CES Put On a Very Technical Auto Show in Shanghai. Here's What Holds Up, and What Doesn't

Mostly, it's just autonomous driving

Shanghai has another auto show, and this one talks about technology.

The Shanghai Auto Show, held every other year, closed two months ago. As usual there were luxury cars and booth models, but nothing at the front edge of technology — nothing like the autonomous driving that internet companies and traditional carmakers both keep talking about.

The new "auto show" is CES Asia, an offshoot of CES, the world's largest consumer electronics show. The traditional consumer electronics market has few new highlights left, and at CES, held every January in Las Vegas, cars have grown more and more eye-catching.

The same goes for CES Asia in Shanghai. BMW and Baidu brought autonomous cars; Honda brought to China for the first time a motorcycle that balances itself and rides itself; Continental showed off hardware for autonomous driving; Nvidia, the graphics-card company that has made a loud entrance into the car market, didn't come in person, but its Drive PX 2 development platform sat in the trunk of Baidu's demo car.

These dozens of companies, large and small, all tied to cars, have together turned CES Asia into a small auto show where digital technology and intelligence stand out most. The new things you can't see at an ordinary auto show, you can find here.

Honda showed a motorcycle that follows you around, and a self-driving stool

Honda brought something unusual this time: a motorcycle that rides itself and balances itself. It still looks like an ordinary gasoline motorcycle — apart from two slender sets of struts on either side of the bike, you can't tell the difference.

During the demonstration, though, the motorcycle stood upright "by itself" in an odd posture, with no kickstand and no one holding it. With a rider on board, he didn't have to put a foot down to steady it, and it didn't fall over. When the demonstrator simply walked off, the motorcycle followed along behind him. Its wheels can change position: at higher speeds the gap between front and rear wheels widens, which helps with fast riding; when city traffic slows things down, the wheelbase shortens, which makes it easier to stay balanced standing still.

The bike keeps its balance by swinging the front wheel through small angles, much the way we ourselves balance a bicycle or a motorcycle at low speed. Honda says the angle between the front fork and the body can change with speed, so the bike balances itself at any speed.

This has a practical use. A motorcycle weighs more than a hundred kilos; it takes real effort to push one out of the garage, and one slip and it goes over.

The technology comes from the self-balancing system in the Uni-Cub, the two-wheeler Honda released earlier. The Uni-Cub goes back to Honda's in-house "Next-Generation Mobility Concept Contest" in 1989, where the company showed a single-ball vehicle that moved by rotating a ball.

In 2008, Honda built the first experimental "U3-X", but it was slow, uneven ground easily stopped it, and it couldn't turn in place. After more work it became the Uni-Cub.

Honda let visitors try the Uni-Cub at the show, and a QDaily reporter got on for a ride.

The Uni-Cub looks like a unicycle, but it actually has two wheels. Once you sit down and pull the lever behind the seat, the Uni-Cub, which had been leaning forward, rights itself on the ground and its two footrests unfold automatically; at that point it is already balancing itself.

You put your feet on the footrests and then lean your body slightly toward wherever you want to go; sensors inside pick up the shift in the upper body's center of gravity and move or turn that way. Sit back up straight and it stops. It runs on shifts in weight, the same principle as a self-balancing scooter.

The difference is that the Uni-Cub never accelerates suddenly, and it keeps itself level through a turn, so it won't tip over. The sensors are very sensitive — even a slight sway of the upper body while sitting still sets the Uni-Cub off and makes it spin slowly in place.

Apart from the lever behind the seat that switches it on and off, the Uni-Cub has no other built-in control levers or physical buttons. Pair it with a phone and you can steer it through an app, turning it into a small self-driving device that carries things and comes to you on its own. It will also follow its owner around — this is basically a self-driving stool.

Honda and Mercedes-Benz have both found ways to make a car "love" its driver

Besides the motorcycle, Honda brought its NeuV concept car to CES Asia. Its selling points are autonomous driving and all-electric power. As Honda imagines it, owners will one day be able to sell the power in their car's battery when they aren't using the car.

