REGISTER | LOGIN
Breaking News
News & Analysis

Autopilot: Flying vs. Driving

Can lessons learned from aviation apply to cars?
3/25/2016 07:14 AM EDT
12 comments
NO RATINGS
1 saves
Page 1 / 4 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
SteveHarris0
User Rank
Freelancer
Cars On Autopilot
SteveHarris0   4/4/2016 1:10:50 AM
It seems to me that creating a failproof autonomous car for use on public roads is a dramatically harder problem to solve, than autopilot for airplanes. Planes don't have to worry about staying on the "road", as long as they are at normal flying altitude there are no obstacles to avoid, no lane to follow, hardly any flying objects to avoid. There is a thing called Big Sky Theory, it goes something like this. Not only do you have completely open space in 2 dimensions, you also have a third dimension, altitude, which makes it incredibly less likely that two randomly flying objects will collide. Hardly any collision avoidance effort is required for the plane, perhaps some simple radar based avoidance algorithm. Cars, on public streets with traffic, is probably some two or three orders of magnitude more difficult of a problem to solve. Perhaps not solvable in my lifetime, to reach the point where a significant portion of traffic is autonomous. Too complex and unpredictable. The demand for liability attorneys would skyrocket, every accident the burden of proof will be on the autopilot to prove that it didn't mess up. I could see a few special routes designated for autopilot cars, but beyond that I don't see it catching on for many decades.

Bit Wrangler
User Rank
Author
Two thoughts
Bit Wrangler   3/29/2016 1:57:59 PM
First thought:  the graph showing a decline in accidents is, in my opinion, only slightly related to increased automation.  Instead, the increased safety is due to careful investigation by the NTSB of each crash, and implementation of their recommendations in a wide variety of areas over the course of many years.  Automation might even be an insignificant factor in the decline of aviation accidents.  One would have to study the NTSB accident reports over the same time period and determine how many early accidents were addressed by increased automation.

Second thought:  as the AF447 crash demonstrated, increased automation caused the first officer on that flight to not even consider that he might have stalled the aircraft.  The Airbus flight computers dropped back to their Alternate Flight Law, which removed stall protection, but the pilots had been trained by repeated usage to expect this protection.  When the computers could no longer provide it, they were not sufficiently proficient in hand flying the aircraft the "old fashioned" way to be able to recover from the system failure.  If your automation quits when it can no longer handle the situation, you had better have a pilot (or driver) who is trained and proficient enough to manuever your vehicle safely through the emergency.  The only saving grace in the case of automated cars is that there is usually a safe place to pull over and stop, an option that is not available to an aircraft.

junko.yoshida
User Rank
Author
Re: lessons learned from Aviation, but a different problem
junko.yoshida   3/28/2016 6:00:54 PM
NO RATINGS
I agree. There are obviously a set of very different problems car OEMs need to deal with in designing autopilot.

However, there are also real problems discovered by the aviation industry that are applicable to drivers of autonomous cars.

I find it fascinating that as scientists concluded, even trained pilots also face, on autopilot, "boredom and distraction, mode confusion, recovery from automation errors, skills degradation, and trust issues are major concerns."

RadioGraybeard
User Rank
Manager
Re: POOTLEs
RadioGraybeard   3/28/2016 4:24:20 PM
NO RATINGS
An aircraft autopilot never has to deal with another aircraft sitting there on a cloud and suddenly darting into its path, like a car driver may have to with another car or kid on a bike or dog or....  It never has to deal with unexpected obstacles in its path. 

Comparatively, I think the aircraft case is easier to design for.  Commercial aircraft fly rather predictable paths: during cruise they're horizontal at a controlled speed.  They don't suddenly climb or descend steeply, they don't suddenly stop, they don't suddenly bank hard and change directions.  The aircraft already talk to each other (TCAS, ADS-B).  Even then the aircraft sometimes gets really lost and says, "I don't know what to do - you take it" 

 

MWagner_MA
User Rank
Author
lessons learned from Aviation, but a different problem
MWagner_MA   3/28/2016 7:26:24 AM
NO RATINGS
All good comments and the right conversations going on.  As a pilot, my view is that the Toyota approach is the more realistic and will generate the best short term benefit while minimizing risk.  Keeping the driver in the loop, but greating assisting in "seeing" problems before split second reaction time is necessary is a huge step forward.  Car-to-Car communications, proximitity alerts (radar), road conditions can help a driver make better decisions while improving the technology and robustness as we work towards higher levels of automation.

perl_geek
User Rank
Author
POOTLEs
perl_geek   3/26/2016 2:50:29 PM
The development of automation in flying has led to a whole new class of accidents, (including some of those referred in the article). They haven't been assigned an official acronym yet, so I suggest POOTLE, for Pilot Out Of The Loop Event.

There is a pilot saying that "The aeroplane should never go anywhere your mind hasn't  been at least 5 minutes earlier".  In other words, stay ahead of the machine and what it's doing.

An autopilot gets into a situation it cannot handle, perhaps because some sensor is U/S, or giving confusing results. It relinquishes control to the pilot, who wasn't expecting to receive it, and does not have a current mental model of the flight. (Altitude, attitude, position, heading, power setting, and speed,) Usually, things become terminally unpleasant before the meatsack can work out why the robot's resigned. 

DonaldB211
User Rank
Rookie
Re: Cars can simply stop
DonaldB211   3/25/2016 8:36:05 PM
NO RATINGS
I would argue that autonomous driving actually has far less opportunity to relinquish control gracefully.

 

An autopilot just needs to set the control surfaces to a near-neutral position and alert the pilot.  There is often time for the pilot to wake, get some coffee, check the morning news and then deal with the problem.


In a road vehicle the system often needs to make a split-second decision.  What should I do about the car that suddenly appears facing me only 40' away when travelling at 65MPH in the middle lane?  Pulling over and stopping is not an option.  There might be several lanes of bumper-to-bumper traffic, and no break-down lane.  And the other drivers certainly won't be expecting a panic reaction just from following a car carrier, rear-towed car, or a mirror-back tanker truck.

 

 

Devashish Paul
User Rank
Author
Mobile Edge Computing will assist
Devashish Paul   3/25/2016 8:13:14 PM
NO RATINGS
You're not getting it. The vehicles will be autonomous.  Networked mobile edge computing will assist them with additional info.  Even driver driven vehicles can/will be assisted by mobile edge computing

elizabethsimon
User Rank
Author
Re: Networks Vehicles
elizabethsimon   3/25/2016 8:10:00 PM
NO RATINGS
As long as the vehicles continue to work when they get beyond range of 4G or 5G. There are areas in the USA where cell coverage is minimal to nonexistant. Due to low population density, it's likely to be a while before the infrastructure exists.

Devashish Paul
User Rank
Author
Networks Vehicles
Devashish Paul   3/25/2016 7:06:29 PM
NO RATINGS
Junko, good article. With networked vehicles using Mobile Edge Computing some of the unpredictablility related to road surface, congestion, weather, and other things that may changed from the last drive through etc, computing can restore predictability in an unpredictable world.  This won't need very expensive infrastructure, just low latency servers/computing nodes running the networks apps that the vehicles communicate to via 4G or 5G. We're exploring some of this at 5G Lab Germany: http://www.prnewswire.com/news-releases/idt-and-5g-lab-germany-collaborate-on-technology-to-enable-network-connected-autonomous-vehicles-300224920.html

 

Page 1 / 2   >   >>
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed