I’m Koji Minoda, a software engineer at TIER IV. During a trip to China, some team members and I conducted research on autonomous driving technology in the country. I’d like to share some of the findings from an engineer's perspective. Last year, a group of us did something similar in the U.S., so this trip was part two of our overseas research.

China, alongside the United States, is a country where autonomous driving technology is thriving in terms of research and development. Companies like Baidu and Huawei are directly or indirectly involved in the development of autonomous driving and advanced driver assistance systems (ADAS). Some services have already been launched but most of the online information about projects is in Chinese. And there are almost no videos posted on platforms commonly used in Japan, which makes it difficult for non-Chinese speakers to get information.

During our trip, we visited Guangzhou and Shenzhen, where we checked out autonomous taxis and buses by companies such as Baidu Apollo, WeRide, and Pony.ai, as well as Level 2+ and Level 2++ electric vehicles (EVs) developed by even newer market players. We also gathered insights on the latest developments from our colleagues in China.

Autonomous taxis, buses

Overall, the ride experiences were exceptional across the board. Some services operate similarly to Uber, where passengers can hail a ride with an app, and the level of the services was impressive. However, unlike Waymo and Cruise in the United States, a safety driver was generally onboard. We also encountered a few overrides during our journeys, which suggests the U.S. might be slightly ahead in this respect.

It's worth noting that many of these services require passengers to have a Chinese ID, so if you’re planning on visiting to check out the technology, be aware of this.

Baidu’s Apollo

Firstly, let's look at Apollo from Baidu, one of China's leading search engine companies. This autonomous driving project is an open-source initiative and it is often compared to Autoware. Apollo’s robotaxis operates in over 10 locations across China, and we had the opportunity to ride in one of their robotaxis in Shenzhen.

The passenger display shows route information.

The vehicles are based on a model developed by Beijing Automotive Group’s ARCFOX EV brand. One of the hurdles in Level 4 autonomous driving is cost, but Apollo has managed to bring the cost of their vehicle down to around 480,000 RMB (approximately 10 million yen), a third of the cost of other options out there. I was curious to know how their system covered the vehicle’s rear and sides with fewer LiDAR sensors compared to Waymo and Cruise.

Apollo’s robotaxi service is similar to that of Uber. Users can select pick-up and drop-off locations to make a reservation with the BaiduMap or WeChat apps. A Chinese ID is required for the registration process.

The safety driver interface in Apollo vehicles was one notable difference compared to WeRide and Pony.ai. It includes buttons for voice memos and accident reporting, allowing the safety driver to record unexpected behavior or problems during autonomous operations. Unfortunately, I didn't have the chance to witness the interface in action during my journey, but it was clearly different from the interface on the passenger-facing display, which was interesting.

The safety driver interface

The UI for passengers was similar to those of Pony.ai and WeRide, displaying things in the vicinity, speed and signal information. Unfortunately, the system wasn’t working properly when I was on board, but I understand it has entertainment features, such as a music player.

As for the ride experience, despite being rush hour, there were few instances of unnatural maneuvers. However, during another person's ride, a buzzer sounded during a left turn, prompting the safety driver to override the system.

On the day of our journey, there seemed to be fewer vehicles in operation, resulting in significant wait times. I waited for over three hours, but the pick-up location might be to blame. If you plan to ride with Baidu in Shenzhen, I recommend allowing plenty of time to avoid disappointment.

WeRide

WeRide is an autonomous driving developer based in Guangzhou, China. In the vicinity of the company’s headquarters, WeRide autonomous driving technology powered not only taxis and buses but also cleaning vehicles. Guangzhou seems to have developed into a futuristic city of autonomous driving.

Given that WeRide was granted approval for the first fully autonomous vehicle road test in China in 2020, it's perhaps unsurprising that in Guangzhou, pedestrians didn't seem particularly intrigued by autonomous vehicles. Autonomous driving has already become a part of daily life.

WeRide autonomous bus

Near the headquarters, members of the public could ride two shuttle buses free of charge. The ride was smooth, but the driving felt a bit aggressive at times, especially when it seemed like the vehicle was cutting it close to avoid parked cars on the roadside.

One interesting moment was when the bus honked its horn because the vehicle in front was driving slowly. In Japan, not many people use their car horns. It seemed somewhat aggressive for an autonomous vehicle to honk at other vehicles when there wasn't any imminent danger. However, on China’s urban streets, I noticed that car horns were used frequently. It struck me as a fascinating cultural difference.

Autonomous test vehicles line a road at WeRide’s headquarters

Numerous test vehicles operate continuously in the area around WeRide’s headquarters, reflecting the company’s commitment to autonomous driving. It reaffirmed my desire for us to also advance autonomous driving in Japan without delay. We weren't able to ride in any of the autonomous vehicles operating near the headquarters, but we were able to use an autonomous taxi service in another part of Guangzhou, and the car model was the same.

A robotaxi powered by WeRide technology

We used the taxi service during rush hour, when the roads were heavily congested. There were many challenging situations where a conservative autonomous system might have struggled, but this system displayed decisiveness akin to a skilled driver.

