Autonomous vehicles – how skeptical should we be?

I have gone up and down on the prospects for  autonomous vehicles (AVs). There are a lot of technical hurdles, and probably as many social issues such as how liability laws will be written. The Google car has been over-hyped. But today I received a claim that AVs are not feasible until 5G wireless networks are ubiquitous.


 In the midst of all of the hype and hoopla over self driving cars, let’s pause for a reality check: None of the many rosy media discussions of the future of autonomous vehicles show any awareness that there are no credible network plans at all to support these vehicles – at scale  – in cities and in the countryside.  Meanwhile among the engineers, there is a growing consensus that  autonomous vehicles  will need  dense networks supporting gigabyte low latency streams for every vehicle. e.g. at least “300gBytes/ per month” of  *coordinated*  secure networks of LTE WiFi DSRC-V2V meshes , and satellites according to Andeas Mai, Director Smart Connected Vehicles at Cisco http://viodi.com/2015/06/15/300-gbytes-of-data-per-month-per-car/

And:

“It is very important that cars can communicate with each other and communicate with other participants in the city, such as pedestrians or cyclists. You have bring all of this together to have a smart city where cars can autonomously drive.”

 [Source: Dave Farber’s IP list. I don’t want to hold the original poster responsible for what I’m doing with his comments; you can check the archive for his original post.]

This is way-overblown. Real time communication would be nice, but it’s certainly not necessary.  For example it’s not the approach used by us humans today. Instead we use use a distributed model: local “sensors” and “computation” to detect potential collisions, and avert them. The corresponding AV approach will be putting good machine vision in each car.

If the networked approach is not necessary, is it superior? A  ubiquitous networked approach would in theory be able to anticipate a child darting out from behind a parked car- as long as the child and her dog are carrying  smart sensors and a radio.  (not likely!) The local vision approach could not detect them until they reach the street. But probably neither centralized nor autonomous  approach could handle a cyclist hitting a pothole and tipping over into the path of the car.
In any case, requiring AVs to show perfect collision avoidance is foolish. The  design goal for AVs should be something that is as good as humans in almost all situations, and modestly better than people in normal conditions.
Adopting autonomous cars is as much about societal norms and preferences as it is about technology.

By the way, the  aviation industry is currently transitioning from centralized control via the air traffic system and radars, to local control. This is called the Automatic dependent surveillance system. Aircraft signal each other via local radios. But the aviation world is a lot simpler than the auto world, at least for this purpose.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s