My name is Nathaniel Fairfield I'm a software engineer at Waymo. Back when we
started this we were the Google self-driving car project and self-driving
cars didn't really exist. Thinking about our first thousand miles and then our
first ten thousand miles and our first hundred first million you've actually
gotten breadth and diversity of mileage and you start to see some of the
patterns of driving. The 10 million mile milestone is about 10 lifetimes of
driving, it's a huge amount of experience that we've gained across the country and
all kinds of driving conditions and it takes that kind of experience to learn
all of these lessons to really make that possible. How does Waymo see the ball
before anyone else? It's always looking, it's always looking in all directions
the lasers are what lets it spot a small moving object that's approaching towards
the car and estimate its position and velocity very precisely. So this is a
very interesting case where there's actually a couple with a dog at the same
time a jogger is just passing by and the jogger swings around the dog out of the
bike lane into the road. The car doesn't just understand where the agents are it
actually understands how they're gonna interact in the near future and is able
to make those predictions in the blink of an eye before you can really
understand exactly what's going on. This is night construction which you
know as a human driver is one of the more stressful things. The car handles it
really really really well because the car sees as well at night as it does
during the day for the lasers and for the radars they see where the cones are
where that aisle of traffic that's being laid out here by the construction zone
is going. Red lights and green lights they're
really important but people sometimes don't obey that. In a case like this
where the red light runner happened the car is actually checking all around in
all directions all at once and so it can see when there's somebody coming who
looks like wait they're not really slowing down for their red light and at
some point we decided wait they actually look like they're gonna run the red
light let's stop and let them go through and then after they've gone you know on
we go. So the thing that's amazing about this one is that you can't see in in
dust storms like these there's actually a pedestrian out there that the car
sees perfectly well in advance they're just walking across the road the car
sees it using its lasers and radars and starts slowing down before you as a
human even understand what's happening. A regular person has a lot of built-in
assumptions about how the world works. The car is really thinking and is
designed to consider all the worst case stuff that we have ever seen and how
that can play out so that makes it in many ways a very careful driver that is
thinking through all the repercussions that you might not be doing. It doesn't
get sleepy or drowsy or drunk or distracted by cell phones or distracted
by the children in the backseat. The car is always paying attention to all of
these factors and trying to think ahead. In the early days we used to joke how
many engineers does it take to drive a driverless car because there'd be a
bunch of people watching the thing and so every mile it drove they were a whole
pile of people attending it. Those first 10 million miles were about learning and
growing and proving things out. The next 10 million the next hundred million the
next billion miles are about the cars just driving themselves driving safely
driving people where they need to go and really making it easy and safe for
people to get around and that's super exciting
Không có nhận xét nào:
Đăng nhận xét