March 26, 2024 • 28 mins
Today, on the inside track, we get a chance to look at how the autonomous motorsports team from the Technical University of Munich (TUM) has been writing a manual for the world of autonomous racing. To help explain the process, Simon Hoffmann, the leader of the TUM team, sits down with us to discuss how they became the first-ever winners of the Indy Autonomous challenge, some of the obstacles they overcame to get their car on the track, and how the team is able to have real-time conversations about the car without driver feedback.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
And that was basically the process, so there was always
a human involved currently checking the data, seeing what where
we have problems, adjusting parameters in our software, tuning the
software going out again and trying to increase speeds, accelerations
and see how the car behaves.
Speaker 2 (00:23):
The world around us 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 Indian
(00:45):
Autonomous Challenge, and see why we call this the inside track.
As many racing engineers were to test, there is no
playbook or owner's manual when you're racing on the cutting
edge of technology. Much like some of the first form
the one racing teams tearing through the streets of Monaco
in the nineteen fifties, today's engineers are building out their
(01:06):
racing stacks from the ground up. This includes the code,
the computers, and even the car's ability to feel what's
happening out on the track. Today, on the inside Track,
we get a chance to look at how the Autonomous
Motorsports team from the Technical University of Munich has been
writing the manual for the World of Autonomous Racing. To
help explain the process, Simon Hoffman, the leader of the
(01:29):
Tomb team, sits down with us to discuss how they
became the first ever winners of the Indie Autonomous Challenge,
some of the obstacles that they overcame to get their
car onto the track, and how the team is able
to have real time conversations about their autonomous car without
any driver feedback. It's great to have you on the
(01:58):
inside track. I'm just really interested in learning a little
bit first about you and how you got into this
interesting competition. So, at what point in your life was
robotics or AI something that became of interest, and then
how soon was that something that you were able to
channel into the Indie Autonomous Challenge.
Speaker 1 (02:17):
So I think in particular, this started pretty pretty early,
back back in school, when I was interested in how
technology works, how different robots and robotic applications, car advanced
driver assistant systems or functioning. I think that's when when
my passion for the topics raised, and of course I
(02:40):
looked into different studies which I could could pursue to
develop more or to to gain more experience in this field.
But I think back then my interest started.
Speaker 2 (02:54):
Did you always have an interest in automotive or is
that something that you stumbled into a little bit later.
Speaker 1 (03:03):
Well, I would say that the basic interest was always there,
but my main interest was across all of robotic applications.
This came more in university when I got into different
lessons touching these applications. I started back then started studying
(03:24):
mechanical engineering at first, or mecatronics, which is more like
a combination of mechanical electronics or informatics. So I wanted
to get the whole picture around robots and the way
they work. And in this area, we have a pretty
(03:46):
pretty good automotive industry here here in Munich, and also
the university has a lot of lectures in this area.
I got got interested more into the application of robotics
in tomortive industria.
Speaker 2 (04:01):
So let's dive in then too. You're getting involved with
the autonomous challenge. At what point during your academic career
there at the Technical University of Munich. Did you become
aware of this and how quickly did you want to
sign up? When you started to learn about what the
IAC was all about.
Speaker 1 (04:19):
So I did my master's thesis here at our share
where I'm currently currently working. Back then it was more
about control system of systems of electric vehicles, so actually
a totally totally different topic. But in after after I
finished my master's thesis, I got the well, the opportunity
(04:41):
to start as a research associate PhD candidate at our
chair working at tele opraded driving, which means basically remotely
controlling cars, so real road vehicles remotely controlled, which actually
comes into place as soon as automated systems don't work
anymore or get to the their functional limits, which is
(05:06):
more more or less or use case for autonomous driving
on public road because if you imagine there is no
driver inside the Weekly anymore, and the Weekly has to
navigate through cities on its own, it might still in
the beginning or even in the future, get to functional limits,
and of course it would be would be a shame
of just standing around there blocking graphic delaying each its
(05:30):
passenger's journey. And then that's where Intel operation comes into
place and that's also the main topic of my PhD
I pursued. And then what about Actually about one year
ago I finished my work on this and I got
the opportunity to join the autonomous Racing team at our chair,
(05:53):
which I was pretty excited about because for me it
actually was was the next the next step working on
the fallback systems and then also working on the real
autonomous week And yeah, it was pretty excited about new challenge,
new challenges and new things to work on. And yes,
(06:13):
in the beginning of this year I started being being
team lead of our autonomous racing team. And yeah, pretty
pretty happy right now with our team and things are
going on, I can imagine.
Speaker 2 (06:26):
So, I mean, the team actually has been a front
running team from the very beginning, going back to the
very first event at the Indianapolis Motor Speedway. And so
it sounds like you joined as the team already had
established some level of success. So what was it like
jumping in kind of mid stream on this project that
was already fairly well developed but still needed to further
(06:50):
advance as the complexity of the challenge became that much greater.
