Hello everyone, I’m Mr.Yu. Today, let’s talk about some things in the cockpit.
Coincidentally, GeekCar’s traditional skill, “Intelligent Cockpit Intelligence Bureau,” has entered a new series. Following “Wonderful Car Machine Intelligence Bureau,” in order to facilitate everyone’s familiarity, we originally wanted to use the previous name. However, I believe everyone has already realized that it is no longer just about telling stories about “car machines” now.
Today, both traditional car companies and new forces are striving to understand and interpret people’s needs for experiences, and constantly working hard to introduce more wonderful digital experiences into cars. Of course, intelligence is not just about putting a computer or smartphone on your center console. The evolution of each new thing is a long process, regardless of right or wrong, only different in paths and forms.
As an industry media that has been observing cutting-edge automotive technology and products for a long time, GeekCar launched the “Wonderful Car Machine Intelligence Bureau” program two years ago. We tried to use our eyes to observe and distinguish together with everyone, what is truly intelligent, and what is “not that true.”
In the first two seasons of the “Wonderful Car Machine Intelligence Bureau,” we evaluated dozens of different car machines. Through constant experimentation, comparison, and optimization, we have established a complete intelligent car machine evaluation system framework. In early 2021, we also produced the “GeekCar Intelligent Cockpit Orange Book 2020,” which includes the intelligent cockpit product list in different value directions, ten cabin product introductions selected by GeekCar, and many soul-searching questions based on industry status quo.
Not only because these are things we like, but we also hope that more car companies, supply chain partners, media colleagues, and users can obtain in-depth understanding of each other and equal communication through GeekCar. And, try to bring more valuable voices to the industry through our influence.
Best of all, we have made many good friends. The GeekCar Cockpit Communication Group has nearly a thousand predecessors and fellows working in the frontline of intelligent car cockpit field. They discuss various professional topics every day, exchange insights, and inspire creativity.In today’s fast-paced automotive industry, we all realize that the iteration speed of intelligent in-car experience is getting faster and faster. So, are the current definitions and evaluation frameworks of experience still applicable? Is there a possibility for an upgraded iteration?
Fortunately, we have encountered the intelligent car evaluation team at China Automotive Innovation Corporation (CAIC). This team is at the forefront of the automotive intelligence industry, including experts in mapping and navigation, IoT, V2X, and human-machine interaction.
More importantly, this team has the accumulation, ideas, and execution capabilities, as well as the passion to win.
We all realize the same problem. Therefore, both sides have decided to cooperate deeply and jointly build a more scientific and rigorous intelligent cabin evaluation system with more dimensions, based on their own accumulated experience and resources.
To better adapt to new technologies, products, and trends, the previous hardware + ecology + interaction evaluation system will no longer be applicable.
The new evaluation system focuses more on the functional, spatial, interconnected, and human-machine interaction experiences of intelligent cabins. Each dimension includes dozens to hundreds of quantifiable evaluation items.
GeekCar and CAIC Intelligent Car Evaluation Team will jointly conduct a series of intelligent cabin evaluation activities based on this system.
At the same time, the evaluation objects will not only target newly launched vehicle models but also cover the models presented to everyone through our previous evaluation system-for example, today’s XPeng P7-to verify the rationality of the new framework.
Most importantly, this system will not only be used for the evaluation activities of the new season of “Intelligent Cabin Intelligence Agency” but also be brought to the forefront of the intelligent automotive industry by partners of the CAIC Intelligent Car Evaluation Team.
Since we have all chosen to continue down the road of automotive technology, we must not stop our scrutiny and thinking about the present, and our judgment of future trends.
As the first issue of this season, we first invited an old friend, the XPeng P7.
Speaking of an old friend, it has been exactly one year since our last review of XPeng P7’s car system.
Interested friends can refer to our second season of “Wonderful Car System Intelligence Agency” and the issue of “XPeng P7 Car System Evaluation: It Can Actually Be Better” in the link above.
As we said before, the value of OTA is to make devices continue to export technology-oriented experiences.Translate the following Markdown Chinese text into English Markdown text, in a professional manner. Keep the HTML tags inside the Markdown and output only the result.
