On the road with Mercedes’ autonomous test car

ABOUT six hours into our road trip from Sydney to Canberra via Wollongong, riding shotgun in an S-Class development car set up to demonstrate advanced autonomous driving features, Mercedes-Benz’s driver-assistance tech chief Jochen Haab hits the brakes hard to ensure we avoid rear-ending the vehicle in front.

There was every reason to expect that the specially modified S560 would have stopped by itself in time.

But with traffic parked at a red light just after a corner, and with us rounding that bend at a decent speed, the wide angles involved meant the radar did not have a clear view of what was up ahead and so the car was slow to react.

With the boot stuffed full of computer hardware, and cameras positioned inside and around the car, Mr Haab logs the incident (via steering wheel controls and a microphone) for review at a later date.

This will feed into a global program that has covered more than 9.5 million kilometres since 2010 – with 175 test vehicles taking 1.2 million measurements, logging 430,000 ‘events’ and analysing more than 200,000 of these in detail – and is one of many “trigger points” in our eventful trip on the Australian leg of Mercedes’ so-called Intelligent World Drive.

It is also just the sort of thing that Mr Haab and his team seem to thrive on because it adds to, or reaffirms, the data that is building up and guiding development of next-generation software and intelligent systems as the German manufacturer moves into higher levels of vehicle automation.

“Once you know it, the solution is almost there – it’s just knowing it,” Mr Haab explained. “If you know what’s happening, you can do something about it.

It’s like in real life. It’s all about knowing – ‘doing’ is the easier part.”

This applies to various situations involving camera- and map-based faults – usually to do with other cars and speed limit recognition – that we encountered on our drive, some of which can be found in virtually any country while others are unique to Australia.

The S-Class engineering car, which is set up for the current Level 2 (partial automation) but with modifications that allow demonstrations of what Level 3 (high automation) feels like, was impressive in the way it tracked cyclists and various other road hazards via camera and radar, keeping a safe distance.

Down the road a little further, however, when exiting a narrow bridge on a winding stretch of road, a 10km/h speed sign was apparently interpreted as 110km/h and, well, you can imagine that the car’s response triggered a fresh incident report.

This appeared to be another wide-angle issue, with perhaps a shadow next to the ‘10’ which make it look like 110. In any case, the car’s misinterpretation was a clear demonstration that this situation requires further analysis.

Our weekend drive meant variable school zone speed limits – a known issue – were not put to the test, but temporary roadwork limits were not always recognised and an exit lane’s speed limit sign from a freeway was mistakenly picked up, slowing the car down to 80km/h in the 110km/h zone.

At a roundabout or stop sign, the automated driving systems could identify the road layout and apply the brakes accordingly, but still relied on the driver to give way.

Back in heavier, multiple-lane traffic, another ‘false positive’ was logged when a car in front slipped into a right-hand turning lane and the S-Class was still under the impression that it was directly ahead of us, jumping on the anchors unnecessarily.

“The problem is that the car does not have intuition – yet,” Mr Haab said after this event.

“We have to teach the car intuition. You will know when the car goes, but what if, for whatever reason, it just slams the brakes, then you as a driver are in trouble. The automated car has to account for that, be prepared for that, and the only way at the moment is to slow down and pass it very slowly.”

All these situations underscore the challenges Mercedes and others in the business of building highly intelligent, self-driving cars have ahead of them before graduating, in any significant way, to the next level of automation – Level 3 – which takes a lot of responsibility away from the driver and puts it into the hands of the vehicle itself and its constructor.

But it’s still a two-way street, or what Mercedes calls “conditional automation”. It’s not all down to the car manufacturer while the driver watches TV or reads a broadsheet paper on the commute home, as images depicting autonomous cars often suggest.

As well as ensuring the systems work flawlessly, Mr Haab and his team are working through issues such as how to ensure the driver of a Level 3 car understands when he or she is expected to take back control of the vehicle when the situation demands it.

These include developing a ‘dead man’s switch’ – whether by contact on the steering wheel, or detecting eyes placed on the road, for example – to ensure the driver is alert enough to take control of a situation that the car cannot handle.

If there is no response, the vehicle will go into emergency mode, pull over and ring emergency services.

“We give you as much support as we feel is safe and we can take the responsibility,” he said. “And if not, we tell, you, ‘Okay, sorry, you have to be alert – we can still do the steering for you, but just stay in the loop. Don’t look at your movie … But we can still help you steer.’“It would be extremely wrong and extremely arrogant to say we can deal with anything. You have to come up with solution strategies – how to deal with the unexpected.

“On the highway, we can (already) do Level 3. I can drive as if it was Level 3.

But what if one of those seagulls came and shit on my camera?” Or a rock knocks out a radar sensor? Or… “You have to have a strategy to deal with that.”