Jump to content


Photo
- - - - -

driverless cars to hit california ( sorry!)


  • Please log in to reply
454 replies to this topic

#401 Bloggsworth

Bloggsworth
  • Member

  • 7,475 posts
  • Joined: April 07

Posted 27 January 2014 - 10:02

The legal arguments will probably hinge around timing. Did the autonomous system which found itself unable to cope with the situation hand control back to the driver with sufficient time available for the driver to have any affect on the situation, and was it even possible for the driver to override the computer when the driver perceived the danger when the computer's sensors didn't? If the computer gives up only 1 second before the collision and beeps "Over to you matey!", then there is no way the driver could avoid the accident, so how could he be responsible - It will be a legal minefield. How will such cars cope with a sudden snowstorm which blinds their electronic visual cortex, when heavy rain and/or hail confuses its radar? Will the car just stop and beep "Unable to proceed" while behind it, the latest and most expensive Mercedes with a marginally better visual system carries on and...



Advertisement

#402 mariner

mariner
  • Member

  • 1,381 posts
  • Joined: January 07

Posted 27 January 2014 - 18:28

The legal arguments will probably hinge around timing. Did the autonomous system which found itself unable to cope with the situation hand control back to the driver with sufficient time available for the driver to have any affect on the situation, and was it even possible for the driver to override the computer when the driver perceived the danger when the computer's sensors didn't? If the computer gives up only 1 second before the collision and beeps "Over to you matey!", then there is no way the driver could avoid the accident, so how could he be responsible - It will be a legal minefield. How will such cars cope with a sudden snowstorm which blinds their electronic visual cortex, when heavy rain and/or hail confuses its radar? Will the car just stop and beep "Unable to proceed" while behind it, the latest and most expensive Mercedes with a marginally better visual system carries on and...

All that is true but I think it starts with a very basic question.When the "user" gets in the car and tells it where to go, is he/she  then ( in Uk legal parlance) " in charge of the vehicle"?

 

The pilot of a modern plane can basically power up the systems, plug in the ramp and destination co-ordinates and then start the engnes when cleared for push back. The plane will then pretty much fly itself but he/she  is in charge legally and operationally a soon as he/she  climbed in. On that basis the care "user" is always in charge and, therefore , should be prepared to take over manual control at any time. I agree he/she could argue it "happened too fast " in mitigation but I think the onus is on the user to prove it was too quick.If he /she was looking forward at an upcoming collision risk-  why would the law expect any lower standard of vigilence than an old fashione driver?


Edited by mariner, 27 January 2014 - 18:30.


#403 munks

munks
  • Member

  • 344 posts
  • Joined: January 03

Posted 27 January 2014 - 18:39

What? Signage like this on the A3 in Raynes Park?

A computer would be able to process all those signs MUCH faster than a human.



#404 Bloggsworth

Bloggsworth
  • Member

  • 7,475 posts
  • Joined: April 07

Posted 27 January 2014 - 21:12

In this one, what would it make of the 30MPH limit sign?

 

4_Sign.jpg



#405 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 27 January 2014 - 22:42

Ha Ha!

I think an autonomous car would have less trouble with that mess than any human!



#406 275 GTB-4

275 GTB-4
  • Member

  • 7,001 posts
  • Joined: February 03

Posted 27 January 2014 - 22:48

A computer would be able to process all those signs MUCH faster than a human.


Sure and the computer may mis-identify some for a variety of reasons and use wrong assumptions in its decision making process, generally, humans are fairly adept at picking out individual trees in the forest rather than vice versa.



#407 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 27 January 2014 - 22:52

I recently went to the UK research site where the world's first auto landing syem was developed for BEA. The standard set was one failure in 10 miliion reckoned to be ten times better than a human pilot risk level.

 

That started me thinking about this thread - not the abilty or not of systems to drive cars but the operating philosohy.

 

Auto land is RATHER critical so it is well developed.Nonetheless a plane has one big advantage over a car - it can move anywhere in 3D space if needed subject to nearby aircraft so as to retry any approach.

 

A car can't to do that - it can only move slightly left or right, accelerate or stop. So in many ways its a tougher environment.