Since the hardware as the “body” has not changed, let’s see if the “soul” inside the current XPeng P7 has grown well.
- This review is based on the XPeng P7 smart system Xmart OS 2.6.1 released on August 3, 2021.
Interactive Hardware
The XPeng P7 adopts the mainstream design of a central control screen and a dashboard with dual screens.
The dashboard size is 10.25 inches, and the central control screen size is 14.96 inches. There are physical buttons at the 3 and 9 o’clock positions on the steering wheel that can be easily touched by the thumbs of the left and right hands, and the sensitivity is high.
There is a protruding camera between the 12 o’clock position of the steering wheel and the dashboard, which indicates that XPeng P7 is equipped with DMS-related functions.
Most of the interactions that require physical contact occur in these few areas.
Visual Elements
The entire interface of the central control screen is dominated by cool colors. Under normal mode, it is in blue, white, and gray, while under dark mode, it is in black, gray, and blue. We can see that the color style of each level in the XPeng P7 central control screen is relatively uniform.
There are two benefits to doing this. On the one hand, a moderate cool color temperature is more suitable for the visual habits of Easterners at the physiological level, ensuring basic comfort. On the basis of aesthetics, the relatively small color layout helps to control visual fatigue in visual interaction.
Smoothness
The XPeng P7 is equipped with Qualcomm Snapdragon 820A as one of the representatives of the car-grade high-performance processors, providing hardware-level guarantee for the smoothness of the central control screen’s touch screen, which is close to the overall smoothness of mainstream tablets and leaves a deep impression.
Speaking of this, some friends who know the cockpit may ask, isn’t it popular to use Snapdragon 8155 and even further reserved 8195 now?
However, in fact, whether performance parameters such as calculation power can be intuitively reflected in daily use is not necessarily faster the higher it is-at least it will not be as exaggerated as changing from the traditional mechanical hard disk we used ten years ago to a solid-state hard disk with a read and write speed of several GB per second.
It Starts with Maps, But It’s More Than That
The main map resource of XPeng P7 is Gaode Map. The overall interaction logic is similar to the Gaode Map on mobile phones , so it’s easy to use it if you’re familiar with it. At the same time, the “Seamless Navigation” function that allows you to send addresses from your phone to the car’s head unit is also available, supporting apps such as XPeng, Gaode Map, WeChat, Baidu Map, Dianping, and Meituan.
In our evaluation system, we also include the sharing of location information among multiple platforms and users through maps, social, and lifestyle applications. In terms of cross-platform sharing, XPeng P7 currently performs well.
LBS (Location-Based Services) is a new concept introduced in this season’s evaluation, which introduces scenarios such as parking, refueling, charging, consumption, travel, check-in, and epidemic maps. XPeng P7’s in-car map performs well in presenting basic information such as parking, refueling, and charging, and can also associate peripheral scenic spots with travel destinations, which is helpful for trip planning.
However, the iconic team-up feature of Gaode Map is not available in XPeng P7’s in-car map, which may be regrettable.
We noticed that XPeng P7 supports the linkage between in-car map accounts and mobile phone accounts. After logging into the mobile phone account on the in-car version of Gaode Map, the frequently used places, search records, and other information of the latter will be synchronized with the in-car version together with the settings, which is convenient to use. However, changes made to the settings in the in-car map cannot be synchronously updated to the mobile phone, which may cause some inconveniences depending on the user’s personal usage habits.
It’s Just Like the OTA You Get Just in Time for That Last Exam Question
On the day when we drove this white XPeng P7 back, we received the new push of Xmart OS 2.6.1.
Apart from the changes in the assisted driving aspect, XPeng’s intelligent cockpit voice assistant “Xiao P” has improved the overall sound performance.After upgrading, Xiao P’s voice has significantly improved in terms of natural and fluent intonation as well as content expression.
It is reported that the voice part of XPeng’s intelligent cabin is composed of XPeng’s new generation of large-scale online neural network and small offline splicing engine. It is said that in the latest OTA upgrade, XPeng’s self-developed “end-cloud fusion multi-level cache” technology solves the problem of delay and network jitter.
