can we go
one hundred fifty, one hundred and sixty and seventy and
eighty on a road course? Right? So far, we've done
you know, like eighty, but we still got a long
way to.
Speaker 3 (07:09):
Go and thinking about your participation and in the autonomous challenge,
and you know, you're you're very much a student. You
you started out as a an undergraduate student that was
part of the team, you're now working on an advanced
graduate degree. Kind of what is it about the IC
(07:34):
and the participation in this particular collaborative project that is
different from your other colleagues and student experiences and sort
of what does that meant to you as you as
you develop and progress through your through your studies.
Speaker 1 (07:52):
Yeah, so, I mean for me, it's it's kind of twofold.
One is that it's just really cool having a race car.
You know, it's really cool like writing code for a
race car watching it go really fast, and these are
like full size things, right, and you know that is
that is honestly the coolest thing about the the i C. Right,
(08:13):
Like Ice could have said, Okay, let's do you know
electric race cars, or let's do like go cars, right, No,
let's like push the state of the yard with full
size human human sized race cars. And I think that
is like the thing that is just so unique about
this challenge and nothing like it exists, And so for me,
that was that was also the most exciting and terrifying
(08:36):
part because when I when I first started the project,
like like like you said, Paul, I was, I was
an undergrad getting close to finishing, and my advisor came
to me, my undergrad advisor, and said, like, hey, you
guys should do this indie autonomous challenge thing. You know,
it's a it's this really cool university challenge. You guys
(08:57):
being our robotics club. You guys have done like all
these Internet of competitions before. And I said to him,
I said no, because this is insane. It's insane to
think about, you know, like a race car like going
at really high speeds like autonomously. That's that's just insane
to me. But he, you know, he convinced us like
what's what's what's the worst that could happen? We bought
(09:20):
the car, you know, we like raised all this money,
made this happen. You know, uh, crashed into the wall
at Ims at sixty nine miles per hour and then
like you know, we were passing Hawaii at sixty nine
meters per second. You know, it's like one hundred and
fifty seven miles per hour. You know that's that's like,
that's like where we've come and the journey, and and
(09:42):
for me, like the the thing that you know, the
initial thing is like, hey, this is really cool, is
really exciting. But the thing that kind of has like
kept it kept it going for me is that a
race car is like the most exciting thing that you
can is is this really cool thing that you can
get students it's excited about. And for our team in particular,
(10:03):
a lot of our members are undergrad It's the majority
of our members are undergrads coming in with no experience,
and they go to University Pittsburgh, which is I mean,
it's a good public university, but it's not known for robotics,
and a lot of these a lot of these students
they didn't know they wanted to do robotics. They didn't
know that they liked you know, AI and machine learning
and controls and all these these interesting interdisciplinary fields. And
(10:28):
they came on our team and they got really excited
because it's such a cool problem, and then it was
really easy for them to just kind of get sucked in.
Speaker 2 (10:37):
And Paul, we've talked about this before. The race car
is a vehicle, and that's true in the most literal sense,
but it's also a vehicle to promote ingenuity. It's a
vehicle to attract the kind of interest that Andy is
talking about. And if it hadn't been something so visceral,
it's something that had real stakes attached to it, would
(10:57):
the level of interest have been there for you and
for the rest of the folks on your team. It
sounds like perhaps that wouldn't have been the case.
Speaker 1 (11:04):
No, I definitely don't think so. I definitely don't think so.
I mean, I've seen it time and time again where
you know, people join our team and you know they
they have like some level of interest, but like classes
are more important, and you know they don't they don't
pull the all nighters working on the car. And then
one of the unique things about our team met pit
RW is that we we allow everyone. If you want
(11:26):
to come field testing, if you want to go out
to Indiana, you know, go to Indianapolis and like test
the car and Lucas Oil I mean, and gets stung
by a thousand bees, You're you're more than welcome to.
But the second that people see the car, it's like
night and day in terms of like in terms of
their involvement and their motivation level, you know, you'll have
students who you know have been just around, you know,
(11:47):
coming to meetings. They'll like, you know, have like a
few lines of code here and there. The second they
see the car and they see how real it is
of you know, their code running on on this this
amazing platform, you can't you can't get them to stop
doing the all nighters.
Speaker 3 (12:05):
So one thing that we've been talking about through this
podcast series is the role that AI plays in the
in the Autonomous Challenge, but even more broadly in mobility
and even more broadly than that in society. I just
heard you talking about kind of the pride that you
(12:27):
and the other team members take in your code coming
to life in the form of these race cars and
the AI drivers that are behind the proverbial wheel. Can
you talk a little bit about kind of the literally
the personality of that code and the way in which
your AI driver for midpit r W is different from
(12:52):
say the AI driver of the team from Technical University
of Munich or or you guys were up against the
AI tech racing team of of why and Berkeley and
some other universities. Kind Of what is the actual personalities
Because I think people assume that everyone is trying to
(13:13):
write essentially the same code to solve for the same
exact problem, and therefore perfection would be all the cars
operating exactly the same, passing each other the same way,
and that clearly is not the case. Just like motorsports,
you have different personalities and different risk tolerances of the drivers.