 

More fundementally the one thing a flight crew does NOT do is let the sytems land , or even fly, the plane without monitoring. You seldom see the flight crew back in the cabin as a plane auto lands at , say, Heathrow in bad visibilty.

 

Despite landing sytems with better than one error risk in ten million the airline industry still expects the "driver" to be monitoring the system ready to intervene. That is very, very different to the sales pitch for  driverless cars. " It will drive itself safer than you can".

 

However good the systems become they will fail. Then the legal reponsibility of the " driver" wil be at issue. It would not surprise me at all if the " driver" is held to be always responsible for monitoring the car just as the pilot is during auto land etc.

 

That may sound trivial or obvious but it removes two big social advantages of driverless cars - the elderly/blind and the  young will not be able to use them autonomously and thus they wont add to total personal mobility beyond current cars. Also no driver wil be able to stop watching the road ahead etc. unles he wants to drive without valid  insurance cover.

The bar is not set very high when it comes to road safety:

- World annual road deaths - 1 million +.

- World annual airline deaths about 500.

Consequently it will be quite easy for autonomous vehicles to be an order of magnitude better than human drivers.

 

On personal mobility. Google is claiming that AV's will be able to "go park themselves" after dropping the humans at their destination. This clearly suggests that human oversight will not be required.



#408 Buttoneer

Buttoneer
  • RC Forum Admin

  • 16,801 posts
  • Joined: May 04

Posted 28 January 2014 - 15:21

Watch out for hackers.

 

http://www.theguardi...urity-executive



#409 BRG

BRG
  • Member

  • 11,663 posts
  • Joined: September 99

Posted 25 August 2014 - 19:43

The Empire is starting to strike back against the Google rebels....

 

I suppose this means that legislators and the insurance industry are starting to take the concept of autonomous cars seriously.  Whether this will put the mockers on the whole idea remains to be seen.  But if this is the reaction in go-ahead hi-tech California, I can imagine conservative old Britain being even more cautious, once the DoT actually wake up and notice this.


Edited by BRG, 25 August 2014 - 19:43.


#410 indigoid

indigoid
  • Member

  • 384 posts
  • Joined: March 04

Posted 31 August 2014 - 13:42

I was actually.

Only limited by the resolution of the terrain mapping which can easily be quite high. Car knows roughly where it is courtesy of GPS. Terrain map has the gradient info so the system can infer pretty accurately where the terrain is headed beyond the water's edge.

 

 

 

I don't think the data actually exists for that, at least not on a global scale. NASA got down to 1 arc-second (~30m at sea level?) by combining SRTM and other data. My partner is doing fun stuff with that data set at her work. Supposedly there will be a 12m data set sometime this year from this: http://www.astrium-geo.com/terrasar-x/

 

I'm assuming Google's Street View hardware logs altitude as well as the usual location. They'd be silly not to, really. So they probably have the data already for the majority of places that self-driving cars are likely to want to go.



#411 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 31 August 2014 - 23:46

We are talking about roads aren't we? Perhaps in the third world they don't care too much, but levels are essential engineering data for drainage design etc. What are surveyors for?



#412 mariner

mariner
  • Member

  • 1,381 posts
  • Joined: January 07

Posted 25 September 2014 - 08:54

There is a lot of promise in these sytems but I still worry that the the part of the software industry driving it , and able to fund it, simply doesn't have the relaibilty mindset to do something so safety critical.

 

http://www.bbc.co.uk...nology-29358986

 

If the world's richest software busines can't even get things right on its 8th iteration of a core product how long wil it take them to produce the virualy infallible software and systems needed ro driverless cars?



#413 imaginesix

imaginesix
  • Member

  • 5,690 posts
  • Joined: March 01

Posted 25 September 2014 - 09:24

The driving software only has to be safer than humans at the outset, and then it can improve gradually with subsequent iterations. A pretty low threshold to entry if you ask me.



#414 GreenMachine

GreenMachine
  • Member

  • 791 posts
  • Joined: March 04

Posted 25 September 2014 - 09:49

Not cars, not public roads, but getting closer ... http://www.canberrat...924-10lb25.html



#415 275 GTB-4

275 GTB-4
  • Member

  • 7,001 posts
  • Joined: February 03

Posted 25 September 2014 - 21:13

Not cars, not public roads, but getting closer ... http://www.canberrat...924-10lb25.html


Correct...still confined to a defined area with easily controlled boundary's and criteria...let em out on the road and see how they go (let me get well out of the way first tho!)

