The reality is that sci-fi KITT is almost here.

enter image description here

Recently I read articles Delphi car driving itself across the country as well as Tesla cars will be self-driving this summer. As of 2013 four states (Nevada, Florida, California, and Michigan) have enacted some type of legislation to permit autonomous vehicles. This indicates legislators recognizing the emerging technology.

enter image description here

Also these autonomous vehicle have many sensors to gather data real-time which are processed by complex software algorithm. Radar technology is one such used in front and rear cross traffic alert systems.

enter image description here

Among the many potential advantages, few traffic collision is one of them. Likewise there are many obstacles too. Among them are disclaimer such as the one by Toyota "Do not rely exclusively on rear cross traffic alert or blind spot monitoring systems. These are driver assist devices and are not a substitute for safe and attentive driving "


Is the reliability of systems such as rear cross traffic alert system, in a self-driving car the primary obstacle impeding the success of self-driving car with human as fall back option?


  • 2
    $\begingroup$ With all of the autonomous vehicles out there, I think all of the technical issues have been solved. The key remaining issues are political. For example, what about liability? Who is responsible when a self-driving car kills someone? $\endgroup$
    – Dave Tweed
    Mar 22 '15 at 16:39
  • $\begingroup$ Adecuate sensors for driving in heavy rain and compensation for slippery (e.g. Icy) roads comes to mind from a recent article i read... I wish i could provide the citation, but I seem to have misplaced it. $\endgroup$
    – Paul
    Mar 22 '15 at 20:18
  • $\begingroup$ @DaveTweed: There are still a lot of technical issues to be resolved. $\endgroup$ Mar 24 '15 at 22:16
  • $\begingroup$ @DaveTweed Isn't that just a matter of insurance? Suppose I lease a self-driving car from Google, it could be a three-way contract including me, Google, and insurance company. Whether Google pays the insurance company and includes the price in the fee, or whether I pay the insurance company directly, doesn't really matter all that much. If the insurance is one with sufficiently high fees to not have any copay, I don't really see why liability is such a big problem. $\endgroup$
    – gerrit
    Mar 17 '16 at 12:59
  • $\begingroup$ @gerrit: No, insurance is just one way of financing liability, by pooling resources. It does not address the fundamental question of who is liable in a given situation. $\endgroup$
    – Dave Tweed
    Mar 17 '16 at 13:12

It's hard to know what all of the remaining technical obstacles are. Most of the obvious technical challenges have been solved already for a single self-driving car in amongst human-driven cars. If self-driving cars were to become normal, this raises some possible unknown issues:

  • What would be the effect of every car using LIDAR, RADAR, SONAR to understand their surroundings?
  • Would those 'active' direction and range sensors cause problems/interference with other cars or other equipment when densely packed together on busy streets?
  • Would co-ordination between vehicles become required/desired to prevent undue hesitation or incorrect action when vehicles see other vehicles from different vendors (with different rules and assumptions about traffic behaviour)?
  • Once pervasive (and therefore a worthwhile target), could the 'active' sensors be confused, interfered with, or possibly spoofed with malicious intent (insurance scams, causing collisions, murder, terrorism, etc.)?

My personal opinion is that as self-driving cars become more widespread, we will need to use 'passive' sensors (i.e. cameras) a lot more and infer the environment using image processing (computer vision) of visible light and other (non-visible) electromagnetic radiation where appropriate. While it may be possible to filter out interference, prevent interference with other (non-cars) equipment, protect against deliberate signal manipulation, etc. I think that computer vision will ultimately be the simplest and most reliable way for the car's computer to 'see' and understand its environment.

The technical advances needed in this area are mostly related to parralelisation of processing multiple inputs in a timely fashion. Standardisation of output from processed images (such as stereoscopic depth analysis, material/texture analysis, object recognition and tracking, etc.) will allow for more modular and parallelised systems as well as interoperability with off-the-shelf components of a larger system. Using more than two cameras for radial depth perception rather than the standard two-camera stereoscopic analysis. These are some of the advances needed, if computer vision will end up being the primary input for self-driving cars.

  • $\begingroup$ I wonder what will happen when 4 self-driving cars all meet at a roundabout simultaneously, and they all have to give way to each other. With human drivers, one of them will take the initiative and enter the roundabout, but with only software, could they all end up waiting for each other and hence waiting forever? $\endgroup$
    – AndyT
    Mar 24 '15 at 22:42
  • 1
    $\begingroup$ Heh, yeah deadlock does seem possible. The algorithms would need ways around deadlock. One simple solution is to wait a random amount of time then go for it unless someone else already has. This is like the solution used in the case of packet-collisions in some computer networks. $\endgroup$
    – jhabbott
    Mar 25 '15 at 0:29
  • $\begingroup$ Yesterday in The Netherlands, a group of 50 self-driving cars were reportedly driving simultaneously on the motorway, although from a closer look those do not appear to have been fully autonomous. $\endgroup$
    – gerrit
    Mar 17 '16 at 13:02
  • $\begingroup$ @AndyT a central traffic management network that gives commands to self-driving cars would solve that. which i feel is really whats needed to make masses of self-driving cars work perfectly. $\endgroup$ Mar 18 '16 at 16:42

Current sensor and computing technology permits a vehicle to be operated statistically safer by a computer than by a human.

"Statistically" is the key word here. Every day there are situations where someone will die in a car accident regardless of the decision making involved. Neither humans nor computers can see the future. The difference between the two is the human is socially permitted to make mistakes while a computer is not. If we truly want safety we should be building track based transportation systems where physics, not computers, ensure that collisions do not occur.

Your Toyota reference for example stated "Do not rely exclusively on rear cross traffic alert or blind spot monitoring systems." because the driver is the permissible liability; the scapegoat so to speak. No company is going to take responsibility where it doesn't have to; especially when a human still has partial control of the system.

Liability will be the largest obstacle to autonomous vehicles. 92 people died every day from car crashes in the US in 2012, but the blame was easily socially distributed to those individuals involved. If google or some other large corporation sells an autonomous car to everyone in the US and it only killed 1 person per day, people would be outraged, sue, and shut the whole operation down.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.