The NeuV has an "emotion engine" meant to make driving easier, more pleasant, more efficient and safer. Toyota has tried something similar: the Concept-i it showed at CES 2017 gradually learns a user's tastes and preferences and responds accordingly.

Honda engineer Jarad Hall has explained that the "emotion engine" uses a camera to read facial expressions, a microphone to catch the tremor in a voice, and sensors to register how a person is sitting. The system teaches itself from this data, works out the driver's preferences, likes and interests, and then offers customized driving services and options.

In short, Honda wants to love its drivers a little more. For now, though, it can't offer any information about the NeuV's development or mass production. A product like this won't reach consumers any time soon.

There's another problem: plenty of carmakers have been working on the driver's emotions for years, but whether people need to have a conversation with their car is hard to say.

Mercedes-Benz has taken up the idea of looking after its passengers too — by inviting them to drink tea in the car. The center armrest in the back of a Mercedes-Maybach concept car has been turned into a storage box for a tea set, with cups and a teapot inside; on an LCD screen in the armrest, passengers can choose pu'er, black tea or oolong. But this kind of localization aimed squarely at the Chinese market is too mechanical. It feels as though Mercedes-Benz China drew up a user profile: financially independent middle-aged and older people.

Mercedes-Benz also launched a concept system called the Serenity Pavilion, which it says evaluates and analyzes data on a driver's driving behavior, heart rate, sleep quality, stress level, traffic conditions and weather.

But the staff at the booth weren't clear on how the system actually works. They only repeated that there were a lot of sensors around the seats, that these produced some data, and that the system used it to adjust the cabin temperature and the scent from the air conditioner's fragrance system to suit the owner's physical and mental state at the time.

If the driver is very irritable, for instance, the system will slow the car down while it's moving; if he is tense, it may switch on the seat massage.

Putting a tea set in a car is a strange sort of "innovation" — the hot water for brewing could scald a passenger when the car hits a bump. And driving demands constant attention: if the system senses that the driver is tense and turns on the seat massage, it may startle him instead, which is a bigger safety risk.

The Serenity Pavilion will probably never get past marketing.

Baidu opened up rides in its autonomous car, but the course was no challenge

Outside Hall N3 there is an enclosed area about a quarter the size of a football pitch, which Baidu had fenced in with waist-high barriers to demonstrate its autonomous cars.

A Great Wall Motor WEY VV7 sat outside the barrier for static display, and a Haval H5 sat on the test track inside for rides. Unlike the Ford self-driving car QDaily had tried before, the Baidu car on show here had no lidar or cameras visible on the roof or around the body. It achieves Level 3 autonomous driving with a single monocular camera mounted on the front windshield, and from the outside it looks no different from the same model sold in showrooms.

Baidu had laid out a slow route full of sharp turns. The track was temporary asphalt, with white arrows painted down the middle to show the direction of travel and white lines along the sides for the camera to read the lane. Beyond the asphalt, the open ground was lined with red-and-white tire walls, the kind used on go-kart tracks.

For safety, Baidu put a driver in each self-driving car. No one was allowed in the front passenger seat; two people could sit in the back. The driver wouldn't set off until the passengers had buckled up.

Once the car was moving, the driver pressed a button near the gear lever and it switched into autonomous mode, holding its speed between 20 and 30 kilometers per hour. In the three minutes we rode along, the computer-driven car behaved naturally, taking corners the way an ordinary human driver would: slow down, touch the accelerator, turn the wheel at the same time, come out of the corner, speed up, then do it all again before the next turn.

After the first lap, a staff member at the side of the track set a left-turn sign on the lane. The point was to show that the car's camera could read the sign, send a left-turn command to the driving system and carry it out.

Through all of this nobody touched the steering wheel, the brake or the accelerator; the car did everything itself. The Baidu engineer playing driver kept telling us, as we drove, that his hands were off the wheel.

The onboard computer and the rest of the hardware used for autonomous driving, including the Nvidia Drive PX 2 platform, lay loose in the trunk, unsecured. The cables carrying the data were tangled together every which way. Baidu didn't allow photographs of the trunk.