As mentioned in our previous Hongmeng cabin topic, the product structure of “cloud + car dual voice engine” will become a trend. This not only ensures that the vehicle’s voice interaction ability will not suddenly disconnect, but also helps to improve the performance of intelligent voice.
Visible and audible: just a little bit more to go
For basic voice interaction such as navigation, music, and in-car settings, XPeng P7 can easily and accurately execute them. However, when it comes to third-party applications outside of the ecosystem, the most that Xiao P can do is to help you accurately find the corresponding application and open it, and follow-up actions still need to be completed through finger touch. The ability to achieve visible and audible through voice implementation does not fully cover all applications of this cabin.
Mijia onboard: a good start
As a relatively hot product concept in recent years, car-home interconnection has received a lot of attention from many manufacturers. Whether it is the well-known Huawei, Xiaomi IoT ecosystem, self-owned brands, or joint ventures, they are constantly exploring how to introduce a better car-home interconnection experience into the cabin. Our intelligent cabin evaluation system also includes multiple evaluation dimensions involving IoT fields.
XPeng P7’s cabin system has integrated Xiaomi’s Mijia App, which can control the startup and shutdown of home Xiaomi smart home appliances.
Unfortunately, the Mijia application on our P7 car end only supports on-off operations, and cannot perform more detailed operations and controls for each device card like on the mobile end. Nevertheless, on the way home after work, we can still tell “Xiao Ai” at home through Xiao P to turn on the air conditioner and start the sweeping robot. This way, when we arrive at home, we will be welcomed by a clean and comfortable environment.### Superb voice recognition? It’s all designed
AI voice interaction capability has always been an important part of the information released by XPeng P7 and even the entire XPeng brand. For the AI voice assistant in the intelligent cockpit, it is not only required to listen accurately and understand, but also to “do it right”.
It should be noted that all voice actions need to be performed under the dialogue state recognized by both the driver and the vehicle, which means that the voice control of XPeng P7 temporarily does not support wake-up-free operations such as play control and map zooming.
Free-wake-up, as a common voice interaction function, is actually very common in car machines and cockpits, but it is not seen in XPeng P7, which is known for its voice, which is somewhat inexplicable.
Here we do not define the pros and cons of such settings, although it seems somewhat inexplicable. As the interaction and evaluation deepen, we understand the reason behind it, and we will talk about it later.
We found that XPeng P7 supports “Hello XPeng, navigate to Beixinqiao” and other One-Shot-style voice interactions. When the wake-up word and command are connected to convey instructions to XPeng P7, the cockpit system can also respond quickly.
At the same time, we also noticed that XPeng P7 does not temporarily support recognition of pure English and dialects, but in situations where Chinese and English are mixed, such as “Navigate to Beijing SKP” and “Go to Galaxy SOHO,” XPeng P7 can accurately recognize and execute related actions.
In some specific voice interaction scenarios, such as telling a joke or asking how to make a certain dish, XPeng P7 will implement it by searching and calling external voice resources. At the same time, if XPeng P7 thinks it cannot do it, it will directly tell the user that it is currently not able to. This “proactive ending” mechanism avoids the embarrassment of staring at each other between the user and the machine.
If you interact with XPeng P7 in the driver or passenger seat in the front row, its response speed and quality will surely impress you. But if it is in the back seat, XPeng P7 seems to be less flexible. XPeng P7’s cockpit does not have a special explanation for recognizing multiple sounds, but only the front row supports double sound area recognition, which can be voice-controlled in the car.After analysis, we believe that the XPeng team intentionally put certain restrictions on the cabin voice interaction mechanism to prevent the sensitive voice from causing false awakenings and misoperations. We all know that frequent false awakenings not only greatly affect the user experience, but also make people doubt the “intelligence” of the AI assistant.
Just like our human philosophy of living, those things that can be said or not, most people choose not to say them to avoid trouble.
The emergence of internet power in traditional industries often means more radical technological and product paths. But in Mr. Yu’s view, the XPeng team is not radical in communicating the cabin voice experience to the consumer group, and even appears somewhat cautious, which includes effective management of user expectations.