Speaker 1 (06:55):
Well, it feels kind of kind of more more relexing
if you if you're starting, and you also already have
a basis which which works. But still the challenge or
the challenges increased from race to race, which got more
more complicated. And there's still a long way way to
go to get to an autonomous racing or to to
(07:17):
a raising format like real real drivers would race, or
real race card drivers like we see every every week
in formally one for example. And there's still still a
lot to do. But yeah, I was pretty glad about
what what I already could could find in our software stack.
It's very solid software. We with this software we managed
(07:40):
to win the first Indio Autonomous challenge and so really
really good, good basis to start with. And but yeah,
the next challenges are coming.
Speaker 2 (07:51):
And it's it's got to be some pressure too, right
to know that you're jumping into an organization that's already
had success and now they're looking to you to keep
that success going.
Speaker 1 (08:01):
Yeah, of course, if you're if you're joining a winning team,
you only can can lose. Basically as well, if you win,
everyone says yeah, sure, of course the last time, Well
of course, yeah, and then if you lose, it's on you.
But I think this is not how we work here
at the new University. For us, we are always looking
(08:24):
into what research question we can we can answer with uh,
with these challenges, and the challenge is just a fun
side side part of answering those research research questions.
Speaker 2 (08:39):
Yeah, let's let's dive into that a little bit, because yes,
it is an objectively an interesting and cool project to
be a part of. But at a certain point you
have to justify the continued investment by the university in
a project like this. So what are the questions that
you continue to ask, the answers that you continue to
clean from being involved in this project and presumably continue
(09:02):
to looking into the future as well.
Speaker 1 (09:05):
I think first, first of all, we maybe need to
do to make a start a step back. So if
we're looking into the automotive industry, we already see some
cars driving autonomously on on on roads. We are progressing there,
probably not as fast as we initially expected. And for us,
the most important point or let's let's say in other
(09:29):
restaurants to so if we want to see autonomous weekles
driving on car or on public roads really reliable, we
need to get up to a reliability of let's say
ninety nine point ninety ninety nine percent and maybe maybe
even more and realizing probably ninety five percent, I think
(09:51):
we're we're pretty close to but those less percents are
I would say the hardest to achieve. We are always
talking about edge cases, for example, so bad weather conditions,
driving in the on the limits of the car, or
things the car has not seen before, so new areas
(10:12):
which you see really you don't see that that often
on public roads, but if they appear, they are quite
big challenge to the car and maybe also for for
human driver. But if you want to drive reliable and
save on the car, you need to manage those those situations.
(10:33):
And the Indian Autonomous challenge for us is an opportunity
to do research on those edge cases. So we are
driving on the limits, we're driving an unstructurally why with
other cars, so there are no lanemarkings. The other car
can use the whole track to defend its its its raceline.
(10:53):
And this is basically research on the edge of autonomous drive.
And we hope with doing research in those areas, we
can limit this last percent and or can can reduce
the last percentage to make autonomous driving safe and reliable
on public roads.
Speaker 2 (11:14):
It sounds like you're trying to plan for the unexpected,
But how do you go about doing that when it
comes to actually writing the codes, writing the algorithms for this?
How do you plan for something that you yourself might
not be able to foresee.
Speaker 1 (11:29):
Well, on the one hand, we are having a lot
of simulations, and we can do a lot of simulations.
We can test things that you won't expect on public
roads that you can think about your even don't want
to test it on public roads because it might be dangerous,
So we can do this in simulation and also on
(11:50):
the real racetrag I mean it's a safe environment. We're
not putting any drivers on risk, and so that's maybe
one side of the metal. So we can test a
lot because it's no danger, we can simulate a lot.
And the other part is that we are already driving
(12:12):
in those areas that would be unexpected for public road traffic.
I would say so driving with two hundred and seventy
kph through corners at the Las Vegas Motor Speedway, that's
probably something real car wouldn't experience. So if you can
(12:33):
manage those situations, we expect a road car to also
function at maybe road conditions that are wet or snowy
at lower speeds, for.
Speaker 2 (12:46):
Example, stepping away from the competition. One question that I have,
and it comes because we've had this conversation with a
lot of folks on this show, is of the challenge
of taking the autonomy that you see in the autonomous
chain challenge, where there are cars that are exclusively driven
by AI drivers, and trying to take that to the
(13:08):
public roads where you have the extra variable of human
drivers interacting with the AI drivers. What kind of challenges
do you foresee as we see a greater proliferation of
autonomous cars going on in public spaces interacting with human
drivers on a more regular basis in the future.