#416 CSquared

CSquared
  • Member

  • 625 posts
  • Joined: December 09

Posted 26 September 2014 - 00:36

There is a lot of promise in these sytems but I still worry that the the part of the software industry driving it , and able to fund it, simply doesn't have the relaibilty mindset to do something so safety critical.

 

http://www.bbc.co.uk...nology-29358986

 

If the world's richest software busines can't even get things right on its 8th iteration of a core product how long wil it take them to produce the virualy infallible software and systems needed ro driverless cars?

imaginesix said it. Driverless cars don't need to be anywhere near infallible, they only need to be less fallible than the average commuter, and that's a pretty low bar, at least on the roads around here.



#417 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 26 September 2014 - 04:06

Perhaps the software for autonomous cars would be tested to different standards of reliability and durability than for the latest smartphone?



#418 bigleagueslider

bigleagueslider
  • Member

  • 860 posts
  • Joined: March 11

Posted 30 September 2014 - 05:16

What benefit is a "driverless car" if at least one person is still usually riding inside? I can see benefit of a car that can operate on its own when positioned in a single lane of traffic with vehicles ahead and behind. Since this would improve traffic flow on freeways/highways with normal driving conditions. But I could never imagine autonomous cars ever being safe enough for driving in conditions where there is ice/snow.



#419 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 02 October 2014 - 04:00

You're kidding aren't you?

 

Check some of the videos of autonomous cars driving autocross courses etc - faster than humans. Same will apply for low traction conditions.

 

http://www.youtube.c...h?v=krJmTZ-TcMc



Advertisement

#420 275 GTB-4

275 GTB-4
  • Member

  • 7,001 posts
  • Joined: February 03

Posted 02 October 2014 - 08:29

You're kidding aren't you?
 
Check some of the videos of autonomous cars driving autocross courses etc - faster than humans. Same will apply for low traction conditions.
 
http://www.youtube.c...h?v=krJmTZ-TcMc


Watched it three times and couldn't see and snow and ice...did I miss something?

#421 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 02 October 2014 - 08:50

Well the computer is doing a better job than humans on a high mu surface. Adapting to a different mu is easier for a computer than a human. Do I really have to point this out?



#422 Canuck

Canuck
  • Member

  • 1,662 posts
  • Joined: March 05

Posted 02 October 2014 - 12:45

Based on that video? Yes. That looks like little more than a well-programmed series of actions. It's not self-driving, it's running a set of instructions that define specific, pre-determined, non-input based control movements. It's automated, not self-driving. Quite apart from the non-example in the video, I would agree that technology is vastly superior in detecting and addressing changing surface conditions. More than once, in my ancient, over-powered winter-driven Cheryl pickup, I'd notice the engine speed not matching the ground speed in between blaring stereo tracks. The lack of a locking differential kept me out of the icy ditch more than once as one tire started slipping while ascending some long, black-ice covered hill.

#423 desmo

desmo
  • Tech Forum Host

  • 13,072 posts
  • Joined: January 00

Posted 02 October 2014 - 15:05

On a circuit course I assume drivers will be using a mix of open and closed loop control techniques, so I don't think the criticism that the automated car is "not driving" is entirely consistent with how things normally work. In the video it is implied that the automated car is drifting on what appears to be an inconsistently wetted a track- which if it is the case is a pretty good analog for winter driving and not something likely to work with an open loop control system running a preprogrammed routine.

#424 imaginesix

imaginesix
  • Member

  • 5,690 posts
  • Joined: March 01

Posted 03 October 2014 - 01:43

You're kidding aren't you?

 

Check some of the videos of autonomous cars driving autocross courses etc - faster than humans. Same will apply for low traction conditions.