Xiao P works hard to understand you
Even for us humans, understanding depends on being able to “hear clearly”.
Based on a new evaluation system, we have separately restored the three main scenarios of fully closed windows, one-third diagonal windows opened front and back, and maximum air conditioning airflow in the car, to test the wake-up rate and voice recognition accuracy of the cabin intelligent voice assistant in both static and dynamic modes.
The results show that even if the air conditioning in the vehicle is turned up to maximum in stationary state, Xiao P’s response accuracy rate to voice wake-up is still very high. Moreover, even higher levels of road and wind noise have not had too much impact on Xiao P’s response on the highway.
In the process of repeated voice interaction testing, Xiao P only failed to respond to our call in the high wind noise environment where the window was opened to about one-third. In addition, we noticed that when the air conditioning airflow in the car is at the highest level, there is a certain degree of delay in Xiao P’s response speed to the cabin voice, which occurs about 30% of the time. It is likely that in environments where external noise is high and voice commands are received, Xiao P will perform noise reduction processing to determine the user’s correct intention.
When playing music in the car, as well as interactive commands for broadcasting, cross talk, and other audio content, Xiao P had a probability of about 10% of misjudgment, manifested as command recognition failure; when the car is in a relatively noisy situation such as multi-person chat, coupled with other background noise, this probability will increase to about 30%.According to reports, the upgraded Xiao P achieved a score of 4.49 in the international standard for voice quality evaluation called MOS, which is the highest score achieved by an AI voice assistant, second only to humans.
It is known that the evaluation of intelligent voice assistants involves many dimensions and details. Under different evaluation systems, the results for the same product may differ depending on the methods and dimensions used. Setting aside the element of PR, it is currently difficult to say which aspect of intelligent voice evaluation is more superior.
However, one thing that can be certain is that the Xiao Peng P7’s performance in terms of voice-related experiences is excellent overall. If the Xiao Peng team can further adapt Xiao P to a series of applications within the ecosystem at a higher degree, it is believed that they will receive a higher evaluation in terms of user experience and aesthetics.
Despite a good product, there are still some issues that need to be addressed. Here, we will discuss some points worth reflecting on in each evaluation period.
Hardware Aspect
-
Center console: Although the touch performance of the center screen is smooth and natural, when some operations such as dragging the progress bar of some video applications are performed, the large bulge of the center console under the screen causes the operation to feel awkward and sometimes gets stuck. Fortunately, the Xiao Peng team did not design any blind operation interaction functions for the center screen, otherwise it would inevitably affect the experience.
-
Steering wheel control: Similar problems also occur on the steering wheel. It is unclear whether it is due to the “meaty” hands of the editors who participated in the evaluation, but when they rotate the steering wheel left and right, which requires a larger range of motion, they accidentally touch the “voice wake-up” and “mute” buttons located at both ends of the steering wheel control. Of course, this is the result of a combination of personal physiological characteristics, habits, and design. We also believe that this small problem can be solved through details adjustments or shielding in industrial design.
Interaction Aspect
First of all, it needs to be acknowledged that the introduction of Alipay Mini Programs and the integration of a large number of third-party applications have brought more diverse usage scenarios to the Xiao Peng P7’s cockpit. However, we have also encountered some issues in actual use:1. Apps from the first party under XPeng ID have all been integrated, with the interaction level basically controlled within two layers, making them very user-friendly. However, the interaction framework of XPeng’s cockpit system differs from that of Alipay mini-programs and third-party applications. This leads to a disjointed user experience. Although the touch response speed of the central control screen is fast, the user experience of the aforementioned three parties has not been leveled. In addition, different application systems still maintain their own return mechanisms, resulting in a certain degree of fragmentation.
- There are multiple scenarios in the mini-program that require entering phone verification codes, but the area for users to input the codes is very small. Combining with the first problem above, we believe that both in terms of visual elements and operational logic, this part has not been fully system-adapted.