Speaker 1 (13:28):
So this is basically a or big research research question
on itself. So there's mixed traffic between autonomous vehakers and
real vehicles or vegles driven by human drivers, makes it
more and more difficult. Because if you have a lot
or only autonomous speakers on the road, they could communicate
(13:49):
with each other, you might not need any traffic lights
and so on. It would make things a lot a
lot easier. What makes it difficult is that those yeah
autonomous weeklers need to communicate with people on the road
other human drivers and they can't directly communicate to the
(14:09):
human They need to communicate its intentions via lightings or
or et cetera traffic lights and so on. So that's
that's a big, big point for us for our racing competition.
I would say it does not make a huge difference
because we don't have a weekle to vehicle communication. So
(14:33):
for els, it does not make a difference if the
other race car is driven by human or by AI driver.
But in general, of course there's there's a huge difference
if we have a mixed traffic between human drivers and
AI drivers on public roads.
Speaker 2 (14:50):
One question that often comes up is what would happen
if you put a human driver out with the cars
of the Indie Autonomous Challenge at this stage, how would
you view that potential matchup, What do you think the
outcome would be and how close our way to seeing
an AI driver being somewhere on par with what a
human driver could do.
Speaker 1 (15:11):
Currently, saying, if we have a professional race car driver,
we drove at the historic from the one racetrack in
Mansa this year, we are still a bit behind professional
race car drivers. We need to get slower and slower
to the limit of what we can do. Learn the limit,
(15:32):
and learn it also online, which is pretty pretty important.
I think I would say we're a few seconds away
from a from a real race car driver, but expecting
in the future we're getting closer and closer and maybe
sometime we can even beat a real race car driver
on a on the racetrack yard. I think one of
(15:55):
the most important things, or the huge things that make
it make get difference, is if you're alone on the track,
or if you're on the track with other cars, so
you need to interact with them. It's like my behavior
influences the other behavior, the other's behavior and so on.
So that's also something you need to take into consideration,
(16:16):
not just the lab time itself that you're driving in
the end. But yeah, from today's point of view, I
would say we're we're not getting to or we are
not at the point where what the human race car
driver could do, but I think over the next years
we're getting closer and closer.
Speaker 2 (16:38):
When that point is reached, will it be like chess
where a computer program is always going to beat a
human grand master? Do you see that as a as
a potential outcome.
Speaker 1 (16:50):
I couldn't imagine. Yes, I think for for race cause
it's more more difficult than for for for chess. A
chess computer can take a long time for calculations and stuff.
Reacting inside a race car, you have to be quite
fast making your decisions otherwise it's too late, and so
(17:10):
that's another challenge. But if a real race car has
all the informations it needs to make its decisions, a
Thomas autonomous driver could possibly make better decisions than a
real driver.
Speaker 2 (17:27):
Yeah, really interesting. Well you mentioned the new challenge of
the road course going two months getting a chance to
compete there at that historic venue after years spending time
perfecting what the car can do on oval tracks, admittedly
different ones with the Indianapolis, Texas in Las Vegas. Describe
to people who might not understand just how big of
(17:49):
a challenge it is to adapt the software that you
had put together for the oval tracks to a completely
new challenge that was the Monster road course.
Speaker 1 (18:01):
So the the the challenges we had on the ovals
back back then in the last races were mainly on
multi vehicle driving. So the interaction of the different race
cars making the right decision and went to overtake seeing
seeing the gap to overtake, overtaking the right moment, and
and so on. And what was new this year at
(18:21):
the road course racetrack was the longitudinal and lateral dynamics
that changed. We had to break in the right spot
in the right moment, accelerate at the right moment, decide
in a really narrow gap which velocities we can go
through the corners. Either we are too slow and lose
(18:44):
too much time, or we are too fast and just
spin out of the track. And so this was a
major difference to last year's competitions. Another point is we
had a lot of trees and overpasses which were over
the track, which makes it more difficult for the week
(19:09):
to get its own position so to localize it itself.
And when driving at really high speeds, small eras over
time increase and your own position drifts away from your expectation.
And that's also a quite big, big challenge we had
(19:29):
this year on the monthsA road course.
Speaker 2 (19:33):
I'm really interested in how you went about optimizing the
different turns of that track. For a human driver, they
might watch on board footage of other human drivers to
learn breaking points and speeds. They might look at data
from other drivers things like that and then ultimately they
have to go out and experiment for themselves. How did
you approach it, so that presumably you had some baseline
(19:56):
knowledge of what kind of performance might be available, and
then you can learn to optimize from there.
Speaker 1 (20:03):
So our starting point was basically your simulation. So we
had a pretty good simulation of our week the tire
road contact, and this gave us a kind of baseline
on what is possible in the different different turns. And
of course we started with a few percents below below
(20:24):
that limit and tested, watched the data, increased the speed
the velocities through the turns, and that was basically the
progress so the process, so there was always a human
involved currently checking the data, seeing what where we have problems,
adjusting parameters in our software, tuning the software, going out
(20:46):
again and trying to increase speeds accelerations and see how
the car card behaves. So in this process there were
still human involved. But in the future, we're trying to
make the weak to learn this limit by by itself,
So just send it out and slightly increase the the
(21:12):
target target velocity around around the track and then learn
for each curve maybe or for each turn, how fast
the car can go through this turn.
