Tuesday, May 13, 2014

Commuting to work

For most of my life I have been lucky to work just 3 km from home, so I have not been exposed too much to the commuting woes. For example, when I arrived in the Silicon Valley, the 101 freeway had two lanes in each direction separated by a wide median strip planted with oleanders. Over the years, the median strip has disappeared and 101 became a freeway with four crowded lanes in each direction. For the last two or three years, a fifth auxiliary lane is being added in each direction in the portion between Marsh Road (Facebook) and 85 (Google, LinkedIn, Microsoft), so I have heard a lot of yammer from my coworkers.

For the past year I have been a commuter myself, barreling down 23 km to the San Jose airport every day. Unfortunately there is no usable public transportation, so I am condemned to this daily freeway maltreatment. It starts after 2 km, when I enter 101 on Embarcadero, where about 20% of the drivers illegally cross two double lines at a 90º angle to force themselves in a passing lane before the actual freeway entrance, while pushy Gbusses force themselves from the high occupancy vehicle lane to the exit lane and a few 100 m later a slew of cars try to make a –90º turn from the leftmost lane to the San Antonio Road exit.

During the past year I have tried to develop a driving model that would reduce my stress, but not very successfully. Last Saturday finally was able to see a sophisticated model in action and it was an eye-opener: I got to ride a Google Car from the Googleplex down 101 to the 280 interchange and back.

Sitting behind the "driver" I had a good view of the laptop on the lap of the lady in the front passenger seat, displaying the car's model of the surroundings based on the lidar spinning on top of the car and also a radar in the front of the car, an inertial sensor in the rear wheel axis and last but not least on countless hours of tweaking the model based on the feedback of skilled professional drivers like Anja—our pilot on this trip—who rides full-time for her work.

On the console we see the freeway lanes, our projected route, and the surrounding vehicles. When a vehicle creates a dangerous situation, it is marked with a danger sign. The model recognizes the lights of emergency vehicles and can pull over according to the law. However, it ignores other car's blinkers. In fact, the American driving culture is that the other drivers are your enemies and you do not want to warn them by letting your intentions to be known: the blinker is either never turned on or left blinking.

While as a human I can model a few cars around me, Google's algorithm can model many cars around our self-driving car, in all directions. When our car gets in the blind spot of another car, the icon of that car is flagged with a danger sign. With a surprising frequency, the flagged cars cut us off at a dangerously close distance. Since I am not driving, I can look in the offending cars and can never see those drivers turning their heads to check the clearance. Therefore, they are all driving erratically without looking, resulting in the car being cut off, breaking and propagating this backwards to the following cars.

Like computers can beat humans at chess because they can predict a larger number of steps, Google's car is better than human drivers because it can by far model more vehicles than a human can. Yet, humans are too stupid and reckless for Google's algorithm to be completely foolproof. For example, at one point in Santa Clara we were in the right lane and a big truck tried to pass us driving above the speed limit and on the shoulder. Our pilot Anja recognized, maybe from the truck's exhaust fumes, that he did not have enough torque to pass us and the shoulder turned into a ditch a few meters further ahead. This would have left the truck driver to either go full speed into the ditch or ramming us, so she floored our brakes.

Those reckless drivers are in part professional drivers who spend their working day on the freeway driving trucks, taxis, limos, etc. This indicates that most humans are unfit to drive cars.

But are driver-less cars the answer? When I was a teenager, I thought that by 2014 I could fly to the moon with TWA or PanAm and get to Paris in a couple of hours on a Trans Europ Express (TEE). It would never have crossed my mind that in 2014 I would be driving a car on a freeway full of incompetent erratic drivers.

The mistake being made by the Caltrans agency is to build those auxiliary lanes. Instead, they should have built a train like the S-Bahn on that old median strip. A skilled professional train driver could get me to work in a few minutes, safely and without stress.