How does that how does that come out in in
ural's code? From from your perspective.
Speaker 1 (13:36):
It's interesting. I've never really thought about the personality of
So we named our car Betty, So I'm just gonna
refer to her as Betty, but after after after the
late Betty White, and definitely I mean compared to the
first year, she's a little bit more conservative, you know, like, yes,
we you know, we were always willing to like push
(13:58):
to really high speeds and uh and and whatnot. But Betty,
you know, even though she has a really good understanding
of the of the cars around her. I will say,
probably our biggest like our biggest strength is is our
ability to to see the other cars. I think that
is one of our like secret sauces that that works
(14:19):
really well. Betty tends to be a little more a
little more conservative, give a little more space where needed.
If you notice, like during during the event, you know,
we always we always take a very very inside line
just just to give the other gether agents more space
because ultimately, I mean, in this competition, survival tends to
(14:39):
tends to be the deciding factor, right, and in our
case it was the amount of fuel. But you know,
it's a matter of you know, just just making it
to the higher speeds. She she tends to, she tends to,
you know, sit back and watch and uh and and
and make her moves, make her moves when when she
feels that it's it's right. But hopefully, I mean, we're
(15:02):
working on giving her some some some new some new
spice that will hopefully allow her to be a little
bit more aggressive.
Speaker 2 (15:12):
Well to that end, is Betty's personality on a road
course different than it is on a noval or is
it too soon to say?
Speaker 1 (15:19):
It might be a little bit too soon to say,
but she definitely, she definitely took to Monza pretty well.
That was actually one of the surprising things.
Speaker 3 (15:26):
Yeah, when you think about the role that that AI
plays in your studies and the kind of discussions that
I'm sure have to be happening on campus with you know,
chat GPT and all these other kind of learning based
AI tools coming about, and yet here you are applying AI,
(15:49):
machine learning, kind of advanced robotics to a use case
of a race car and mobility kind of How much
is the disc ession that you guys have about the
way you're contributing, what you're contributing and what it means
to the future of mobility or do you even have
time to think about that? Do you just simply think
about the next competition, the next run just kind of
(16:14):
trying to understand, you know, the mindset of a team member.
Do you have time to to step back and think
about the bigger picture and how you're contributing what that
will mean, you know, throughout your career ten twenty years
from now.
Speaker 1 (16:28):
Yeah, and we'll say we're we're kind of at the
stage with a lot of the technology where the problems
aren't you know, do you have the best language model?
Speaker 2 (16:41):
Right?
Speaker 1 (16:41):
Are you using GPT four or GPT three? Right? Our
problems are more more tangible than that, right. Our problem
is is that, Okay, does the trees get in the
way and you lose GPS? You know, and like like
how do you how do you make a robuff system
around that. So one of the I guess, like this
is a more philosophical answer. One of the things that
(17:01):
I really had to learn a lot as a as
a researchers, as a when I started my masters was
understanding like where where's the right places to like focus
your attention because there is so much right now right
like right now, there's just an explosion of advance fits,
and it's really easy to just kind of like your
head to just spin like, Okay, you wait, should we
(17:24):
be using this? Should we doing that? Should we doing that?
And that that definitely was like early on in our team,
we definitely had a little bit of that, that head
spinning and you know, the things that we were focusing on.
But what's what's really good about a competition and what's
like really grounding about it is that there is a deadline,
(17:45):
there is a set requirement that you have to do,
and that is a really good motivation for really grounding
the approach and really focusing on what are what are
the problems that we're actually trying to solve. So in
you know, there's a lot of fantastic tools that are
being developed right now right and I think of chat
(18:05):
TVT as a tool I think of, you know, you
know these generitive models like stable diffusion and and all
these you know, all these really great advancements, they're just tools, right.
We haven't quite found the problems though, for mobility, for
autonomous racing that we can apply those tools to. And
(18:26):
I think the way you advance the field is you
can't think of you can't think of things in terms of, Okay,
how do I use this tool? It's what are the problems?
And is this tool the right tool to solve that problem?
Speaker 2 (18:40):
And if not, presumably then how do you craft that
tool right to continue to drive this forward?
Speaker 1 (18:46):
Yeah? That is a very like real question, right, And
that that is that is fundamentally what what advanced research
is all about. It's all about like taking your problem,
simplifying it down to the essence of like what are
the problems? Right? So for Manza, the the problem statement
that we came to is how can you go into
(19:09):
a corner at very high speeds with another agent in
front of you and come out of that corner either
in front of the other agent or in a position
to successfully complete an over an overtake. That is that
is like, that is the problem, right, That we're we're
we're trying to address because it's it's there's multiple levels
(19:31):
to it. Right, you have you have to be able
to handle corners. You have to be able to right corners,
left corners, tight corners, you have to be able to
handle Okay, how is this What is this agent going
to do? Right on an oval? It's a lot simpler.