Speaker 2 (21:25):
But going back to the challenge of transitioning from the
oval to the road course, what percentage of your code
or your your programming would you say carried over and
how much did you have to completely rewrite for the
different discipline.
Speaker 1 (21:40):
I would say we didn't have to rewrite anything to
to to make the transition. It was more like just
parameters in a different direction. We had to make some
some trade offs different for the the oval or the
road course. For example, can row the tuning the as
(22:03):
I said, we had we have to break in the
right moment, accelerate at the right time, and that's what
where we had to make adjustments. But in general, we
always try to keep our software generalistic so to work
at different different tracks, different conditions. Also to work if
(22:26):
they're one is, if there's one other car on the track,
if there are four other cars on the check. We
tried to keep our software very general to work under
all those conditions, so we didn't have to rewrite anything.
Of course, we had to create a new simulation model
(22:46):
for the Monster track and so on, but the software
that was running in the car did not really have
to to change. We have to had to use different parameters,
but no major adjustments.
Speaker 2 (23:00):
Yeah, that's really interesting. There's a final few things for you. Then,
you've spent a lot of time in your research in
this field here in recent years. How much of it
would you say you expect to see trickling down into
what we encounter in our daily lives. What kind of
things do you think this is going to make an
(23:20):
impact on in our daily lives in the next five
years or so.
Speaker 1 (23:26):
That's a difficult question, But if you're looking back, I
think racing was always a driver for inventing you stuff
that came into public cars later on. So for example,
in formula one, all those aerodynamics, carbon fiber, active suspension
(23:47):
and stuff that was first first invented in racing and
used in racing, and later on we also saw it
on public roads. And that's also our motivation and our
idea that we can develop algorithms for autonomous racing that
can be later on used on public roads. For example,
(24:11):
we're driving at very high velocities, reacting to other cars
at very high velocities, which needs short reaction times. This
could be for example, interesting for autonomous highway evasive maneuvers.
For example, if the car perceives an an emergency on
(24:35):
the highway tie velocities, it yeah needs to take take action,
for example autonomously to save the driver inside the car,
and maybe that's one one of the areas where the
research can be applied.
Speaker 2 (24:50):
You mentioned earlier that there in Munich you are very
much in the heart of the footprint of the German
automotive industry. Do you see the German automo out of
manufacturers taking interest in what you're doing, the research that
you were doing. What can they learn from the experimentation
that you and your team have done.
Speaker 1 (25:11):
Yes. Sure, we're always in contact with different suppliers manufacturers
and exchanging on the challenges we're currently having, also working
together in some projects on certain solutions. I think research
(25:32):
is also of course important for for the automotive industry.
Here in Munich or in Germany or around the world.
A lot of people are doing research in different areas
of autonomous driving, from perception to planning, control, localization and
so on, and this all comes back to industry in
(25:58):
the future, and that that's why are they're they're of
course interested in the being being an early adapter. For example,
what what are we currently working on? What are our
our main challenges we're we're facing. Yeah, also they are
they are interested in the whole India Autonomous Challenge experience because,
as as I said in the beginning, from from my
(26:20):
point of view, the main problems we have to solve
in the automotive industry are those last two to five
percent to to have autonomous cars on the roads. And
as i'm as I mentioned, the India Autonomous Challenge is
driving in one of those edge cases that are that
are missing for the bigger picture. Well, we'll finish with this.
Speaker 2 (26:42):
Then what has the Indie Autonomous Challenge allowed you to do?
Speaker 1 (26:47):
How?
Speaker 2 (26:47):
How can you use this as a jumping off point
to get to where you want to be personally and
professionally in the next five or ten years.
Speaker 1 (26:57):
Very good question. So the in the Autonomous Challenge allows students,
researchers to apply their their software, their their research on
a real car and tested in a real environment. I
think it's always a huge difference doing or validating your
(27:22):
your research and simulation compared to driving it in the
real car. As I also mentioned, the reality is completely
different from simulations most of the time. Or yeah, that's
that's that's one point people get more and more excited
if they're on the racetrack, seeing seeing their car driving
(27:44):
on the roads, and it's the same same for me.
You gain more, more and more passionate about what you're
working on.
Speaker 2 (27:53):
Insightful is always Simon, Thank you so much for joining
us on the inside track. Thanks for joining us this
week on the inside track. That was Simon Hoffman, the
team lead at the Autonomous Motorsports team at the Technical
University of Munich, giving us an inside look at their
autonomous racing cookbook. I'm your host, Ryan Marine. Give us
(28:16):
a follow on social media and join us next time
to hear from Rue Phillips, the former guitarist for Black
Sabbath turned EV pioneer who's working on many of the
new cars hitting the public roads today.
