Advertisement
Advertisement
Self-driving cars and autonomous vehicles
Get more with myNEWS
A personalised news feed of stories that matter to you
Learn more
Having worked on autonomous driving for over a decade, I think now we’re finally at a tipping point, says AutoX founder. Photo: Handout

China’s Professor X says we are at the tipping point for mass roll-out of self-driving cars after tech advances

  • HD mapping and deep learning capabilities are two technologies that have pushed autonomous driving forward

For nearly a century engineers have been working towards the goal of self-driving cars, a technological puzzle that has occupied some of the best minds in science and which has the potential to open up a multibillion-dollar market.

Today we see hands-free valet parking and auto-pilot navigation in the US along with limited robotaxi services in the US and China but how close are we to seeing this technology become part of our everyday lives?

Xiao Jianxiong, founder and chief executive of autonomous driving start-up AutoX, spoke about the future of autonomous driving in a webinar organised by the Post this week.

Also known as “Professor X” after a stint at Princeton University as an assistant professor, Xiao founded his company in 2016 with a goal to “democratise autonomy” and make driverless technology universally accessible.

AutoX, which counts Chinese car giants SAIC Motor and Dongfeng Motor among its investors, signed a partnership with the Shanghai authorities last month to build a robotaxi pilot area in Jiading district.

The Shenzhen-based company has also received approval from the California Public Utilities Commission to operate self-driving ride services to public passengers.

Q: Can you give a broad overview of the current state of autonomous driving and where you see it going in future?

Having worked on autonomous driving for over a decade, I think now we’re finally at a tipping point.

There’s been a few breakthroughs. The first one is HD mapping, which allows us to significantly increase the reliability of auto-driving. The second one is deep learning, particularly the strides made since 2012, which now enables us to crunch a lot of data. This can be channelled to robust algorithms for object recognition and object prediction to provide better perception capabilities for self-driving cars.

Didi follows Google with plan to set up independent self-driving unit

Underlying that there are also other technologies. Big data is a very simple idea but it has taken the artificial intelligence community a few decades to finally figure out that this is the secret sauce.

And it’s only possible when you have a lot of computational power and there has been a significant increase in the last decade. More powerful CPUs, GPUs and deep learning acceleration enable us to crunch more data and build better algorithms.

Autonomous driving has been in the news recently – people have always liked it but now the technology is more mature. It’s an exciting time in human civilisation – finally we are about to build a machine that can drive cars.

Q: How far away are we from true autonomous vehicles – where you can literally step in and fall asleep while the car takes you from point A to point B?

This is level four autonomous driving – a complete self-driving car where the human driver does nothing at all.

From an industry perspective, it depends on the geolocation, because different countries and different cities have different levels of complexity in terms of driving.

For most urban settings in the US, it is getting really close, maybe one or two years more. Robotaxis will be a reality in urban settings by then. That is from a technical standpoint – there could be other challenges.

China self-driving players narrow the gap in US road tests

China is slightly more complicated and it may take a little bit more time. Maybe in three to five years’ time we will see mass deployment of robotaxis in some Chinese cities.

Q: What are the technological hurdles that we still have to cross to get to level four and beyond?

From a science point of view – not much. But to build a viable product you need to figure out the science and the engineering part. You may know how to build a rocket but lack the resources to build a real one – it takes a lot of engineering effort to make it a success.

Building a product that is 100 per cent safe requires mastering a lot of details to ensure all the different components are perfect and reliable. For level four autonomous driving, safety is critical.

That is why to a lot of companies including leader Waymo and major players like ourselves are focused on real testing, gathering data and figuring out what the imperfections are. It’s going to take time to figure out faults and correct them.

Q: How safe are autonomous vehicles now compared to human drivers?

In basic situations, self-driving cars are safer than those with a human driver. That’s because computers don’t suffer from being tired, distracted, impatient, or road rage.

But computers can make other mistakes that a human would consider stupid – such as mistaking a plastic bag floating in the air as a flying rock. Should the car stop for a plastic bag or run over it? Real life can be complex.

Q: With safety, where does liability lie in the event of an accident, injury or damage?

I think in general the insurance industry should be involved – it needs to adapt to the new technology. We usually design a black box within the car where all data is recorded in great detail. The data can be used to review a situation and pinpoint the problem or source of the accident in a fair way. Insurance companies can adjust their premiums accordingly, to reflect risk.

China lagging US in self driving cars but gap is narrowing, says start-up Pony.ai

Q: Is control of autonomous driving done at the edges or at the centre?

To ensure safety, we need to think about the worst case scenario and that is when there is no connection at all. Whether it is the GPS or a 5G network connection that is disrupted, we need to know that the car can drive itself safely.

5G and V2X technology (tech that allows the car to communicate with external objects such as traffic lights) should be developed in tandem so that ‘control from the edge’ and central, mutual-communication technologies work together.

Q: Does 5G really matter to the widespread take-up of autonomous cars?

5G is supplementary, it is not absolutely essential for self-driving cars. In the US there is no 5G and yet Waymo has 1,000 self-driving cars on the street every day. But it will be crucial in terms of mass commercial deployment, when we have a million self-driving cars on the street that are like robots that can go anywhere.

5G helps to monitor the functioning of self-driving cars, it finally creates the bandwidth so that we can stream data in real time to a centralised monitoring system.

Q: How do you view the disengagement report developed by California’s transport authorities? Is it a good gauge of progress?

Autonomous driving companies like Waymo and AutoX are currently focused on cornering by self-driving cars, so they need to stress test the system. We are focused on how to deal with complete system failures, so we can fix those issues.

China’s Pony.ai ramps up its robotaxi fleet by 10-fold to 200

The California DMV report offers interesting data but there is a trade-off. The numbers are interesting only if the test drive is done in complicated road scenarios. But the problem is that sometimes companies optimise their MPI (miles per intervention) by driving on ‘simple roads’ where there is no traffic.

For companies like Waymo and ourselves we want our MPIs to be as low as possible. If your car drives for 1,000 miles with no mistakes this could be a waste of time, money and energy. It is better to go to downtown China for a stress test, increase the difficulty level and fix the issues that are exposed.

 Q: Would cars trained in China be able to drive in the US and vice versa?

If the cars are trained in China, where there are more challenges, they may do well in the US. But if the car is only trained to drive in suburban areas in the US, it may prove difficult to drive in China, maybe impossible.

It is crucial that companies test in China to drive in the country. Not only is it more difficult, the driving behaviour is also different with more bikes and motorcycles, for example.

For more insights into China tech, join our Facebook group, subscribe to our Inside China Tech podcast, and download the comprehensive 2019 China Internet Report. Also roam China Tech City, an award-winning interactive digital map at our sister site Abacus.

Post