 

http://www.youtube.c...h?v=krJmTZ-TcMc

I don't understand most of bigleagueslider's objections, but the concern over poor weather driving is justified. It's not an issue with control in low grip, but in both recognizing spots of low grip in advance, and being able to judge lanes and driver behaviour when everything is coated by a layer of white. Road signs can get obscured, curbing can disappear, piles of snow end up in different places on the road, entire lanes are reshaped by drivers, who tend to follow the tire tracks of the car in front. Right now driverless cars can't even identify potholes. They can't distinguish a crumpled piece of paper from a stone in the road. So we're a long way off even in good weather, and I find it hard to see how they'll overcome the weather concerns I mentioned, while sharing the road with human drivers.



#425 imaginesix

imaginesix
  • Member

  • 5,690 posts
  • Joined: March 01

Posted 03 October 2014 - 01:47

Elon Musk says next year's Tesla cars will be able to self-drive 90 percent of the time.

 

http://www.theverge....ent-of-the-time

 

Hopefully this is just marketing bluster, protected from any requirement to deliver the goods by the wording "able to".


Edited by imaginesix, 03 October 2014 - 01:47.


#426 Kelpiecross

Kelpiecross
  • Member

  • 902 posts
  • Joined: October 10

Posted 03 October 2014 - 03:59


Surely no one really has a name like "Elon Musk".

#427 275 GTB-4

275 GTB-4
  • Member

  • 7,001 posts
  • Joined: February 03

Posted 03 October 2014 - 08:28

Unlike Chess where every known move has been jotted down so that big blue could whip Kasperov's ass with brute force and ignorance....I will bet my Sebastian Lobe in a timed trial against any computer controlled land vehicle from A-B :wave:



#428 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 03 October 2014 - 11:11

You would lose that bet I am afraid. Soon if not already.



#429 BRG

BRG
  • Member

  • 11,663 posts
  • Joined: September 99

Posted 03 October 2014 - 18:50

Surely no one really has a name like "Elon Musk".

I thought it was an anagram, but I tried some online anagram solvers and they came up with nothing. 



#430 Canuck

Canuck
  • Member

  • 1,662 posts
  • Joined: March 05

Posted 03 October 2014 - 21:28

My first reaction to Desmo's use of "assume" and "implied" was "isn't it interesting how our bias filters our perceptions".  I may very well be wrong and Desmo correct as my filters suggest an entirely different conclusion.

 

If you watch again, the entire surface where the car is shown to be drifting is rather uniformly wet despite the pre/post shot that may lead one to conclude that the drifting surface is of mixed mu.  However, entirely apart from that my supposition that the car is simply executing some pre-programmed actions is based on the pylon course.  A self-driving car would not serpentine the pylons, it would avoid the first cone and continue - unless it's exercising a pre-programmed maneuver or series of maneuvers.



#431 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 04 October 2014 - 09:58

It is not possible to drift a car to a predefined course using an open loop control system. Drifting on tarmac - wet or dry - is an unstable system and feedback (throttle, steering or both) is required to "balance" the car in its oversteering attitude.



#432 275 GTB-4

275 GTB-4
  • Member

  • 7,001 posts
  • Joined: February 03

Posted 04 October 2014 - 10:51

ummmm so why can't the appropriate sensors and controllers form a closed-loop system...



#433 Kelpiecross

Kelpiecross
  • Member

  • 902 posts
  • Joined: October 10

Posted 04 October 2014 - 10:59

I thought it was an anagram, but I tried some online anagram solvers and they came up with nothing.


Anagram? "Lone Skum" maybe?

#434 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 04 October 2014 - 11:16

ummmm so why can't the appropriate sensors and controllers form a closed-loop system...

I've lost track of what we are arguing about here - or even whether we are actually arguing?  :)



#435 275 GTB-4

275 GTB-4
  • Member

  • 7,001 posts
  • Joined: February 03

Posted 04 October 2014 - 22:14

I've lost track of what we are arguing about here - or even whether we are actually arguing?  :)


Ahhh huh...right, hmmmmm :well:



#436 BRG

BRG
  • Member

  • 11,663 posts
  • Joined: September 99

Posted 05 October 2014 - 19:15

I've lost track of what we are arguing about here 

You see, that's the problem with human pilots, they lose track.  A computer would have done much better.



