The test drivers at South Korea’s K-City automotive proving ground may have one of the easiest jobs on Earth. K-City, set amid quiet countryside outside the satellite city of Suwon, south of Seoul, is, at 36 million square meters, the size of a small airport.
Its facilities encompass a five-mile long high-speed highway circuit; a tunnel; a tree-lined street, a signalized intersection complete with pedestrian crossings – even a school zone.
It is home to 46 test vehicles – supplied by auto makers including Hyundai, Kia, Ssangyong and Audi; by seven universities; by portal company Naver; by telcos KT and SK; by electronics giants Samsung and LG; as well as a range of other bodies and research institutes. The vehicles are jointly used at K-City by some 80 companies and research institutes.
A super high-speed 5G wireless telecoms network has been installed at the 15.6 million euro facility and equipment to be added later in the year includes weather-generators that will replicate conditions of rain and fog. A cyber version of K-City will also be made available for off-site researchers.
It is operated by the Korea Automotive Testing and Research Institute, which falls under the Korea Transport Safety Authority. The reason test drivers there may not have as much to do as their counterparts at more conventional proving grounds is is that K-City is a test bed for autonomous vehicles – or, if you prefer, self-driving cars.
In addition to advising the Korean government and local and overseas companies and bodies, K-City staff are part of the UN World Forum for Harmonization of Vehicle Regulation’s Working Party 29. In that capacity, they coordinate with researchers at other autonomous driving test tracks around the world – one in China, one in Japan, one is Sweden and two in the United States.
During a spin around the tracks in two self-driving cars and one self-driving bus, Asia Times was given an on-the-ground review of where the technology stands – and what the issues facing it are.
Rev it up
The vehicles at K-City are armed with a combination of sensors that enable hands-free driving. Visual sensors – cameras – provide visual information, such as “reading” lane lines painted on road surfaces. However, visual sensors operate poorly in rain and while they can capture movement, they cannot configure distance and velocity.
Lidars – a portmanteau of light and radar – measure distance by illuminating an object with pulsed laser light. And radars measure movement, but cannot configure distance or velocity.
For optimal results, Hong Yun-seog, head of the Transport Safety Authority’s Automated Driving Division, advises a suite of all three sensor types, with built-in redundancies – “sensor fusion” – for maximum efficacy and safety. So cameras, lidars and radars all operate together, networked in the car via a black box.
Even so, the Mark 1 human eyeball still outranges them all: none of the sensors have a range of more than 200 meters. Another problem is the height of some traffic lights in South Korea: camera sensors might not have a field of vision broad enough to “see” high lights, Hong said.
First, we tried a bus. Once the Automated Driving System, or ADS, was engaged, the driver removed his hands from the wheel and the vehicle navigated itself, via the on-board telematics, around an eight-minute course, including bends, at a genteel speed.
A K-City official noted that as ADS becomes the norm, they will impact not just cars’ onboard electronics, but also interior design: A car’s interior will transition from passenger transport spaces to working spaces and/or living spaces. The experimental 5G bus at K-City offered computer games which are playable on the windows, which are made of transparent displays rather than glass.
For our second test run, we tried a passenger sedan and went for a drive around a mock urban course. The car successfully navigated though lights and a replicated school zone.
Finally, we tried a second passenger sedan on the high-speed highway. Once cruise control was engaged – via a button on the dash – the vehicle took over, keeping its distance from other cars on the road and accelerating and slowing as necessary. For the driver, switching from cruise control and engaging manual was a simple affair – simply grip the steering wheel, or touch either the accelerator or brake pedal with the foot and you have it.
Critical time ahead
In 2015, Seoul set the goal of commercializing “Level 3” ADS-equipped cars for South Korean roads in 2020. “Level 3” means cruise-control highway driving, but not for inner-city streets or small rural roads. In 2015, achieving Level 3 looked like a daunting challenge for Hong and his team.
“There were many complaints among my researchers, they were complaining that it was too hard,” Hong said. But with the speed of the technology advancing, “Right now, 2020 is OK,” he said.
As the era of mass self-driving cars approaches, prices are coming down. As recently as 2009 the price for car lidar was US$80,000. Now, similar sensors are only US$1,000, and Hong expects them to fall to US$100 as they are commoditized. In South Korea, Hyundai is already preparing autonomous vehicles for sale in 2020. A Level 3 ADS function build into a passenger sedan adds an additional US$5,000 to US$10,000 to the overall vehicle price, Hong estimated.
The final phase of testing for Level 3 automatic driving – live, on actual roads, rather than in K-City’s courses – is just getting underway.
The algorithm for “Level 2” driving – that is, in urban areas – is not yet ready, said a driver-researcher. In the more distant future, Hong expects all cars’ AI to be linked via 5G or higher networks, allowing cars to communicate with each other while navigating using sophisticated electronic maps loaded into the Cloud. This will create a seamless experience – even in the heaviest and most complex urban traffic scenarios.
Yet despite the various high technologies – optical and other sensors, telematics, AI – that are converged in ADS, it is still all about people.
“In my opinion, the most difficult thing is human behavior,” Hong said. Asked his biggest concern, he detailed the ADS’ manual over-ride. “When the ADS sensor is not working, it requests the driver take over,” Hong said. “That needs some time.”
Once the ADS senses a fault, the driver is notified by both a light on the dashboard and an audio alarm. The critical question hanging over much of Hong’s research is how long the time shift from automatic to manual control should take.
Hong, who is advising the Korean government on the matter, notes that manufacturers think four seconds is an appropriate takeover time. But Japanese research suggests 10 seconds, while Hong, based on his own findings, believes it should be set at eight seconds.
This time window is not so much technologically critical as legally critical, Hong explained. That is because, until the driver is fully in control of the car after the alarm has sounded, any legal liability – should there be an accident or worse – rests with the manufacturer.
“This is really important,” Hong said, noting that some 20 laws interface on this liability. “It is a legal issue.”
Utterly a useless colomn about ďriverless car (DLC). There is so much of hype and fictional imagination about what these DLCs, nobody has written about what they can not do!
Here is a question: what happens if markers-lines are covered by snow drifts or muddy water in a rain storm. Will DLCs come to a halt or if it happens to be a dead body of a dog on the road? What if an accident can be seen say mile away where a real human driver can take evasive action by turn around and take a side road. Can DLCdo it?