Slashdot Mirror
Cadillac SRX Converted Into Self-Driving Car
fergus07 writes "There's been much talk about self-driving cars in recent times and the latest glimpse into this autonomous future comes from Carnegie Mellon University where researchers have loaded a Cadillac SRX with an array of sensors that allow it to manage highway traffic, congested roadways, and even merging on and off ramps."
Not sure what the news is here, except for that CMU now did what a private company has been doing for a while.
http://en.wikipedia.org/wiki/Google_driverless_car
Having a self driving Cadillac would be WONDERFUL!
Old people...
So what happens when a light has a communication error and the car ends up in a bad crash? who is at fault and what happens to the victims who have bills that are pillaging up while the courts are working out who will have to pay them.
if he's talking about medical bills, pillaging up is a perfectly cromulent usage.
as long as they don't make a 1982 Pontiac Trans Am - this is not newsworthy for me
I wonder what the V2V communication is about. I expect that in an ideal world other automated vehicles could communicate obstacles, road condition, and velocity data to one another, but this seems ripe for abuse or exploitation. In a worst-case scenario, someone could use that to either completely stop traffic or to cause an accident.
allow it to manage highway traffic, congested roadways, and even merging on and off ramps.
This is how you know self-driving car tech is not quite ready, when they are bragging about being able to manage an off-ramp.
Seriously folks, we are not going to have a fully autonomous car by 2020.
"First they came for the slanderers and i said nothing."
But the real challenge is to find a manufacturer who's willing to dodge the sueball coming at them if one of those things goes wrong, even if it is (and it will have to be, and will be) 10 times safer than any human driver. Legal problems are going to snuff out this baby at birth.
This raises an interesting question I hadn't previously considered: in most states, liability insurance is required to drive on public roads. What insurance company will insure an autonomous car?
I am usually the last person to take the position, "the free market will take care of it," but in this case ... maybe it can. When the insurance companies are ready to accept the risk, I think I will be as well.
[Sir Garlon] is the marvellest knight that is now living, for he destroyeth many good knights, for he goeth invisible.
This capability has been implemented for years. The DARPA Grand Challenge has had many capable entrants, including (I believe) CMU. All of the described behavior was required years ago in the Grand Challenge.
See DARPA Urban Challege 2007:
http://archive.darpa.mil/grandchallenge/
To even qualify:
National Qualification Event
The NQE for the Urban Challenge was divided into three separate test areas, each with its own flavor and set of challenges:
The NQE A test course required robots to safely merge into and out of two-way traffic in a tight, circulating course. Needless to say, this led to some hair-raising moments for some of the traffic drivers. Besides the complex timing and scoring being recorded by course officials, traffic drivers would alert officials to aggressive behavior with an ever-popular horn blast. Amazingly, in eight days of testing, only one traffic vehicle was actually struck by a robotic vehicle, a testament to the progress of the teams and DARPAâ(TM)s focus on safety.
The meandering NQE B course tested robots on their ability to stay within a lane as they traversed this 2.8-mile course. One section, affectionately termed âoeThe Gauntletâ required the robots to delicately maneuver through a series of parallel parked cars and road obstacles. A final test on the NQE B course required the robots to find an assigned parking spot between adjacent parked cars, then safely pull into and back out of the spot before proceeding on its mission.
NQE C was traffic intensive, consisting of a series of four-way stop intersections for the robot to negotiate, each with its own arrangement of traffic. Robots had to recognize the other vehicles at these intersections, determine the order of precedence and then safely proceed through the intersection when it was their turn. For the second half of the NQE C course, various road blocks were emplaced and the robots were tested on their ability to recognize the road block, execute a U-turn and dynamically replan a new route to complete their mission.
>> Cadillac SRX Converted Into Self-Driving Car
Audi had this problem back in the 80's.
Seriously, considering the demographic that actually buys Cadillac cars, it's the obvious make which needs this first ;-)
Why isn't this electric?
At which point, WTF is the point of the self driving car?