#437 imaginesix

imaginesix
  • Member

  • 5,690 posts
  • Joined: March 01

Posted 05 October 2014 - 20:42

You see, that's the problem with human pilots, they lose track.  A computer would have done much better.

LOL thread wipeout!


Edited by imaginesix, 05 October 2014 - 20:43.


#438 Greg Locock

Greg Locock
  • Member

  • 4,534 posts
  • Joined: March 03

Posted 05 October 2014 - 21:43

An ESC+cruise control  already knows how to do drifting so far as the throttle response goes, unfortunately that is dialled back in production releases, but you'll find in a Falcon or Territory that it will add throttle to maintain low yaw rates if the tail is out, on gravel roads. Of course what usually happens in a production cal  is that the brakes activate which knocks it out of cruise which then allows the car to slow which is very boring.

 

There is a huge interest in low mu prediction, my guess is once C2C becomes standard we'll build realtime maps of mu, for each piece of road. However the idea of remote sensing it via cameras, radar, lasers etc is hopeless at the moment, the car ends up driving like your grandmother due to the number of false positives.



#439 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 06 October 2014 - 00:12

C2C? Coast to coast?



Advertisement

#440 Greg Locock

Greg Locock
  • Member

  • 4,534 posts
  • Joined: March 03

Posted 06 October 2014 - 01:29

C2C=car 2 car

 

C2X is car to anything, such as the guvmint's central database, the traffic lights, or pedestrians.



#441 imaginesix

imaginesix
  • Member

  • 5,690 posts
  • Joined: March 01

Posted 06 October 2014 - 06:52

AKA as V2V



#442 Dmitriy_Guller

Dmitriy_Guller
  • Member

  • 4,081 posts
  • Joined: July 01

Posted 09 October 2014 - 02:47

Unlike Chess where every known move has been jotted down so that big blue could whip Kasperov's ass with brute force and ignorance....I will bet my Sebastian Lobe in a timed trial against any computer controlled land vehicle from A-B :wave:

I think you overestimate Big Blue's brute force and ignorance.  It's true that you can solve chess by brute force, but it's going to take you billions of trillions of years.  Since all professional chess matches are played with time control, such brute force computer strategy is not practical.



#443 Superbar

Superbar
  • Member

  • 409 posts
  • Joined: June 01

Posted 14 October 2014 - 21:05

Elon Musk says next year's Tesla cars will be able to self-drive 90 percent of the time.
 
http://www.theverge....ent-of-the-time
 
Hopefully this is just marketing bluster, protected from any requirement to deliver the goods by the wording "able to".


Well, Elon Musk tends to get a little carried away sometimes when he talks(much to the stress of the engineers...), but Tesla Motors has been remarkably good at delivering so far. Since a few weeks back ALL Tesla cars are produced with the necessary hardware for the "Autopilot" self driving features, but not the software. At this time only two driver assist features (speed sign warning and lane departure warning) are activated, but more functionality will come as updates whenever it is deemed ready.

#444 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 15 October 2014 - 23:42

Audi claims a driverless Audi RS7 is set to lap Hockenheim at the same pace as a professional racing driver. 

 

 

http://www.gizmag.co...3829df-90270322



#445 Bloggsworth

Bloggsworth
  • Member

  • 7,475 posts
  • Joined: April 07

Posted 18 October 2014 - 07:53

While chess programs use opening books and endgame tablebases, they also crush us is middle game where they can't draw from memory.

For the most part, it's not your Sony being slow. The broadcast data stream is such that it can't start displaying at any point.There is also some image processing the Sony might like to do, because the raw video stream looks so awful. It needs to buffer some frames and compare them so there's a small delay. Broadcast TV is so last century...

Speed is where computers beat us handily. We are terrible. Average driver reaction time is over 2 seconds. Even at best we can't go much lower than 1 second. Computer's car has stopped before we even start braking.

There are of course many scenarios where human decision making leads to better outcome. But there are many where computers do better. There are still positions in chess where a computer can't find the winning move, but human players can.

 

Unmitigated twaddle. The computer chess programs ONLY draw from memory, they have no self awareness.

 

And there's no "Image processing" that the car's computer might need to do when it comes across a situation it hasn't been pre-programmed to deal with before it says "I can't handle this" and crashes?