Drive PX 2 is the autonomous-driving development platform Nvidia announced last May. It supports 12 camera inputs, laser positioning, radar and ultrasonic sensors. The platform uses two Tegra processors, and its GPU is Pascal, which is ten times faster at deep learning than the previous Maxwell architecture.

After the ride, a Baidu engineer told QDaily that the car had been through six weeks of training on a closed course beforehand, imitating the habits of human drivers, and that this demonstration amounted to Level 3 autonomous driving.

In a Level 3 car, the driving system handles the detection of the environment and the driving itself, but a human still has to answer requests such as whether to sound the horn for a pedestrian or slow down in the rain. This follows the classification set by the U.S. National Highway Traffic Safety Administration (NHTSA), which defines five levels in all.

Last November, Baidu demonstrated a Level 4 self-driving car on real roads in Wuzhen, with several hundred thousand yuan worth of lidar mounted around the body. The difference between Level 4 and Level 3 is that the car no longer needs a human to answer requests like slowing down in the rain; it decides for itself.

Relying on radar or relying on cameras are the two most common approaches in autonomous driving today. The best-known example of the first is Mobileye, which worked with Tesla on Tesla Autopilot 1.0; Baidu's system belongs to the second camp, whose best-known example now is Tesla. After Mobileye announced last year that it would not renew its work with Tesla when the contract ran out, Elon Musk's electric-car company switched to guiding its cars with cameras, and the Model S and Model X carrying the new system have already been delivered to customers.

This isn't to say that whichever approach needs less extra hardware is the better one. They are simply two ways of getting a car to move on its own.

Google and Uber also run public autonomous-driving tests on real roads. On a ride in March 2016, The Washington Post found that Google's self-driving car drove like someone who had just gotten a license: the acceleration was abrupt, and braking gave the same impression — experienced drivers tend to ease off the pedal gently, while Google's braking seemed too fierce.

Baidu's self-driving car felt natural to ride in, which is a good thing. But that was its performance in a closed area with no obstacles, no pedestrians, no traffic lights and no cars arranged to follow in front or behind. You can't judge from it how the car would do on a real road.

In March this year, Baidu folded its Autonomous Driving Business Unit, its Intelligent Vehicle Business Unit and its connected-car business into a single Intelligent Driving Group (IDG). Then, at the Shanghai Auto Show in April, Baidu announced Project Apollo, which aims to open its autonomous-driving software platform to partners and help carmakers put vehicles and hardware together into a complete autonomous-driving system.

Besides the rides outside, Baidu also showed a map-collection vehicle, a high-precision positioning and environment-sensing module, and an in-car entertainment system running DuerOS inside the hall. All of them are part of Project Apollo.

South Korea's Hyundai has already signed a factory-fitted deal with Baidu for the DuerOS in-car entertainment system, and Hyundai's Kia brand will launch the first model from that partnership in November. In the aftermarket, Desay SV, Hangsheng Electronics and PATEO have signed agreements with Baidu. But nothing will be on sale before the end of this year at the earliest.

The traditional car industry came to talk about autonomous driving too, and thinks it is the more aggressive side

BMW had the biggest booth in Hall N3.

At the center of it sat the BMW i concept cockpit, the company's idea of what a car interior will look like in the future: once a car drives itself to a high degree, the computer takes over the accelerator, the brakes and the steering wheel, freeing the driver's hands and feet for other things.

But this is a pure concept design built for marketing, and as usual it will probably never reach the road.

Last July, BMW announced it would develop autonomous driving together with Intel and Mobileye. It brought a self-driving car built on the 3 Series to the show, but only for static display.

BMW fitted three cameras to the front windshield and two at the rear, all supplied by Mobileye. Four lidar units are set into the body — 4-line units from Germany's Ibeo. On top of that, seven millimeter-wave radars are spread unevenly around the car.

Millimeter-wave radar is used mainly for precise measurement of speed and distance. It sees a long way and weather barely affects it; its weakness is that it is poor at identifying an object's width and features. The radar in a common parking sensor is millimeter-wave radar. Lidar scans and builds a 3D model of everything around the car; it measures distance precisely and has high resolution, but its performance falls off badly in bad weather, and it costs more than millimeter-wave radar.