And that was basically the process, so there was always
a human involved currently checking the data, seeing what where
we have problems, adjusting parameters in our software, tuning the
software going out again and trying to increase speeds, accelerations
and see how the car behaves.
Speaker 2 (00:23):
The world around us 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 Indian
(00:45):
Autonomous Challenge, and see why we call this the inside track.
As many racing engineers were to test, there is no
playbook or owner's manual when you're racing on the cutting
edge of technology. Much like some of the first form
the one racing teams tearing through the streets of Monaco
in the nineteen fifties, today's engineers are building out their
(01:06):
racing stacks from the ground up. This includes the code,
the computers, and even the car's ability to feel what's
happening out on the track. Today, on the inside Track,
we get a chance to look at how the Autonomous
Motorsports team from the Technical University of Munich has been
writing the manual for the World of Autonomous Racing. To
help explain the process, Simon Hoffman, the leader of the
(01:29):
Tomb team, sits down with us to discuss how they
became the first ever winners of the Indie Autonomous Challenge,
some of the obstacles that they overcame to get their
car onto the track, and how the team is able
to have real time conversations about their autonomous car without
any driver feedback. It's great to have you on the
(01:58):
inside track. I'm just really interested in learning a little
bit first about you and how you got into this
interesting competition. So, at what point in your life was
robotics or AI something that became of interest, and then
how soon was that something that you were able to
channel into the Indie Autonomous Challenge.
Speaker 1 (02:17):
So I think in particular, this started pretty pretty early,
back back in school, when I was interested in how
technology works, how different robots and robotic applications, car advanced
driver assistant systems or functioning. I think that's when when
my passion for the topics raised, and of course I
(02:40):
looked into different studies which I could could pursue to
develop more or to to gain more experience in this field.
But I think back then my interest started.
Speaker 2 (02:54):
Did you always have an interest in automotive or is
that something that you stumbled into a little bit later.
Speaker 1 (03:03):
Well, I would say that the basic interest was always there,
but my main interest was across all of robotic applications.
This came more in university when I got into different
lessons touching these applications. I started back then started studying
(03:24):
mechanical engineering at first, or mecatronics, which is more like
a combination of mechanical electronics or informatics. So I wanted
to get the whole picture around robots and the way
they work. And in this area, we have a pretty
(03:46):
pretty good automotive industry here here in Munich, and also
the university has a lot of lectures in this area.
I got got interested more into the application of robotics
in tomortive industria.
Speaker 2 (04:01):
So let's dive in then too. You're getting involved with
the autonomous challenge. At what point during your academic career
there at the Technical University of Munich. Did you become
aware of this and how quickly did you want to
sign up? When you started to learn about what the
IAC was all about.
Speaker 1 (04:19):
So I did my master's thesis here at our share
where I'm currently currently working. Back then it was more
about control system of systems of electric vehicles, so actually
a totally totally different topic. But in after after I
finished my master's thesis, I got the well, the opportunity
(04:41):
to start as a research associate PhD candidate at our
chair working at tele opraded driving, which means basically remotely
controlling cars, so real road vehicles remotely controlled, which actually
comes into place as soon as automated systems don't work
anymore or get to the their functional limits, which is
(05:06):
more more or less or use case for autonomous driving
on public road because if you imagine there is no
driver inside the Weekly anymore, and the Weekly has to
navigate through cities on its own, it might still in
the beginning or even in the future, get to functional limits,
and of course it would be would be a shame
of just standing around there blocking graphic delaying each its
(05:30):
passenger's journey. And then that's where Intel operation comes into
place and that's also the main topic of my PhD
I pursued. And then what about Actually about one year
ago I finished my work on this and I got
the opportunity to join the autonomous Racing team at our chair,
(05:53):
which I was pretty excited about because for me it
actually was was the next the next step working on
the fallback systems and then also working on the real
autonomous week And yeah, it was pretty excited about new challenge,
new challenges and new things to work on. And yes,
(06:13):
in the beginning of this year I started being being
team lead of our autonomous racing team. And yeah, pretty
pretty happy right now with our team and things are
going on, I can imagine.
Speaker 2 (06:26):
So, I mean, the team actually has been a front
running team from the very beginning, going back to the
very first event at the Indianapolis Motor Speedway. And so
it sounds like you joined as the team already had
established some level of success. So what was it like
jumping in kind of mid stream on this project that
was already fairly well developed but still needed to further
(06:50):
advance as the complexity of the challenge became that much greater.
