Apollo Moon: Baidu’s Autonomous Vehicle Available for Public Use

The story of autonomous driving has been told for so many years. When will we truly experience the driverless vehicles without a driver?

Baidu Apollo said, now is the time.

On October 28th, the Baidu “Apollo Day Technology Open Day” with the theme of “Going to the Unmanned Land” was held in Beijing. At the event venue, Baidu publicly displayed three models of Apollo’s fifth-generation shared autonomous vehicles (also known as Robotaxis) – the Jixu version, the WmAuto version, and the Aian version. They all share a common name – Apollo Moon.

More importantly, Apollo Moon is not the “concept car shell” commonly seen at many auto show launches, but a mass-produced vehicle that can run on the streets.

On the day of the event, Baidu held a test ride experience of the Jixu version Apollo Moon at the Beijing Shougang Park.

Before introducing the actual experience, let’s take a look at the Apollo Moon itself.

As early as June of this year, Baidu Apollo and ARCFOX Jihou released the Apollo Moon mass-produced shared autonomous vehicle based on the Alpha T, and you can review the details by clicking on Baidu Apollo’s “A big step: the fifth-generation driverless car with a cost of only 480,000 yuan.”

The core highlights of the “too long to read version” are as follows:

Apollo Moon is all front-mounted mass-produced. The pre-installation process in front of the production line is completed on the production lines of the cooperative auto companies, instead of buying the mass-produced cars and then adding a “hat” of the lidar.

Naturally, this brings further optimization of the vehicle’s manufacturing process and improved quality.

The entire vehicle’s computing capacity exceeds 800 TOPS, and the roof uses one customized Velodyne lidar, and the front of the car is equipped with one redundant safety lidar. The vehicle is equipped with 13 cameras and 5 millimeter-wave radars (front three and rear two), achieving Level 4 autonomous driving.The underlying architecture of the autonomous vehicle is based on Baidu Apollo “ANP-Robotaxi”. To learn about the specific performance of this architecture, you can click on our previously shot video “ANP Without Lidar: Reliable City Autonomous Driving Assistance?”.

According to the official explanation, “as long as a mass-produced car supports Baidu ANP, it can become an Apollo Moon autonomous vehicle by adding a roof ride-hailing screen and a customized LiDAR.”

The autonomous vehicle runs on the road every day. What if the accuracy of the radar and camera is affected by bad weather?

Baidu Apollo engineers have already thought about this problem and have configured self-checking and self-cleaning functions for these sensors, ensuring safe driving in complex weather and city road conditions.

For autonomous driving safety, 46 safety technologies and 59 travel service designs are used, with full redundancy of all sensors and computing units, and a high success rate of delivering in complex city roads of up to 99.99%.

Behind this is also a very interesting 5G cloud driving system, which we will introduce in detail later.

Finally, the cost is controllable: the overall cost of the Arctic Fox version of Apollo Moon has dropped to 480,000 yuan, which is one-third of the industry average.

After reading the introduction PPT before the test ride, I suddenly realized that although this event was called “Apollo Day Technical Open Day”, there were not many dazzling technical parameters introduced or boasting about how advanced Baidu Apollo’s autonomous driving technology is.

Overall, everything is trying to tell everyone that L4 autonomous driving is no longer elusive. Apollo Moon is a “real car” that can be mass-produced and driven on the streets and alleys.

Next, let’s talk about some highlights that I experienced during my short test ride on Apollo Moon at the Beijing Shougang Park for 2 kilometers in just 5 minutes.

“Have you experienced our fourth-generation autonomous vehicle before?” This is the question that staff asked me the most during the event that day.

And they’ve come to the right person. Before, I had made a video of riding in Baidu Apollo’s fourth-generation autonomous vehicle at a Supercharger station. You can click on Higher Brother’s video to review our experience that night.

When I replied positively and asked what changes had been made in the new version, the staff repeatedly emphasized the same answer: there is no human, really no driver, not even a safety officer.

Me: …

But to be honest, for an autonomous vehicle, this feeling of “not feeling any difference from human driving” is perhaps the most valuable progress.

The Apollo Moon ride is very close to the experience of ordering a ride through apps like Didi.

First of all, the autonomous vehicle is not “arranged” by the staff in advance. Passengers need to use the “Rabbit Run” app in their phones to select the destination and call the vehicle at the designated pickup point.

When Apollo Moon arrives at the pickup point, passengers need to enter their phone numbers on the touchscreen located behind the C-pillar to verify their identity, and the door will open automatically.

After entering the vehicle, fasten the seatbelt, and confirm the start of the journey on the interactive screen behind the front seats.

Of course, if you look at the front, there is no driver or safety officer behind the steering wheel.

During the journey, the autonomous vehicle can accurately recognize traffic lights, make common driving actions such as turning and U-turning, etc.

Overall, the entire ride is quite smooth and seamless, and you cannot feel the difference from human driving.# Previous Headache of Autonomous Driving Systems

What was the biggest headache for autonomous driving systems before? Ice cream cones.

Apollo Moon can now accurately identify ice cream cones and successfully avoid them when encountering construction sections. Obstacles in the construction zone can also be displayed in real-time on the front and rear screens.

In addition to precise perception of stationary objects, Apollo Moon also achieves dynamic recognition of pedestrians and bicycles in motion. During the test ride, when encountering pedestrians, Apollo Moon will brake in advance, slow down and even “honk” to alert them.

