Thursday, August 23, 2012

Hacking the Automated Car


I was pleased to come across a site recently from a group called the Center for Automotive Embedded Systems Security (CAESS).  The Center is a collaboration between the University of Washington, the University of California San Diego, and the US National Sciences Foundation.  It deals with, as the name suggests, making computers in automated driving systems secure.

The computer security of automated cars and automated driving systems is something I have given thought to over the past 15 to 20 years, more as the capabilities of our computers have increased.  There are certain things that can be done to reduce the opportunity to hack automated driving systems.. The one that springs to mind is to hardwire as many functions as possible so that they cannot be hacked.  Hardwire refers to making the physical design of the chip control the program that runs on it.  Such a chip, once made, cannot be reprogrammed, only replaced.  Obviously there are disadvantages to such an approach, but in a systems where safety needs to be paramount, we may need to live with those disadvantages.

Heavy security needs to be built in from the bottom up.  And frankly, what can already be done to hack the rudimentary systems in cars is scarey enough.  Stay tuned for more.

Wednesday, August 15, 2012

Google Cars Get Ready for Prime Time?

Last week a report on TechCrunch, and a post on the Google Blog indicates that Google now believes that their automated cars are safe enough for a single person to be in the car while the car drives the person to and from work and, it is implied, other tasks.

Politics in Florida

Not being an American my opinion doesn't count for much in their politics, however, some things really burn me.  There is a commercial against Jeff Brandes (R-St. Petersburg).  Unfortunately the commercial takes aim at him through his support of automated driving systems, using all the usual scare tactics.



The Democrats (presumably) should be ashamed of themselves.

A Response to "Learning to Let the Car Drive"

I recently read an article on Wired Autopia, Learning to Let the Car Drive about how we would slowly increase the amount of automation in the driving process, and there would be a point when driver assistance would be needed sometimes, but mostly not, and how it would be necessary to keep drivers focused on the task of driving.

I find three of  the assumptions problematic - 1) that drivers will need to stay on the driving task when the car does most of the work, and 2) that cars will stay in their lanes through surface markers on the road  3) avoidance of side hazards, eg deer or children

Regarding the first assumption,  when the car is capable of driving itself in most conditions, say 65% or more of the time, I strongly suspect that we will need to make cars that leap over that last 35% completely.  Think about human nature.  If you don't have to pay attention to something most of the time, then you don't pay attention to it at all.  How will the car companies force people to pay attention to the driving task when it is mostly handled by the car?  We have enough problems making people focus on driving and not texting or not using cell phones now, and we drive manually.  Quite frankly, once we are capable of making cars that handle most situations, we will have to start making cars that handle all situations, at least as well as a human, in very short order.  Even with some sort of alert system, it will require the "driver" to:
  1. shift between the task they are doing, and the hazard that the system alerts them to, 
  2. locate the hazard
  3. evaluate the hazard, and then 
  4. respond appropriately.
Even if each of the above steps takes only one second, which seems to me exceedingly optimistic, then at least four seconds is required before the car begins to do something under driver control.  Most likely it will take much longer.  We would all be better off letting the car drive and react which are at speeds beyond human ability.

The second assumption seems to be that surface road markers, e.g. paint, will be used by the vehicle for way finding and location maintenance purposes.  With all due respect, evidently road conditions of heavy or blowing snow seem not to have been considered, or for that matter the highly reflective road surfaces of wet roads at night, or immediately after a heavy rain when the sun shines on the road surface.  In order for an automated driving system to work effectively we will need to install some sort of edge marking, whether for lanes or for road edges, that is not dependent on the ability of the markers to be visually observed by cameras (maybe magnetic pucks embedded in the pavement).  This is not to say that cameras as an edge finding device should not be on vehicles, but they cannot be the only way a vehicle will maintain its correct position on the road.

The third assumption is that humans will see side hazards sooner, and react more appropriately.  Humans may be able to evaluate whether or not something is a side hazard to a vehicle more appropriately than a computer, but humans could never respond faster when the hazard appears in the driving space.  In some cases, think a child dashing from between parked cars, then even a human cannot respond quickly enough, whereas an automated braking might.  Additionally, humans tend to speed, and a response to a perceived side hazard is to slow down to the speed limit.  An automated system would have the following advantages:
  1. it would already be doing the speed limit, so slowing down would be to below the speed limit and,
  2. an automated system would react faster than a human once the threat was perceived
So I find it problematic that humans are better equipped to handle side hazards, especially given the faster reaction times of computer systems.

It would be interesting to know exactly how drivers will be kept on task when the car does most of the driving.