Speaker 1 (06:55):
Well, it feels kind of kind of more more relexing
if you if you're starting, and you also already have
a basis which which works. But still the challenge or
the challenges increased from race to race, which got more
more complicated. And there's still a long way way to
go to get to an autonomous racing or to to
(07:17):
a raising format like real real drivers would race, or
real race card drivers like we see every every week
in formally one for example. And there's still still a
lot to do. But yeah, I was pretty glad about
what what I already could could find in our software stack.
It's very solid software. We with this software we managed
(07:40):
to win the first Indio Autonomous challenge and so really
really good, good basis to start with. And but yeah,
the next challenges are coming.
Speaker 2 (07:51):
And it's it's got to be some pressure too, right
to know that you're jumping into an organization that's already
had success and now they're looking to you to keep
that success going.
Speaker 1 (08:01):
Yeah, of course, if you're if you're joining a winning team,
you only can can lose. Basically as well, if you win,
everyone says yeah, sure, of course the last time, Well
of course, yeah, and then if you lose, it's on you.
But I think this is not how we work here
at the new University. For us, we are always looking
(08:24):
into what research question we can we can answer with uh,
with these challenges, and the challenge is just a fun
side side part of answering those research research questions.
Speaker 2 (08:39):
Yeah, let's let's dive into that a little bit, because yes,
it is an objectively an interesting and cool project to
be a part of. But at a certain point you
have to justify the continued investment by the university in
a project like this. So what are the questions that
you continue to ask, the answers that you continue to
clean from being involved in this project and presumably continue
(09:02):
to looking into the future as well.
Speaker 1 (09:05):
I think first, first of all, we maybe need to
do to make a start a step back. So if
we're looking into the automotive industry, we already see some
cars driving autonomously on on on roads. We are progressing there,
probably not as fast as we initially expected. And for us,
the most important point or let's let's say in other
(09:29):
restaurants to so if we want to see autonomous weekles
driving on car or on public roads really reliable, we
need to get up to a reliability of let's say
ninety nine point ninety ninety nine percent and maybe maybe
even more and realizing probably ninety five percent, I think
(09:51):
we're we're pretty close to but those less percents are
I would say the hardest to achieve. We are always
talking about edge cases, for example, so bad weather conditions,
driving in the on the limits of the car, or
things the car has not seen before, so new areas
(10:12):
which you see really you don't see that that often
on public roads, but if they appear, they are quite
big challenge to the car and maybe also for for
human driver. But if you want to drive reliable and
save on the car, you need to manage those those situations.
(10:33):
And the Indian Autonomous challenge for us is an opportunity
to do research on those edge cases. So we are
driving on the limits, we're driving an unstructurally why with
other cars, so there are no lanemarkings. The other car
can use the whole track to defend its its its raceline.
(10:53):
And this is basically research on the edge of autonomous drive.
And we hope with doing research in those areas, we
can limit this last percent and or can can reduce
the last percentage to make autonomous driving safe and reliable
on public roads.
Speaker 2 (11:14):
It sounds like you're trying to plan for the unexpected,
But how do you go about doing that when it
comes to actually writing the codes, writing the algorithms for this?
How do you plan for something that you yourself might
not be able to foresee.
Speaker 1 (11:29):
Well, on the one hand, we are having a lot
of simulations, and we can do a lot of simulations.
We can test things that you won't expect on public
roads that you can think about your even don't want
to test it on public roads because it might be dangerous,
So we can do this in simulation and also on
(11:50):
the real racetrag I mean it's a safe environment. We're
not putting any drivers on risk, and so that's maybe
one side of the metal. So we can test a
lot because it's no danger, we can simulate a lot.
And the other part is that we are already driving
(12:12):
in those areas that would be unexpected for public road traffic.
I would say so driving with two hundred and seventy
kph through corners at the Las Vegas Motor Speedway, that's
probably something real car wouldn't experience. So if you can
(12:33):
manage those situations, we expect a road car to also
function at maybe road conditions that are wet or snowy
at lower speeds, for.
Speaker 2 (12:46):
Example, stepping away from the competition. One question that I have,
and it comes because we've had this conversation with a
lot of folks on this show, is of the challenge
of taking the autonomy that you see in the autonomous
chain challenge, where there are cars that are exclusively driven
by AI drivers, and trying to take that to the
(13:08):
public roads where you have the extra variable of human
drivers interacting with the AI drivers. What kind of challenges
do you foresee as we see a greater proliferation of
autonomous cars going on in public spaces interacting with human
drivers on a more regular basis in the future.
Speaker 1 (13:28):
So this is basically a or big research research question
on itself. So there's mixed traffic between autonomous vehakers and
real vehicles or vegles driven by human drivers, makes it
more and more difficult. Because if you have a lot
or only autonomous speakers on the road, they could communicate
(13:49):
with each other, you might not need any traffic lights
and so on. It would make things a lot a
lot easier. What makes it difficult is that those yeah
autonomous weeklers need to communicate with people on the road
other human drivers and they can't directly communicate to the
(14:09):
human They need to communicate its intentions via lightings or
or et cetera traffic lights and so on. So that's
that's a big, big point for us for our racing competition.
