covering the week's top tech stories
with a slight linux bias a deep learning
model has achieved superhuman
performance at gran turismo sport
over the past few decades research teams
worldwide have developed machine
learning
and deep learning techniques that can
achieve human comparable performance on
a variety of tasks
in order to further assess their
capabilities and performances
some of these models were also trained
to play renowned board or video games
such
as the ancient chinese game go or atari
arcade games researchers at university
of zurich and sony ai zurich have
recently tested the performance of a
deep reinforcement
learning-based approach that was trained
to play the car racing video game gran
turismo sport
their findings further highlight the
potential of deep learning techniques
for controlling cars in simulated
environments
juan long song one of the researchers
who carried out
the studies says autonomous driving at
high speed is a challenging task that
requires generating fast and precise
actions
even when the vehicle is approaching its
physical limits autonomous car racing
where the goal is to complete a given
course in minimal time
features some of these difficulties of
controlling a car close to its physical
limitations
to address these challenges and advance
the frontier we consider the task of
autonomous car racing in the top selling
car racing game
gran turismo sport which is known for
its detailed physics simulation of
various cars and tracks
the key objective of the study is to
develop an artificial neural network
based controller
that can autonomously navigate a race
car within a simulated track
without requiring any prior knowledge of
the car's dynamics
to perform well at gran turismo sport
the controller should be trying to
minimize the amount of time
in which it can complete a given track
the researchers first defined a reward
function that formulated the racing
problem
as a minimum time problem as well as
outlining a neural network policy that
directly mapped input observations to
car control commands
subsequently they trained their neural
network's parameters
using reinforcement learning maximizing
the reward that their model would
receive when
when performing well the researchers
trained the neural network on four
playstation 4s and a desktop pc
remarkably after less than 73 hours of
training
their model had already achieved
superhuman performance
outperforming the best known human lap
times in all three reference settings
including two different cars on two
different tracks
the results highlight the incredible
level of precision that the model can
achieve
and its ability to consistently execute
optimal actions
even in scenarios where these actions
are risky and might lead a human player
to push their car off track
so what exactly would you reward an ai
with
chips what i really love about this
story that becca's sharing with us
is that when i think about ai technology
up until this point when it comes to
that machine learning
it's very like step based
so you think about chess for example so
when an ai plays chess
it's planning out the next move and then
it makes that move
and then it plans out the next move and
then it makes that move
what's happening here in gran turismo
sport is that
the ai the machine learning algorithm
the
the brains of the operation has to
continually
be planning ahead and figuring out
how to make those moves and and it does
it so very well
so it makes me think okay first of all
this could be a really really great way
to improve
the computer players in online games or
in offline games imagine being able
like to play a race car game with
opponents that
match your level of skill
and maybe even surpass it
but by using those computer games as
simulations they're training the ai
even further to learn what works in a
simulated environment where
human lives are never at risk so
would they then be able to take that
learned
methodology that learned intelligence if
you will
and move it over to true autonomous
vehicles that are going to operate
on the road and are able to
basically think for themselves and plan
how are they going to pass a car without
actually hitting that car
and those kinds of things that right now
we're just not quite there yet
with autonomous vehicles but could this
be kind of a precursor to the next step
of artificial intelligence based
autonomous vehicles and then that leads
to
everything else i think about space
flight and and
everything that is to come it's really
an exciting time
you