This 3 Series looks much like an ordinary one from the outside, but it costs far more: the parts tied to autonomous driving account for roughly half the price of an ordinary 3 Series, which runs over 300,000 yuan. People buy a BMW 3 Series to get a car that isn't too expensive and still has the brand, the power and the handling.

A German exhibitor at the show gave QDaily his own view: for autonomous driving, he could accept paying 10% more for a car in the 3 Series class, but at 50% more he would look at a higher class of car instead.

At an Innovation Day event, Robert Bruckmeier, the joint vice president in charge of autonomous-driving research at BMW China, said the company is studying China's roads and traffic, and has already demonstrated Level 3 cars and fully autonomous Level 4 cars at the National Intelligent Connected Vehicle Pilot Zone in Jiading, Shanghai.

So far BMW has run 16,000 kilometers of autonomous-driving tests in China. It plans to add about 40 more 740 hybrid sedans, priced above 1 million yuan, for these tests, which may by then attract Tesla's customers, since the Model S and Model X sit in exactly that price range.

There are many ways to work on autonomous driving. Ford and Tesla do it themselves; other companies bring in parts makers, as BMW has with Delphi and Mercedes-Benz with Bosch.

Most people know Germany's Continental for its "Continental" tires, but the company is stronger in chassis and body control. This month it signed an autonomous-driving agreement with Baidu, under which Continental develops the hardware and Baidu works on the algorithms.

"Autonomous driving really is a hot and big topic right now. But we believe autonomous driving is definitely a piece of systems engineering — it isn't something any one automaker, any one supplier or any one tech company can pull off alone," said Liu Yi, vice president of key account management for Continental in China.

Their self-driving car carries one camera and five millimeter-wave radars. Engineer Wang Houming told QDaily that Continental can currently deliver Level 2 to Level 3 functions, and that the next prototype will add lidar and raise the level through partnerships.

Continental showed us the computing equipment stored in the trunk. An aluminum-alloy frame took up most of the space, holding the onboard computer that handles modeling of the surroundings, the human-machine interface and data collection; three or four cooling pipes stood out.

Asked whether the onboard computer suffers blue screens or crashes when it gets too hot inside the car, Continental said it happens now and then: "In summer the temperature in the trunk climbs very easily, which makes the onboard computer handling environment collection restart. We will do more work on cooling to make it run more stably."

A person can feel fatigue coming on, and the right response is to pull over somewhere safe as soon as possible, rest, and then drive on or hand over the wheel. But a human driver has no way of knowing when an autonomous-driving system will fail, and one failure can be enough to wreck the car and kill the people in it.

Next to traditional carmakers, tech companies look far more aggressive about autonomous driving: Google has been working on it since 2009; in 2015 Uber hired more than 50 researchers away from Carnegie Mellon University's robotics institute in one stroke, to develop Level 3 and Level 4 autonomous driving.

Kurt Lehmann, Continental's chief technology officer, told QDaily that this impression does exist on the outside. "Google has been working on Level 5 autonomous driving from the very beginning — no accelerator, no pedals. But they had no store of automotive knowledge, and didn't know how hard building a car actually is." Google has since given up making its own self-driving cars and works with automakers instead, selling them technology.

"Traditional carmakers look more practical, moving up one level at a time. Compared with the tech companies, these are two different research paths, and the goal at the end is the same: full autonomous driving. But before that arrives, consumers can already buy the lower-level autonomous cars traditional carmakers have put on sale. Seen that way, who is more aggressive than whom?" Kurt Lehmann said.

The trouble is that the company that has come closest so far to giving consumers something like full autonomous driving is Tesla — a company founded seven years after Google, now worth more than GM and Ford and the largest carmaker in the United States.

The tech companies still look like the more aggressive ones.