I would say it does not make a huge difference
because we don't have a weekle to vehicle communication. So
(14:33):
for els, it does not make a difference if the
other race car is driven by human or by AI driver.
But in general, of course there's there's a huge difference
if we have a mixed traffic between human drivers and
AI drivers on public roads.
Speaker 2 (14:50):
One question that often comes up is what would happen
if you put a human driver out with the cars
of the Indie Autonomous Challenge at this stage, how would
you view that potential matchup, What do you think the
outcome would be and how close our way to seeing
an AI driver being somewhere on par with what a
human driver could do.
Speaker 1 (15:11):
Currently, saying, if we have a professional race car driver,
we drove at the historic from the one racetrack in
Mansa this year, we are still a bit behind professional
race car drivers. We need to get slower and slower
to the limit of what we can do. Learn the limit,
(15:32):
and learn it also online, which is pretty pretty important.
I think I would say we're a few seconds away
from a from a real race car driver, but expecting
in the future we're getting closer and closer and maybe
sometime we can even beat a real race car driver
on a on the racetrack yard. I think one of
(15:55):
the most important things, or the huge things that make
it make get difference, is if you're alone on the track,
or if you're on the track with other cars, so
you need to interact with them. It's like my behavior
influences the other behavior, the other's behavior and so on.
So that's also something you need to take into consideration,
(16:16):
not just the lab time itself that you're driving in
the end. But yeah, from today's point of view, I
would say we're we're not getting to or we are
not at the point where what the human race car
driver could do, but I think over the next years
we're getting closer and closer.
Speaker 2 (16:38):
When that point is reached, will it be like chess
where a computer program is always going to beat a
human grand master? Do you see that as a as
a potential outcome.
Speaker 1 (16:50):
I couldn't imagine. Yes, I think for for race cause
it's more more difficult than for for for chess. A
chess computer can take a long time for calculations and stuff.
Reacting inside a race car, you have to be quite
fast making your decisions otherwise it's too late, and so
(17:10):
that's another challenge. But if a real race car has
all the informations it needs to make its decisions, a
Thomas autonomous driver could possibly make better decisions than a
real driver.
Speaker 2 (17:27):
Yeah, really interesting. Well you mentioned the new challenge of
the road course going two months getting a chance to
compete there at that historic venue after years spending time
perfecting what the car can do on oval tracks, admittedly
different ones with the Indianapolis, Texas in Las Vegas. Describe
to people who might not understand just how big of
(17:49):
a challenge it is to adapt the software that you
had put together for the oval tracks to a completely
new challenge that was the Monster road course.
Speaker 1 (18:01):
So the the the challenges we had on the ovals
back back then in the last races were mainly on
multi vehicle driving. So the interaction of the different race
cars making the right decision and went to overtake seeing
seeing the gap to overtake, overtaking the right moment, and
and so on. And what was new this year at
(18:21):
the road course racetrack was the longitudinal and lateral dynamics
that changed. We had to break in the right spot
in the right moment, accelerate at the right moment, decide
in a really narrow gap which velocities we can go
through the corners. Either we are too slow and lose
(18:44):
too much time, or we are too fast and just
spin out of the track. And so this was a
major difference to last year's competitions. Another point is we
had a lot of trees and overpasses which were over
the track, which makes it more difficult for the week
(19:09):
to get its own position so to localize it itself.
And when driving at really high speeds, small eras over
time increase and your own position drifts away from your expectation.
And that's also a quite big, big challenge we had
(19:29):
this year on the monthsA road course.
Speaker 2 (19:33):
I'm really interested in how you went about optimizing the
different turns of that track. For a human driver, they
might watch on board footage of other human drivers to
learn breaking points and speeds. They might look at data
from other drivers things like that and then ultimately they
have to go out and experiment for themselves. How did
you approach it, so that presumably you had some baseline
(19:56):
knowledge of what kind of performance might be available, and
then you can learn to optimize from there.
Speaker 1 (20:03):
So our starting point was basically your simulation. So we
had a pretty good simulation of our week the tire
road contact, and this gave us a kind of baseline
on what is possible in the different different turns. And
of course we started with a few percents below below
(20:24):
that limit and tested, watched the data, increased the speed
the velocities through the turns, and that was basically the
progress so the process, so there was always a human
involved currently checking the data, seeing what where we have problems,
adjusting parameters in our software, tuning the software, going out
(20:46):
again and trying to increase speeds accelerations and see how
the car card behaves. So in this process there were
still human involved. But in the future, we're trying to
make the weak to learn this limit by by itself,
So just send it out and slightly increase the the
(21:12):
target target velocity around around the track and then learn
for each curve maybe or for each turn, how fast
the car can go through this turn.