- When multiple tasks are happening in parallel, the system doesn’t plan the execution priority well. It simply interrupts and executes based on the order of the confirmation command. For example, if the user sends out a command to play music when watching a TV show through a third-party video application, the video playback window will enter a black screen state, and the background will start playing music, but the system won’t switch to the music player interface, instead retaining the original screen.
The experience of XPeng P7 in the map and voice modules is excellent. We can see that XPeng has almost scored full marks in the functional experience of these two modules, which indicates that the product team of XPeng has put a lot of effort into optimizing various details and experiences in the basic functions.
But it doesn’t mean that the map and voice modules have no room for improvement. In terms of functional completeness, we can see that some useful map and voice functions are not included in the function list of XPeng P7, such as teaming up in the maps, dialect recognition in the voice, commonly used commands without voice wake-up, etc.
In terms of the ecosystem, XPeng P7 does not provide connectivity between mobile phones and car machines. This shows that they have absolute confidence in their car machine, but they will also lose some habitual users of Apple CarPlay or Baidu CarLife.## 小鹏汽车的挑战和机遇
Small and medium-sized start-up companies such as XPeng Motors are aiming to be disruptors in China’s auto industry. As a translator for the automotive industry, my task is to accurately convey the content in the source language. In addition, I need to polish the wording, correct any errors in spelling and grammar, and make improvements wherever possible to produce a smooth and professional English version of the original.
In addition to the usual entertainment offerings, XPeng has fully integrated the Alibaba mini-program ecosystem, which covers almost all consumer, work, study, and social scenarios. However, XPeng has not customized the UI logic of these mini-programs like the cockpit system. At the same time, how to match these application scenarios in the car, and how to make these mini-programs truly usable in the car, is also a problem that every automaker must face.
The audio system and DMS of the driver’s headrest as innovative features have caught our attention, but how DMS interacts more with the hardware in the car and provides active services is the next issue that XPeng needs to consider.
In terms of IoT and V2X, like most car companies, it can only achieve the level of having this function, and it is not possible to talk about the experience. In addition to the intelligent car machine, how to truly achieve an intelligent “cockpit” requires more innovation and exploration from the entire industry. We hope to see more independent audio zones, headrests, DMS, and configurations like this on the car, and truly grow the intelligent car machine into an intelligent cockpit.
Having dealt with intelligent voice assistants for several years, Mr. Yu is no stranger to how to communicate with AI voice. We can confidently say that XPeng P7’s intelligent voice experience is unprecedented:
One-wake-up, multi-round interaction; quick response, precise execution;
What can be done will be quickly and accurately implemented, and what cannot be done will not be maliciously acting cute or jokingly…
In every aspect related to voice interaction with XPeng P7, it meets our usual imagination of the role of “human-machine interaction port” in the car scenario.
At the same time, we can also see that XPeng P7 has a high degree of application ecosystem integrity. The access of the Alibaba application ecosystem and Alipay mini-programs basically covers the scenarios we use in our daily life.
For example, for frequently used apps like “Ele.me”, because the interaction logic of the car screen itself has not been fully considered, the convenience of its use is not as easy and fast as using our own mobile phone by muscle memory.
Speaking of which, the appearance of some applications in the cockpit system is more like “having” than allowing people to “use”. Just like the Bilibili app “Yun Shiting Xiao Dianshi” on the car machine, which emphasizes the immersive video application, like the TV terminal application, can provide users with content and functions that are far less rich than the mobile terminal.
Of course, these are not problems with XPeng P7. However, at the same time, this phenomenon needs to be given joint attention by the industry.
It should be emphasized that people are willing to pay for intelligent cockpits not only because they look fun.
Thank you for reading this far.>In an era where everything emphasizes being “straightforward”, it is invaluable to be able to discard impatience and anger, and immerse oneself in the space of words and imagination.
In the industry group of intelligent cabins of GeekCar, there have been many practitioners, media colleagues and friends interested in intelligent cabin products to discuss industry trends, product possibilities and creative ideas inside and outside the industry.
As Mr. Yu said earlier: it’s better to chat with people thousands of times than to read thousands of books.
If you also want to join us and feel the collision of ideas and creativity in communication, you can add our WeChat by scanning the code and meet the people with similar wavelengths here.
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.