Pony.ai

Pony.ai was founded in 2016 by former Baidu employees. Its headquarters is in Fremont, California, and it also has a base in Guangzhou. We hailed a ride in a Pony.ai vehicle from the Hong Kong University of Science and Technology campus in Guangzhou using the PonyPilot+ app. Users can also hail rides with the popular ON TIME app.

The vehicle navigated smoothly along a street lined with parked cars. It also managed to find gaps between parked cars to pull over. With moderate traffic and a short distance to cover, there was no need for manual override.

The driver display features control buttons, speed indicators and route information.

The driver display had a slick interface. There was also a passenger monitor facing the rear seats that displayed the route and fare details, as well as the estimated time and distance to the destination.

Chinese autonomous vehicles were all pretty similar. In comparison to the vehicles from U.S. firms such as Waymo and Cruise, the Chinese vehicles seemed to have quicker acceleration and more responsive steering, allowing for finer control.

EVs, passenger cars

In China, it’s not all about Level 4 autonomous driving. The development of Level 2, Level 2+, and 2++ systems is thriving, too. Many such vehicles were on display at malls in Guangzhou and Shenzhen.

NIO ES7

First up was the NIO ES7, which is equipped with Seyond LiDAR sensors. Cars with LiDAR were a common sight. It was surprising because the technology is beyond the stage of concept cars. NIO ES7 is already on the market.

NIO ES7

Marketing materials showcased the fact that the vehicle is equipped with the NVIDIA Drive ORIN system-on-a-chip and included performance information such as the number of tera operations per second. The materials also highlighted that it has 8 MP cameras and LiDAR, which suggests car buyers are conscious of these features.

LiDAR is often viewed as a component that compromises the design and aerodynamic performance of vehicles. However, according to a TIER IV team member who lives in Shenzhen, LiDAR is becoming a sign of luxury among EV buyers, thanks to NIO's advertising. That was also a big surprise.

Li Auto L7（理想汽车）

Another vehicle we spotted was the Li Auto L7, which is equipped with HESAI LiDAR sensors. As a side note, Li Auto has published a paper on arXiv about BEV-CLIP, a model for image retrieval from driving scenes. Perception development and evaluations heavily rely on data. It appears that Li Auto is taking an approach of gathering driving data using onboard sensors, cherry-picking crucial data, and annotating them.

It's unclear to what extent LiDAR sensors are being used in the vehicle’s autonomous driving system, but the fact that the cars are collecting LiDAR data could be an investment for the future. In that sense, the sale of LiDAR-equipped cars marks an important development in the sector.

A display showcases technical details about the Li Auto L7.

Many companies appreciate that the data collected today will lead to market strength in years to come. So what about TIER IV, you might ask?

TIER IV believes in the power of an open ecosystem, where instead of each company hoarding data internally and aiming for dominance based on that data, companies collaborate to collect data and promote development as a whole. To achieve this vision, TIER IV is launching the Co-MLOps project.

Huawei Luxeed S7

Finally, let's talk about the Huawei Luxeed S7. What's particularly interesting about this car is its in-vehicle infotainment system (IVI). It is possible to configure ADAS functions and display recognition results using the Harmony OS-based navigation system.

A display in the Huawei Luxeed S7 shows sensor data.

Naturally, this car is also equipped with LiDAR. I was curious to find out if the recognition results shown in the IVI were from the camera or LiDAR, so I blocked the camera, which distorted the results. It's possible that the current system relies solely on the camera for detecting surrounding objects, and LiDAR is used exclusively for data collection.

Team members conduct an impromptu sensor test.

Other players

EVs from brands like Xpeng and AION were among those being showcased in malls across Shenzhen and Guangzhou. What struck me was how LiDAR-equipped vehicles were almost the norm. I got the impression Chinese manufacturers are approaching car design with a lot of freedom and creativity.

For example, take the Huawei Aito M9, which can display the time in its front lighting. If I were involved in the planning, I might wonder if such a feature was really necessary.

Huawei Aito M9

Then there’s AION, whose vehicle comes equipped with karaoke microphones, projectors, and massage machines. The idea of creating a car with a karaoke system might pop up during a brainstorming session, but whether it would actually be developed and sold is another matter entirely.

Unfortunately, we weren’t able to ride in any Level 2+ or Level 2++ vehicles but I’m eager to do so soon.

Wrap-up

During this trip, I had the opportunity to ride in models developed by Baidu Apollo, WeRide, and Pony.ai, and got to see many other autonomous vehicles. Chinese companies are steadily approaching Level 4 autonomy, focusing on autonomous driving within defined areas. I was also impressed by the smooth ride experience and maneuverability of the vehicles. It reinforced the notion that driverless travel is just around the corner.

Written by Koji Minoda. TIER IV engineers Hidenaga Ushijima, Kotaro Yoshimoto, Shumpei Wakabayashi and Shintaro Tomie also contributed to this blog post.

Koji Minoda ｜Perception team
A graduate of the master's program in aerospace engineering at the University of Tokyo, Koji joined TIER IV in April 2022 following a stint as a part-time engineer from November 2020. He currently leads efforts to enhance the performance of machine learning models through dataset construction on the perception team.