 

Your reaction time may be 2 seconds, but my 75 year-old mother-in-law's certainly weren't - Also, humans have the imagination to envisage the consequences of random events and take the appropriate action; like driving up the central reservation to avoid a car coming up too fast from behind and unable to stop, something that I have done twice in my 56 years of driving. I would suggest that computer driven cars may be unable to take that sort of action, as their programming would be such that they are unable to "leave the highway/carriageway."

 

It is the "Many scenarios" which are the ones where accidents happen, not the routine everyday driving.


Edited by Bloggsworth, 18 October 2014 - 07:54.


#446 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted 18 October 2014 - 08:13

It is the "Many scenarios" which are the ones where accidents happen, not the routine everyday driving.

Yes and I think it has been pointed out previously in this thread that the majority of such scenarios arise from a failure to perceive the threat in time - due to inattentiveness, inebriation, drowsiness, excessive speed, bravado, foolhardiness etc etc.


Edited by gruntguru, 18 October 2014 - 08:13.


#447 Bloggsworth

Bloggsworth
  • Member

  • 7,475 posts
  • Joined: April 07

Posted 18 October 2014 - 22:24

Yes and I think it has been pointed out previously in this thread that the majority of such scenarios arise from a failure to perceive the threat in time - due to inattentiveness, inebriation, drowsiness, excessive speed, bravado, foolhardiness etc etc.

 

But not always by a driver - Drunken pedestrians, children thoughtlessly chasing a ball between parked cars, unleashed dogs. Will driverless cars look underneath parked cars for the feet of children and animals, will it look for reflections from the sides of parked cars or shop windows to garner advanced warning of other vehicles? Will it spot an approaching car radio aerial above the hedge on a country road, or hear the throb of a tractor which may suddenly emerge from behind the hedge? What would it make of a couple of reflective metallised party helium balloon drifting across the road? Would the radar perceive them as solid objects and brake violently?


Edited by Bloggsworth, 18 October 2014 - 22:26.


#448 imaginesix

imaginesix
  • Member

  • 5,690 posts
  • Joined: March 01

Posted Yesterday, 01:06

But not always by a driver - Drunken pedestrians, children thoughtlessly chasing a ball between parked cars, unleashed dogs. Will driverless cars look underneath parked cars for the feet of children and animals, will it look for reflections from the sides of parked cars or shop windows to garner advanced warning of other vehicles? Will it spot an approaching car radio aerial above the hedge on a country road, or hear the throb of a tractor which may suddenly emerge from behind the hedge? What would it make of a couple of reflective metallised party helium balloon drifting across the road? Would the radar perceive them as solid objects and brake violently?

This offers some of the answers; http://www.technolog...riverless-cars/

 

Basically, if you can think of it, they will test for it. Of course it will take generations to optimize electronic drivers, but it's not hard to imagine that they will be a big leap forward in safety over human drivers even when they're launched.



#449 desmo

desmo
  • Tech Forum Host

  • 13,072 posts
  • Joined: January 00

Posted Yesterday, 03:14

But, but even if they prove quantitatively much safer than human drivers, think of all the bizarre and unlikely scenarios one could concoct where they might not work! The children! Be afraid! The children!

#450 gruntguru

gruntguru
  • Member

  • 5,361 posts
  • Joined: January 09

Posted Yesterday, 07:24

But not always by a driver - Drunken pedestrians, children thoughtlessly chasing a ball between parked cars, unleashed dogs. Will driverless cars look underneath parked cars for the feet of children and animals, will it look for reflections from the sides of parked cars or shop windows to garner advanced warning of other vehicles? Will it spot an approaching car radio aerial above the hedge on a country road, or hear the throb of a tractor which may suddenly emerge from behind the hedge? What would it make of a couple of reflective metallised party helium balloon drifting across the road? Would the radar perceive them as solid objects and brake violently?

For every accident caused by driverless cars (including all the fantastic scenarios dreamed up by the Luddites of the world) there will be dozens avoided by removal of human error. Infallible drivers like yourself will still have the option of piloting your own vehicle while enjoying the vastly reduced risk to yourself, from sharing the road with autonomous vehicles.