Google recognizes 272 is a serious deficiency in the testing of its autonomous car (but draws a positive conclusion from them)

Yesterday, Barack Obama was headed to their citizens to deliver his State of the Union, a speech in which he would speak of the challenges facing the Western world, from security and terrorism, to climate change, through the importance of science and technology development. Although time spent tip-toeing about this issue, Obama, he rebukes to the u.s. to work in “the transportation of the XXI century”. The 44th president of the united States was concerned, among other things, to a technology which these days is in the mouth of all, the autonomous car. In fact, it is expected that this week the united States increase, with either laws, or subsidies, institutional support to the development of the technology. Meanwhile, Google recognized 272 serious errors in its autonomous car. Errors that required that the technician in charge to take the controls immediately. Errors for Google are not a jug of cold water for the autonomous car, but a valuable experience to ensure that in a few years the autonomous car arrives at the dealers, and it is safe.

autonomous cars that are being tested in California are registering all the faults suffered by the technology, and all its accidents, to provide a periodic report to the authorities.

what were those errors? The failure of an autonomous car arrives at the time that the vehicle suffers a disconnection unplanned stand-alone mode. That is to say, the moment at which the autonomous car requires the “driver” to take the controls having detected a fault in the technology, or when the situation requires that the “driver” to take over the controls of the car to avoid a situation of danger, or even an accident.

Recall that the legislation of the various states require that after these autonomous cars are experimental, that are circulating already in open road, is a technician monitoring at all times the march. The Department of Motor Vehicles (DMV) of California also requires that each manufacturer, such as Google, send a report with the accidents and mistakes suffered by their autonomous cars.

google-car-coche-autonomo-errores-03

These mistakes they demanded that the technician monitors the movement of the autonomous vehicle to take the controls immediately. The technicians of Google, who have been specially trained to do this, they were given a reaction time average of 0.84 seconds.

Thanks to these reports, we know that since 2009 Google cars have traveled nearly 4 billion miles, more than half of which have elapsed in stand-alone mode. In all these years, and miles, the autonomous cars of Google have experienced 17 minor accidents, none of which was guilty of the autonomous car.

why is it so important for Google that these errors occur?

In these moments the most important thing is not to convince the other drivers that autonomous cars are safe, but to get that you really are safe, that their reliability is close to 100%, and which may be circulated in our streets without the need of a technician, or a driver, to monitor their progress to take the controls immediately in case you experience a problem. For Google, therefore, is not a problem that its autonomous car will suffer an automatic disconnect or forced, and their technicians have to take the controls of the car. Their technicians have been trained specifically to maintain the concentration and be able to take the controls immediately when a succession of these incidents. According to Google, the time of reaction half of their technicians in the 272 incidents recorded would have been 0.84 seconds. Many of the incidents would not have meant a collision in the case that the driver had not taken the controls, and in any case would have occurred an accident as a result of these errors.

To Google each incident is an opportunity to get to know the aspects which have to be improved in your autonomous car to ensure that reliability close to 100%.

google-car-coche-autonomo-errores-01Google has sophisticated computer simulation systems that can replicate the conditions that led to one of their cars suffer a problem that required the intervention of the driver.

When there is one of these rulings, in addition to a driver to take the controls of the car immediately, starts a protocol analysis to determine what has happened, and how it can be solved. The car of Google records thousands of parameters, and their technicians are equipped with advanced computer systems that are able to create simulations which can recreate, with all luxury of details, how the incident occurred. These simulations are not only able to recreate the environment, to get the autonomous car of Google is in the same situation that led to the error, and put the rest of participants in the scene (other cars, pedestrians…), but also to get the reaction of the autonomous vehicle in simulation is exactly the same as it had in the failure real.

Google mentions an interesting trend, that is an error, and the following, are increasingly separated in time, or better said, the distance traveled by their autonomous cars. In 2014, for example, experienced an error of this kind every 4,000 miles or less. In 2015, the autonomous cars of Google experienced an average of an error every 8,000 kilometres (first and fourth quarter of the year), or even an error each more than 13,000 kilometers (in the second quarter of the year).

In any case, Google continues to believe that the important thing is not that these errors are each time more distant in time, but these errors still occur, and are treated according to its protocol, to design an autonomous vehicle more advanced and reliable.

google-car-coche-autonomo-errores-04

why these errors occur, these disconnections of the autonomous car?

The majority of the errors have been produced by a discrepancy in the perception of the vehicle. That is to say, an autonomous car detects conflicting messages in the signaling and, when in doubt, forces the driver to take the controls. Sometimes, these discrepancies generate an error in the software, or the hardware. Many of the errors have occurred at the time that the technician has taken the controls to the realization that the autonomous vehicle was performing a manoeuvre that was not appropriate.

finally, we also have not forgotten that the majority of errors are occurring on urban roads, and not on interstate highways or freeways, in which the reliability of the autonomous car of Google is very high. In any case, it is also true that Google recognizes that its autonomous cars are making a significantly greater number of miles in urban environments, on roads, in both cities, which are benefiting from increased traffic, intersections, lane changes, etc., represent a greater challenge to the development of the driving autonomously.

Exciting, don’t you think?

Source: Google
In motor: