Traffic Cop or Inflatable Wacky Waving Tube Man – Autonomous Vehicle has no idea.
With autonomous vehicles, also known as self-driving cars, it seems that we could be on the threshold of a cleaner and safer highway system, but are the obstacles too much to overcome?
As we’ve discussed, there are plenty of questions regarding the implementation of autonomous vehicles, such as “Who’s at fault in case of an accident?” or “Will autonomous vehicles share the road with human drivers?” for which regulators are still working out the answers. Additionally, there are limitations, since even the best computers in the world fall far short of the human brain’s capacity to think.
In general, autonomous vehicles use satellite positioning for a rough idea of their location, then fine-tune their position using maps and sophisticated camera and ranging systems.
The onboard computers need to constantly distinguish between pedestrians and Inflatable Wacky Waving Tube Man, for example, or between a rock and an errant fast-food bag. Additionally, bad weather can be a big problem for autonomous vehicles.
Rain interferes with camera and ranging equipment, and snow can obscure road markings, which the car needs to navigate. Add in construction zones, road changes, deer crossings, and stick-ball, and the results range from simple delay to catastrophe.
The problem is that autonomous vehicle computers still fail miserably when it comes to reproducing human brain activity. Even the fastest supercomputers in the world are hopelessly outmatched by the human brain.
The K Computer, in Japan, is the world’s fourth-fastest supercomputer, capable of 8.192 petaflops (8,192 quadrillion operations per second), and was recently used to model human brain activity. Using 705,024 processor cores, 1.4 million GB RAM, and approximately 9.9 MW power, K Computer was able to model one percent of the human brain for one second.
It took forty minutes to accomplish the task. Autonomous vehicles don’t carry nearly the same computing power, so it’s a long road ahead before they’ll be ready to tackle the everyday driving situations that human beings have been managing for over a century.
Granted, we’re not expecting autonomous vehicles to write a sonnet or contemplate the universe, but can they successfully navigate in an ever-changing and uncertain traffic world?
The folly of this article… is assuming that autonomous cars will be (or should be) engineered to behave like humans… with limited visual sensors in a narrow spectrum, and complex pattern recognition (“reproducing human brain activity”).
Humans cause 30,000 deaths each year in the US alone from driving. If this is the measure… I think we should avoid engineering autonomous vehicles to behave as humans.
Instead, lets build them to play to the advantages of computers… which are far more suitable to driving than humans.
The only reason they need to be engineered as such is because they have to interact with us. Pedestrians and other drivers are a HUGE problem that can’t be solved by algorithm short of AI. IMHO, the ONLY way to successfully implement autonomous drive is to get rid of all the human drivers.
That may be an eventuality, but algorithms have no problem detecting the position and motion of objects regardless of who or what is controlling them
Green Pessimist!
Change your headline. There is nothing in the article that suggest that autonomous vehicles “ARE HAVING” problems with weather.
Just speculation.
Lidar and subsonic sensors are not confused about the difference between a pedestrian in the road, and a wacky waving inflatable tube man on the side of the road who the computer reads as having no real substance, like a flag waving in the wind.
Lidar and subsonic sensors do BETTER than humans for seeing through rain, snow and fog… and obscured lane markings are easier to spot because of the difference in density between the freshly driven snow/slush, and the surrounding untouched snowfall.
Yes, but their ability to come to a snap decision that is both effective and efficient, yet results in no loss of human life, is still lacking. The Google Car would be paralyzed in any modern city with human drivers. Currently I’m in Lima, some of the worst traffic in the world. Could an autonomous vehicle eliminate the dozen or so deaths I’ve read about in the last week here? Sure! I think the ONLY way it could do it would be if EVERY SINGLE VEHICLE was replaced with autonomous vehicles AT THE SAME TIME. Just try to convince every single ignorant dumbass driver here that it’s a necessary change.
I’ve driven in Lima as well as other South American countries in Central America too.
Yes they pose unique challenges because human drivers do not even obey the simplest of traffic laws.. such as staying in lanes.
however, the mistake is thinking that it needs to eliminate 100% of all fatalities.
it does not, it just needs to reduce them enough to be significant
no, not the point that it has to eliminate fatalities, but it has to be able to respond without being paralyzed by the indecision of others. by design, autonomous vehicles are not “aggressive,” but perhaps the epitome of “defensive” driving. for now, i can imagine that open highways and US-style congestion can be handled by current designs, but inner-city insanity requires a different algorithm.
According to Google, they’re STILL working out the kinks when it comes to driving in snow and rain, which is almost a non-issue in Silicon Valley, but what about the rest of the country? Here’s an interesting article, covering gaps in Google Car’s programming, such as its inability to discern a traffic cop’s motions or the difference between a fast-food bag and a boulder, or even the ability to avoid an open manhole. There’s a long way to go… http://www.technologyreview.com/news/530276/hidden-obstacles-for-googles-self-driving-cars/