Baidu is clearly the darling of autonomous giants. A few months ago, Tesla chief Elon Musk visited China, the second largest market for the company, to release FSD (full-self driving) robotaxis in the region later this year.
However, this was made possible only through a partnership with Baidu’s Apollo, which is already slaying in the autonomous segment with its fleet of robotaxis making it the largest autonomous ride-hailing service in the world.
Baidu’s robotaxis are currently available in four cities in China, including Wuhan, Beijing, Shanghai and Chongqing. In Wuhan alone, it caters to a population of seven to eight million covering an area of 3000 sq km.
“Here [Wuhan], we have 6000 driverless rides per day. So, all of them combined together is quite a big achievement,” said Helen K Pan, general manager and board of directors for Baidu Apollo, California, in an exclusive interaction with AIM.
Though robotaxis currently operate in only four cities, their autonomous driving service is available in 11 cities across China.
In May, Waymo, a self-driving cab service, confirmed that they are operating more than 50,000 paid trips every week across three cities in the US.
Level 5, Not Yet
Baidu has been working in the driverless segment for over a decade and has developed various levels of autonomy. It currently operates at Level 4. However, their technology borders on Level 5 too. “Typically, we call it L4 and L5, with L4 [autonomy] implying in geo-fenced regions,” said Pan.
“With the government policy, we are actually more in the L4 region.”
As of April 2024, the cumulative test mileage of Apollo L4 has exceeded 100 million kilometres.
While Baidu has been advancing its autonomous capabilities for over a decade, the West is already competing in this space with Tesla and Waymo. However, Level 4 is what they all are at.
“I think we are all at L4 today; and with the government regulation, it’s not possible to do L5. Another thing is that, I think, all of us providing this technology haven’t been tested in all the scenarios. We wouldn’t have this confidence to claim that we have the L5 capability.”
Baidu vs the World?
Pan believes that the training conditions for Baidu’s vehicles act as the biggest differentiator when compared to Tesla or Waymo.
“China has a couple of unique market technology challenges compared to the US and other countries,” said Pan. “The roads are in a diversified condition in China. The driving behaviour, the population density, and the driving scenarios are much more complex.”
Pan also elaborated on how the aggressive driving behaviour of drivers in the Chinese market, especially in certain regions of the country, contributes to building rich data. “Outside China, you’ll take a longer time to see, but you will face those corner cases here more frequently. So, that’s a benefit,” she said.
In addition to rich data, costing is another differentiator. “What I see in China is that the cost of the robot taxi is really low. We have to have the cost constraints much tighter than the rest of the world.”
Interestingly, Pan comes with a rich experience of working with autonomous technology. Before Baidu, she worked with Google and Waymo on their self-driving technology.
Challenges Remain
Though comfortably placed in the autonomous vehicle (AV) space, Baidu is still experimenting to navigate the challenges associated with it. “If you look at it from an AI point of view, I think it’s challenging. AI algorithms are evolving,” said Pan, who explains that you need real-time and extremely high-accuracy data.
Pan confirmed that the company is also looking at developments around large models. “We’re providing this development, an end-to-end large model and we are able to tailor all the diversified scenarios.
“It will also help us migrate from one city to the next, once you have this large model instead of a very customised one for a particular city. So, I think that’s quite a big breakthrough for us to be able to further expand our technology into multiple regions in China,” said Pan.
Safety Regulation
Pan discussed the government’s apprehensions associated with AV technology.
“The government actually wants to embrace this new disruptive technology. But on the other hand, they don’t know how to really regulate it. They want to foster this technology, and make sure it’s not chaotic and that they can really control this one,” she said.
To address this, Baidu is actively working with the Chinese transportation department to help form regulations to ensure safe deployment.
“We run a lot of simulations, and we also do a lot of early testing with the safety driver behind the wheel, and proving our technology is working. Then, we gradually remove the safety driver, and become completely driverless. So, when we reach the driverless stage, we know our technology, and we have the confidence,” said Pan.
Furthermore, there are 10 levels of redundancies built into the autonomous driving systems, and AI algorithms. The Chinese government also mandates specific test conditions and routes to obtain a driverless certification.
When asked if 100% safety can be achieved, Pan was pragmatic in her answer. “Nobody can say everything is 100% safe, but I think it will evolve. If you look at it 5-10 years from now, we’ll still continue to improve our safety.”