To get to the point where they actually are safer than one being actively piloted by a human. Furthermore even if they never become completely self-driving there will be a lot of very useful spin off technology that is going to come from this research. The legal framework for these cars can be updated when appropriate. That is the easiest problem with the technology since we already know how to do that.
I just don't see people actually wanting this technology, and since we'll never convert all of the cars on the road to this system
Yes people do actually want this technology. In fact I'd go so far as to say people who don't even think they want it actually do even if they don't know it yet. There already are self-parking cars, cars with automatic speed control/braking, stability control, ABS, traction control, navigation aids, drive by wire steering/braking, cruise control, and more. All those things are portions of a driverless vehicle. It's by no means a solved problem but we already rely on a host of technologies to make us better at driving than we could be unassisted.
I would dearly love to be able to get to/from work (~30 minute each way commute) without having to waste an hour every day with the non-productive task of driving. It is a huge waste of my time. It potentially solves other problems as well like helping handicapped people, reducing drunk driving, freeing up huge amounts of non-productive time and more.
We don't actually have to convert all the cars to driverless. It would potentially only take a fraction of them to be self piloting to improve road safety. I assure you that you do not really want my 94 year old grandmother behind the wheel of a car. I would welcome a self driving car to take her around. Any driverless solution will have to be robust enough to deal with unpredictable events at least as well as a human. A well designed system could have better situational awareness than any driver. My field of vision is only about 160 degrees even when I'm not distracted. A computer would have 360 vision day or night, be able to communicate with other vehicles regarding position and speed and direction, be able to react faster than any human, be far less prone to distraction, and actually obey the rules of the road. The engineering obstacles are large but so are the potential benefits.
See comment subject.
A Self Driving car has a whole new layer of computers, servos, pumps, linkages, regulators, etc. that can and will break. It's bad enough you can be stranded in the middle of nowhere when you alternator fails, but now your car may shut down Friday, at 5:15 on the 405 Freeway because a safety feature disables it.
Bad idea all the way around.
When Fascism comes to America, it will call itself Anti-Fascism, and tell you to give up your guns.
My personal experience is the rest of the drivers on the road will still do stupid and random shit, which has a good chance of negating any of the benefits of a car on auto-pilot.
You are assuming that the self driving vehicle would be less able to deal with "stupid and random shit" than a given human. Computers can potentially have greater situational awareness and also react faster than humans. A car could communicate with other vehicles regarding position and speed and road conditions. A car can have 360 degree vision day or night, clear or foggy. A car can estimate distance to the car ahead of it to within millimeters whereas people are often wrong by dozens of meters. Cars can directly sense and respond to unusual road conditions far faster than any human (see traction control) could ever hope to.
When you have a huge fraction of your cars still being older and not using this technology, a lot of the assumptions about how this safe will be goes out the window.
Not necessarily. If a driverless system is well designed there is a good chance it will be BETTER at dealing with all the lunatics out there driving badly. We're nowhere close to that now but it's hardly inconceivable.
Why would that necessarily be any harder for a computer to handle than a person?
33 miles on city streets and highways with speeds up to 65 miles per hour.
This is most impressive. It seems that self driving cars are getting very close to a reality for the public.
Every time an automated car story shows up, a zillion people feel the need to show how sure they are that human drivers can handle more situations than the computers. First of all, why can't they (the ignoramuses posting this stuff) ever accept that hundreds of very smart engineers, not only at Google Research, have taken all sort of 'whoops what happened there' situations into account?
Second, why are these posters incapable of noticing that other transportation systems such as the DC subways or nearly all modern jet aircraft, currently function very well completely autonomously? Yes, Cthulhu might pop up in the middle of a subway tunnel, but a human operator will do no better than the computer in avoiding it.
One more reference: several USAF fighter aircraft are designed for high maneuverability, and are in fact unstable. Only tightly-bound wing surface control loops, fully computerized, keep the things flying in the intended direction.
https://app.box.com/WitthoftResume Code: https://github.com/cellocgw
Audi had this problem back in the 80's.
No they did not. Those allegations against Audi were driver error and a fradulent news report.