Right on an oval, you say, okay, they're probably going
to follow the race line right for a corner. There's
a lot more decisions about what they can do. And
(19:52):
so there's a ton of like sub problems to this
like overarching problem. But for us, it's it's really grounding.
It's it's really useful to have a grounding problem to
focus on rather than just focus on all of the
all of the Really, at this point, it's kind of noise, right.
It's there's going to be always new things, but until
(20:15):
you have a problem to apply it to, it's it's
it's not useful to you.
Speaker 3 (20:20):
It's just a shiny new object that may be a
distraction more than it is a part of your solution.
When when you think about the pioneering work that you
and the other teams have been doing in this field,
of high speed automation right in excess of one hundred
(20:41):
miles an hour multi agent, and you think back to probably,
I guess a couple of generations before you. Now we're
talking to Sebastian Thrun who who won the darp and
Grand Challenge, and there was about low speed kind of
level five navigation in a desert where the trains are
(21:04):
completely different, and I think the winning team was averaging
speeds of nineteen miles an hour. And from that we
saw a lot of work in this kind of urban
suburban low speed robo taxi as the focus of a
lot of the institutions, including you know where you are
now at Carnegie Mellon's done a lot of work in that.
Do you feel that you guys are working on a
(21:27):
new field. Do you feel like a pioneer where instead
of you know, further perfecting the low speed urban suburban
robotaxi use case, you're tackling this use case of extreme
high speed automation. And where do you think that goes
in terms of the connection back to inevitably, back to mobility,
(21:50):
back to passenger safety on highways.
Speaker 1 (21:53):
Yeah, I don't know. I never really thought of ourselves
as pioneers, I think because so much of what we
rely on on, you know, is is just learning from
from what has come before us. If we're pioneers, that's
just because you know, we're like standing on the shoulders
of the true pioneers. But I will say I think,
(22:15):
I think like kind of where where you know, the
high speed autonomy kind of fits in is I think
it's a really good vessel to develop solutions to problems
that are maybe not relevant in the ninety percent case
for passengers, but more in that ten percent. Right, So
in the context of self driving vehicles, there's always this
(22:39):
this talk about the last ten percent, Right, how do
you get a system to work ninety nine point nine
to infinity percent of the time? And part of that
is you need to be able to handle you know,
very fast moving objects. You need to be able to
handle snow. You need to be able to handle ice. Right,
(23:00):
you know, a race car driving at the limit, right,
you know, at one hundred and eighty one hundred and
ninety two hundred miles per hour, that is not much
different than a passenger vehicle driving on black ice. The
way that the handling of the tires, so the things
that we are developing, I think are very applicable to
(23:20):
that last like ten percent. I don't necessarily think of
it as a as a totally separate thing because so
much of it is is the same, right, we deal
with a lot of the same problems, maybe like actually
we deal with the same problems, but kind of too extreme.
You know, a lot of these self driving car companies
they develop their own custom hardware, They develop their own cameras,
(23:42):
their own light ours, so they're able to have you know,
dedicated compute. You know, they don't have to rely on
you know, commercial off the off the shelf parts put
together into a platform. They they also have you know,
significant more engineers resources, and they also have a little
(24:02):
bit of a larger time budget than we do. Right,
you know, we have to make decisions. We measure late
and see in the order of you know, microseconds. They
you know, can maybe get away with the millisecond, right
at least they they will say that they're targeting the
microsecond because obviously, you know, faster reaction time is always better.
(24:25):
But for us, you know, it's it's not a matter
of like a safety case, it's a matter of doesn't
work at all.
Speaker 2 (24:36):
That was part one of our conversation with team leader
and mentor Andy Saba. Andy's passion for motorsport and robotics
is something everyone can aspire to. Join us next time
for the rest of this incredible conversation, where we talk
about Andy's philosophy of teaching students to explore the thought
process behind decisions and not focus solely on a deadline
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
My Favorite Murder with Karen Kilgariff and Georgia Hardstark
My Favorite Murder is a true crime comedy podcast hosted by Karen Kilgariff and Georgia Hardstark. Each week, Karen and Georgia share compelling true crimes and hometown stories from friends and listeners. Since MFM launched in January of 2016, Karen and Georgia have shared their lifelong interest in true crime and have covered stories of infamous serial killers like the Night Stalker, mysterious cold cases, captivating cults, incredible survivor stories and important events from history like the Tulsa race massacre of 1921. My Favorite Murder is part of the Exactly Right podcast network that provides a platform for bold, creative voices to bring to life provocative, entertaining and relatable stories for audiences everywhere. The Exactly Right roster of podcasts covers a variety of topics including historic true crime, comedic interviews and news, science, pop culture and more. Podcasts on the network include Buried Bones with Kate Winkler Dawson and Paul Holes, That's Messed Up: An SVU Podcast, This Podcast Will Kill You, Bananas and more.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.