Why Self-Driving Cars DON'T Just CRASH

Techquickie ·Techquickie ·2019-05-06 · 1,117 words · ~5 min read
Floatplane YouTube

Transcript

JSON SRT VTT 85
0:00 i don't think anything says we're living in the future quite like the
0:04 proliferation of self-driving cars now granted they're not exactly all over the
0:09 roads yet but given the huge investments that automakers and tech companies are
0:13 making in them it seems like it's a matter of when rather than if
0:19 but how exactly do they work i mean most
0:22 of our trains aren't even self-driving and those things are quite literally on
0:27 rails well as you'd expect self-driving cars
0:31 are jam-packed with tons of equipment to help them both see and understand their
0:36 environment some of it is rather familiar such as a data connection for
0:40 traffic info and a gps transponder to allow the car to know where it is but
0:46 while gps is fine for providing driving directions in a human-controlled car it
0:50 has a margin of error of up to several meters and besides to control a car with
0:56 such a slow and error prone system would result in more than a few fender benders
1:01 so to start with self-driving cars also use freaking laser beams to build a map
1:07 of what's around them lidar which you can learn more about up
1:10 here can be built into units that can spin
1:14 360 degrees shooting invisible lasers in
1:17 all directions then measuring how long it takes for each beam to hit an object
1:23 and bounce back this allows the car to judge not only the distances but even
1:27 the shapes of the objects immediately around it giving the car even more
1:32 information is traditional radar to help it gauge speed gyroscopes and
1:36 accelerometers to provide more movement data than would be possible with a
1:39 traditional speedometer and high resolution cameras now you might think
1:44 well we've already got frickin laser beams
1:48 what are the cameras for well although the car's other systems give it a pretty
1:53 good idea of what's in the general vicinity the cameras really help provide
1:57 a complete picture as they can see in color this helps your car distinguish
2:02 let's say a caution sign from a construction sign so you put all of this
2:06 together and the car has a coherent 3d
2:10 map that provides the data that it needs to make decisions for example lidar and
2:15 video can determine whether that thing up ahead is telling the car to stop or
2:20 yield whether the vehicle in front is a small sports car or a large truck so that it
2:24 can decide whether to pass or whether that two-wheeled contraption is a
2:29 motorcycle or a bicycle so the car can get around the cyclist and give her a
2:34 bit more space also many self-driving cars have
2:37 ultrasonic sensors in the wheels so the car will know how close it is to the
2:41 curb and other vehicles while parking and even microphones so that they can
2:46 hear a police or ems siren to get out of the way in a timely manner i mean that's
2:51 going to be one improvement on the roads around here once self-driving cars are ubiquitous people can be so rude they're
2:57 going to an emergency anyway that's how all the data gets collected
3:02 but what about the processing an experienced driver can take in
3:06 information from the environment filter out everything that's not important and
3:10 make all the kinds of decisions we talked about before in a fraction of a
3:14 second so of course then you need a lot of
3:17 computing power for a self-driving car to pull off the same thing this is made
3:22 possible by running many processors in
3:25 parallel to crunch these numbers it's actually quite similar to how desktop
3:30 gpus work these types of processors also happen to
3:33 be more trainable with machine learning so for example you can teach a car what
3:38 a pedestrian looks like by showing it a large data set with lots of photos of
3:43 people crossing the road but what if a self-driving car didn't only have to
3:48 rely on its own sensors and processors well another goal is to have these cars
3:52 communicate with each other while they're on the road so they can
3:56 actually tell other cars what they're doing and why enhancing safety and
4:02 taking some of the guesswork away from the individual vehicles and if you think
4:07 about it having many cars working together like this in unison could help
4:11 with lots of other issues too like easing congestion by coordinating
4:15 movement so that traffic can keep flowing smoothly
4:18 also in the future especially once faster 5g connections become more common
4:23 we could even see smart infrastructure that communicates with the cars i mean
4:28 think about a parking garage that could tell a car that it's too full to
4:31 accommodate it or transponders in construction zones that could tell them
4:35 to slow down and prepare for narrower lanes of course we do still have a long
4:40 way to go before this kind of tech is commonplace but given how many
4:43 semi-autonomous cars are available for purchase today and how much money is
4:48 being poured into this whole endeavor along with the ever-rising processing
4:52 and data speeds we continue to enjoy the hope
4:55 is that in the near future our car trips will be much easier and safer since well
5:01 over 90 percent of car accidents are attributable to human error and speaking
5:05 of self-driving cars today's episode was brought to you by ibm spectrum storage
5:10 did you know that a self-driving car can generate up to 15 terabytes of data
5:16 every hour and lots of this is actually not even used while they're driving so
5:22 it's really common in the ai industry to have to ingest just tons of this raw
5:26 data and transform it after the fact while the car's not even driving into
5:32 intelligence and ibm is using spectrum storage to help automakers manage all
5:37 this data so it's optimized for ai and machine
5:40 learning with industry-leading GPU accelerated servers and ibm spectrum
5:44 scale software-defined storage combined with groundbreaking performance and
5:49 simplified deployment so get the fastlane to insights into
5:53 whatever field you're in by learning more about data storage for ai and ibm
5:57 storage solutions for autonomous driving at the link below so thanks for watching
6:02 guys like dislike check out our other videos leave a comment with video
6:05 suggestions and don't forget to subscribe and follow
6:10 not this one i mean this one too but but but also Techquickie that's a button
6:13 below