If you think LeEco's Faraday Future promises a lot, here's an even bigger promise

China's National Modern Energy Holdings, the main buyer of the former Swedish carmaker Saab, set up NEVS in Trollhättan, Sweden, in 2012. This time, under the NEVS brand, it brought an extravagantly styled plastic concept cabin called In-Motion, plus two electric cars.

NEVS says In-Motion assumes Level 5 autonomous driving and explores how passengers can better pass the time while they travel, and how to make the most of the space inside the car.

As NEVS imagines it, passengers can customize the lighting theme inside, and use a phone app to set up the riding environment, clean the cabin, move the seats and adjust the mood lighting. In-Motion also wants to use AR to project all sorts of interactive information onto smart window glass.

This is basically the same thing BMW's i concept cockpit talks about. The difference is that BMW spent decades building aircraft engines and then decades building cars, and sold 2.3 million cars last year. NEVS has yet to build a single production car.

NEVS now has a 700-strong engineering team at its Swedish headquarters and 2,200 employees worldwide. Its bigger challenge at the moment is getting its all-electric 9-3 and 9-3X into production and on sale as soon as it can.

According to material NEVS provided, the two cars have a maximum range of 300 kilometers. The company got its Chinese production license this year, will start building electric cars at its Tianjin plant by the end of the year, and expects to put them on sale in China in 2018.

It isn't clear how much technology NEVS has in hand, or whether its production and launch plans will run on time. Back in December 2015 it signed a framework agreement with Panda New Energy to deliver 150,000 electric cars and electric buses. But a car company with not a single product it can sell, coming out to talk about concepts, is a strange thing.

The car accessory people talked about most at the show was the head-up display (HUD)

The head-up display (HUD) uses optical projection to put important information — speed, navigation prompts — onto the glass, so the driver can read it without looking down.

It first appeared on military aircraft, precisely so that pilots could see what they needed without looking down, and would have to check their instruments less often. Now the technology has moved into civilian use.

But car HUDs and aircraft HUDs are not the same. Fighter jets use diffractive HUDs, whose image focus can be set at infinity. Even when you look far into the distance, the numbers on the display stay sharp.

Cars all use LCD projection, where the focus still can't reach infinity. That means you can't read the numbers while you are watching the road. And by the time you are actually reading the numbers, your focus has left the road.

It works like focusing a camera: the center is sharp and everything around it blurs. To read the HUD clearly, you have to shift the focus of your eyes. That distracts you easily, which creates a safety risk.

Mid-range and high-end carmakers usually fit HUDs to their top trims, mostly projecting straight onto the front windshield, which disturbs the driver least and gives the best brightness and color. An engineer at Pioneer, the Japanese in-car systems supplier, told QDaily that carmakers at this stage mostly use LCD- or DLP-projection HUDs, whose problem is low contrast, which tends to leave a gray background.

Laser-scanning HUDs can get rid of the gray background, but laser diodes are still sensitive to temperature and can't meet the automotive-grade requirement of working at up to 85°C.

The car accessory market saw a wave of HUD startups in 2015, with companies like Carrobot, CarPro and Hade launching crowdfunded products.

These products came in two parts: a main unit the size of a palm and a screen of five to six inches, onto which a phone app projected the information. Later, in-car Android entertainment systems, which command a premium more easily, became the mainstream direction for startups. But the enthusiasm itself is open to question — there is no important function that plugging in a phone can't handle.

At this year's CES Asia, the HUD was once again the most talked-about car accessory.

There are plenty of trade shows every year, and the car industry doesn't have that many new things to talk about.

Traditional carmakers like Mercedes-Benz, BMW and Volvo took up most of the floor last year, and it's the same names again this year.

Next to chassis, suspension and transmissions, "smart" is still what carmakers most want to talk about. Wave your hand to open the door, order your car around with your phone, share a car through an app — it costs them very little explaining to make consumers understand what's going on.

But unlike phones, which have to change shape every year or two, the gap between one new car and the next is far too long. New technology often takes ten years or more to travel from concept car to high-end model to ordinary model.

Of all these new things, autonomous driving is probably the only one that has made any real progress.

Unless otherwise noted, the cover image, the images in the text and the video were all shot by the author.