Unfortunately, the overall test ride on that day was relatively smooth and did not encounter more extreme traffic conditions such as food delivery drivers peeking out from nowhere.

But overall, it is still true that “experience without experience” is the most precious – except for the absence of people in the front row, there is almost no specific difference from taking a ride-hailing car in daily life.

Of course, it is difficult to vividly display this feeling with words alone. If you are interested, it is recommended to go to the Beijing Shougang Park and call an Apollo Moon with “Luobo Kuai Pao” app for an actual experience.

After seeing the excellent performance of the Apollo Moon, everyone may still have a question in their hearts about autonomous driving technology. What if the system encounters a BUG despite being advanced and accurate in recognition?

Of course, the engineers have also thought of this problem, and their solution is quite straightforward: if all else fails, just call another designated driver and hand it over to manual operation…

However, this designated driver will not come in person to solve the problem, but will take over in the cloud.

This is another interesting new system that Baidu unveiled during the event besides the first-hand experience of Apollo Moon – Apollo 5G Cloud Valet Enterprise Edition.

# 5G Cloud Driving is dedicated to providing remote assistance to autonomous driving. For example, in the face of temporary road changes or traffic control, upon receiving a request for help, remote drivers can take over the autonomous driving vehicle through “5G Cloud Driving” to help solve the problem.

And this cloud driving system looks like a game cockpit? At first glance, it does indeed resemble one!

Let’s take a closer look at what the steering wheel of the cloud driving system looks like… Well, Baidu Apollo’s engineers probably drew inspiration from Sony PS game consoles.

However, as it pertains to driving safety, 5G Cloud Driving is certainly more than just an electronic game.

Baidu stated that the system, with its multiple patented technologies, can address safety issues with different fault or risk levels, efficiently handle complex real road scenes, and form four advantages of “safety, efficiency, scale, and multiple scenarios.”

In terms of safety, the system has a comprehensive safety hierarchy design that includes active safety, safety warning, and safety foundation functions. It can monitor the state of the driving cab, network, and unmanned driving vehicle in real-time, and make safety treatments based on different fault or risk levels to further escort autonomous driving operations comprehensively.

In terms of efficiency improvement, the system has a real-time remote driving function that can respond to autonomous driving vehicles’ remote assistance requests in milliseconds, enabling efficient handling of various complex scenarios on real roads and helping vehicles quickly get out of trouble.

In terms of scale, 5G Cloud Driving gives operators the ability to remotely control remote driving on the cloud, and through on-board identification and reporting Cases, cloud scheduling, and cabin rescue can be achieved on a large scale that facilitates efficient allocation of driving resources while also accelerating the deployment of autonomous driving vehicles on a large scale.

In terms of scenarios, the enterprise version of 5G Cloud Driving has already been deployed in a variety of business scenarios. In addition to the passenger car industry, the system also supports remote driving of operational vehicles in smart mines, unmanned ports, and warehousing logistics scenarios. With only one driving cabin, N unmanned vehicles can be deployed, reducing operating costs.

However, this also raised several questions for me at that time, and I received answers in my discussions with Baidu engineers.The first question, of course, is whether network transmission quality will affect safety.

According to the reply from a Baidu Apollo engineer, the system and the vehicle’s autonomous driving system complement each other. If there is a safety risk caused by unstable network transmission, the vehicle’s autonomous driving system will intervene and take emergency actions such as stopping in real time.

It sounds like a big challenge for the algorithms.

The second question is whether this system will be a standardized product or adapted to different vehicle models.

The answer I received is that the hardware cabin will be a standardized product, but the software will be adapted for different types of cars and commercial vehicles. For the operators in the back end, the “driving feel” of remotely controlling different vehicles will be different.

It sounds more like a driving simulation game. Friends who have played games like Need for Speed and Euro Truck Simulator will understand that even using the same console, the driving experience of different vehicles is different.

This also makes me look forward to a “cloud test drive” activity organized by Baidu Apollo in the future, allowing everyone to experience the real car driving experience of remotely controlling a car with software. I believe it would be a good marketing idea.

Finally, let’s summarize the “three elements” of Baidu Apollo’s autonomous driving evolution according to the official statement:

Front-end mass-produced vehicles: Apollo autonomous driving vehicles have evolved from a single fourth-generation vehicle to three fifth-generation vehicle models including Apollo Moo Arctic Fox Edition, WM Edition, and AE Edition, providing more choices for “carrots on the run” users.

AI senior driver: The safety mileage of autonomous driving road tests has tripled from 6 million kilometers to 18 million kilometers.

5G cloud driving: From a single remote rescue scenario to three landing scenarios including “shared travel services, special operations scenarios, and smart city services”.

Until the first half of 2021, Baidu Apollo’s autonomous driving service platform, “Luobo Kuaipao,” has launched passenger transportation services in five cities in China, including Beijing, Shanghai, Guangzhou, Changsha, and Cangzhou, serving more than 400,000 passengers.

According to the plan, in the next three years, “Luobo Kuaipao” will be deployed in over 30 cities across China, using 3,000 autonomous vehicles to serve over 3 million users.

So, it looks like we may not be able to buy our own fully autonomous vehicles in the next three years, but we might be able to hail a driverless ride-hailing car in certain specific scenarios and routes.

It’s quite amazing when you think about it.

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.