Is Berlin too chaotic for self-driving cars? – Roadshow

Riding in a modified Jeep Grand Cherokee down the Bundesstraße through Berlin’s Tiergarten, the driver takes his hands off the wheel and the car does a reasonable job of maintaining its lane position. Joern Ihlenburg, engineering manager at Magna, the automotive equipment supplier that turned this Grand Cherokee into a self-driving car, points out that the lanes in Berlin are substantially narrower than those in the US.

The lane width makes this self-driving test a real challenge, exacerbated by the Grand Cherokee, which is wider than typical European cars.

The test becomes untenable, however, when we approach the traffic circle around the Victory column, the 19th-century landmark in the center of Berlin’s Tiergarten. The Grand Cherokee’s driver takes hold of the steering wheel as we approach the Victory column, and Ihlenburg, sitting in the back seat with me, explains that this first generation of Magna’s Max4 Autonomous Driving Platform isn’t up to handling this situation.

I look at the chiaroscuro of mixed lines on the pavement, with five major roads entering and exiting the traffic circle’s five lanes. It’s a mess, so I get why the car’s computer can’t handle it. Human drivers can barely cope.

Ihlenburg remains confident. “The hardware is up to the task,” he says, “it now comes down to the software.”

The hardware making up Magna’s Max4 platform consists of the typical self-driving car loadout of cameras, radar and lidar, this last sensor type using a laser to judge distances between it and any solid object in the vicinity. Magna integrated all these sensors into the Grand Cherokee, with a stereoscopic camera in front of the rearview mirror on the windshield, while radar and lidar are camouflaged in the bodywork.

Under the cargo floor of the Grand Cherokee sits a computer that processes all the incoming sensor data, commonly called sensor fusion, then outputs driving decisions.

With this test car, Magna is working toward what’s known as Level 4 self-driving in engineering circles. That means the car can handle all driving tasks by itself, but still has a steering wheel, brakes and an accelerator so that a human can choose to drive.

Magna, one of the larger automotive equipment suppliers in the world, intends to sell the Max4 system to its automaker clients. The projected date most automakers give for self-driving cars is 2021, a daunting timeline for Magna.

For the Victory column traffic circle, the car’s camera will need to follow the lane lines, distinguishing the ones defining its path of travel from others that cross over from the incoming roads. At the same time, its radar and lidar need to detect the cars entering and leaving the circle to avoid collisions. With all this information, the computer should be able to tell the car how to steer around the circle, and make lane changes to take its exit.

Ihlenburg believes his team at Magna can accomplish this programming, although it represents a major challenge.

As the car drives itself in the left lane of the Bundesstraße, it taps the brakes once or twice when its sensors pick up some large billboards on the median strip. That’s just another thing Ihlenburg will have to iron out.

As we discuss programming cars to drive themselves, Ihlenburg says that, instead of strictly following the rules, cars will have to drive more like humans to be practical. Mostly, that means edging around vehicles that might be stopped in an intersection for a left turn, or paused at a curb but still partially in a lane.

Magna has just begun developing its self-driving software, which puts it behind company’s such as Waymo, the Google self-driving offshoot. But solving for the kinds of traffic problems Berlin represents will give it a good start for many developed countries.