Speaker 2 (21:25):
But going back to the challenge of transitioning from the
oval to the road course, what percentage of your code
or your your programming would you say carried over and
how much did you have to completely rewrite for the
different discipline.
Speaker 1 (21:40):
I would say we didn't have to rewrite anything to
to to make the transition. It was more like just
parameters in a different direction. We had to make some
some trade offs different for the the oval or the
road course. For example, can row the tuning the as
(22:03):
I said, we had we have to break in the
right moment, accelerate at the right time, and that's what
where we had to make adjustments. But in general, we
always try to keep our software generalistic so to work
at different different tracks, different conditions. Also to work if
(22:26):
they're one is, if there's one other car on the track,
if there are four other cars on the check. We
tried to keep our software very general to work under
all those conditions, so we didn't have to rewrite anything.
Of course, we had to create a new simulation model
(22:46):
for the Monster track and so on, but the software
that was running in the car did not really have
to to change. We have to had to use different parameters,
but no major adjustments.
Speaker 2 (23:00):
Yeah, that's really interesting. There's a final few things for you. Then,
you've spent a lot of time in your research in
this field here in recent years. How much of it
would you say you expect to see trickling down into
what we encounter in our daily lives. What kind of
things do you think this is going to make an
(23:20):
impact on in our daily lives in the next five
years or so.
Speaker 1 (23:26):
That's a difficult question, But if you're looking back, I
think racing was always a driver for inventing you stuff
that came into public cars later on. So for example,
in formula one, all those aerodynamics, carbon fiber, active suspension
(23:47):
and stuff that was first first invented in racing and
used in racing, and later on we also saw it
on public roads. And that's also our motivation and our
idea that we can develop algorithms for autonomous racing that
can be later on used on public roads. For example,
(24:11):
we're driving at very high velocities, reacting to other cars
at very high velocities, which needs short reaction times. This
could be for example, interesting for autonomous highway evasive maneuvers.
For example, if the car perceives an an emergency on
(24:35):
the highway tie velocities, it yeah needs to take take action,
for example autonomously to save the driver inside the car,
and maybe that's one one of the areas where the
research can be applied.
Speaker 2 (24:50):
You mentioned earlier that there in Munich you are very
much in the heart of the footprint of the German
automotive industry. Do you see the German automo out of
manufacturers taking interest in what you're doing, the research that
you were doing. What can they learn from the experimentation
that you and your team have done.
Speaker 1 (25:11):
Yes. Sure, we're always in contact with different suppliers manufacturers
and exchanging on the challenges we're currently having, also working
together in some projects on certain solutions. I think research
(25:32):
is also of course important for for the automotive industry.
Here in Munich or in Germany or around the world.
A lot of people are doing research in different areas
of autonomous driving, from perception to planning, control, localization and
so on, and this all comes back to industry in
(25:58):
the future, and that that's why are they're they're of
course interested in the being being an early adapter. For example,
what what are we currently working on? What are our
our main challenges we're we're facing. Yeah, also they are
they are interested in the whole India Autonomous Challenge experience because,
as as I said in the beginning, from from my
(26:20):
point of view, the main problems we have to solve
in the automotive industry are those last two to five
percent to to have autonomous cars on the roads. And
as i'm as I mentioned, the India Autonomous Challenge is
driving in one of those edge cases that are that
are missing for the bigger picture. Well, we'll finish with this.
Speaker 2 (26:42):
Then what has the Indie Autonomous Challenge allowed you to do?
Speaker 1 (26:47):
How?
Speaker 2 (26:47):
How can you use this as a jumping off point
to get to where you want to be personally and
professionally in the next five or ten years.
Speaker 1 (26:57):
Very good question. So the in the Autonomous Challenge allows students,
researchers to apply their their software, their their research on
a real car and tested in a real environment. I
think it's always a huge difference doing or validating your
(27:22):
your research and simulation compared to driving it in the
real car. As I also mentioned, the reality is completely
different from simulations most of the time. Or yeah, that's
that's that's one point people get more and more excited
if they're on the racetrack, seeing seeing their car driving
(27:44):
on the roads, and it's the same same for me.
You gain more, more and more passionate about what you're
working on.
Speaker 2 (27:53):
Insightful is always Simon, Thank you so much for joining
us on the inside track. Thanks for joining us this
week on the inside track. That was Simon Hoffman, the
team lead at the Autonomous Motorsports team at the Technical
University of Munich, giving us an inside look at their
autonomous racing cookbook. I'm your host, Ryan Marine. Give us
(28:16):
a follow on social media and join us next time
to hear from Rue Phillips, the former guitarist for Black
Sabbath turned EV pioneer who's working on many of the
new cars hitting the public roads today.
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.