It's like...the Cadillac of self-driving cars.
a day will come when bus/cab drivers will be like the elevator operators
I would love to see how this handles someone in the next lane suddenly moving into the car's lane, sideswiping it. It could look at the other side's lane or oncoming traffic too. For example, no oncoming traffic means safe to swerve into oncoming lane to avoid the sideswipe. Oncoming traffic or quickly upcoming traffic from behind, and it could just decide to accept the sideswipe as the least-harmful choice. It would be difficult for a human to take all this into account in under a second when just in regular driving mode (not racing, where they tend to keep 360 awareness).
The main advance is the progression towards real-world sensor selection and packaging. If you look at all the cars which completed the Urban Challenge, and the Google cars, you'll notice the spinning Velodyne laser sensor on the roof. It is a great sensor and makes autonomous driving much easier. Unfortunately, that sucker costs more than most luxury cars and would never be deployed the real-world since nobody wants a spinning can on their roof.
Carnegie Mellon would not have won the Urban Challenge without that sensor or the others littered all over the exterior of the car. The major advance for this new Carnegie Mellon car is comparable performance with cheaper sensors fully packaged within the car. This is a big deal since (a) economics limits which sensors you can buy and (b) the car body and shape limit the size and location of sensors. These obviously limit your overall sensing capability.
The new car also has better computer packaging. Most autonomous vehicles have no trunk space and frequently have no back seat room. For a historical perspective, Carnegie Mellon's Navlab 1, which found a spot and parallel parked autonomously in 1992, had racks of computers and an extra air conditioning system to handle the heat load. Urban Challenge vehicles also had racks in their trunk areas. The Cadillac SRX team was able to cram all the computational gear out of sight. This is really Moore's Law, etc but it is still a respectable achievement.
V2V is peer-to-peer and really focused on reducing reaction times. It allows the car ahead to instantly tell the car behind it is braking. This means less latency for corrective action. This also helps non-autonomous cars since V2V equipped vehicles could, theoretically, suck up some of the shockwaves present in current highway driving.
Then live closer to your work.
Great solution! So when are you buying me a new house?
With 30,,000 dead every year in the US alone this is tech whose time has come. The Google car has already distinguished itself on California's roads as a safe driver. At this point Robots cannot do worse than sensory-limited, distracted, angry, drunk, stoned, tired people behind most wheels. I have been driving for 40 years in some of the worst places. Bangkok, Istanbul, Moscow, Yerevan you name it. And I have had some close calls,
I am so ready for a self driver. Even without a network these cars seem to do well among human drivers. Imagine when all cars are talking to each other. Beautiful. If I want to drive I will go to a track. Or hop into a simulator.
You are traveling down a street at significant speed. A little girl runs into the opposite lane and a car in that lane swerves, to avoid the little girl, and is now on a head on collision course with you. To the side of the road is another child standing in the grass by the curb, if you turn off the road he will be killed. What now?
As a human driver, You're taking the head-on vehicle. It will probably kill you, but that's the right choice and, more importantly, that's YOUR choice.
What does the driverless car do? If the car chooses to go straight, is it not deciding to kill you?
This is a very straightforward scenario, that happens every day.
Sorry to hear about your accident and injuries, but you lost me when you wrote "...when you are between a concrete barrier and an elderly driver that doesn't notice your horn there aren't many options." Now, I admit that I wan't there, and there may not have been any options as you assert, but generally if one is hitting the horn, one would be better served taking some other action instead; like hitting the brakes or otherwise taking evasive manuvers.
You mentioned defensive driving courses. As I understand it, one of the main tenents of defensive driving is "be prepared to control your vehicle, because you can't control the other persons vehicle" (Blowing the horn is essentially attempting to control the other vehicle.)
To get back on topic: This is the thing that the autonomous cars are actually quite good at, reacting to exactly what is happening, not what the "driver" is expecting to happen. That and not over-reacting; such as the tendency for people to steer out of one accident and right into another (often a worse, head on accident).
McFly777
- - -
"What do people mean when they say the computer went down on them?" -Marilyn Pittman
when you consider the demographics.