Author｜Lu Yao

Over the past few decades, tech giants have spared no effort in investing in the research and development of smart wearable devices, among which smart glasses are one of the most popular categories.

Many industry pioneers even regard smart glasses as an excellent carrier of AI technology, believing that they not only have the functions of traditional glasses but also incorporate advanced technologies such as speech recognition, natural language processing, and machine vision, providing users with more intelligent and convenient services.

However, despite being fiercely pursued by Meta, Google, Apple, and other major companies, smart glasses have not yet become an indispensable part of people's daily lives. There are many reasons for this, such as excessive weight affecting wearing comfort, poor battery life, and practicality and technology levels that are far from satisfying everyone.

In the first half of this year, Meta's glasses with new AI functions brought a new breakthrough to the entire industry. However, similar to most similar products on the market that are currently connected to large foreign models, this high-tech product connected to Llama 3 is not user-friendly enough for domestic users, especially Chinese users.

This to some extent reflects the awkward situation of the industry: even though the AI technology boom is surging, there is still a lack of AI glasses on the market that are truly suitable for Chinese usage scenarios. However, Baidu Duer AI glasses unveiled at Baidu World 2024 may break this stalemate.

At that time, the existing market landscape may change.

01

Chemical reaction between hardware products and AI

The discussion about whether glasses will become the most important smart device in the future can be traced back to 2012 when Google launched Google Glass. With functions such as taking photos, navigation, and information push, AI glasses were brought into the public eye and were once considered a possible replacement for mobile phones.

However, despite the promising vision, this product was discontinued in 2015. Many users complained about its lack of wearing comfort, inconvenient operation, and high price, which deterred most people. After the setback of Google Glass, AI glasses entered a relatively stagnant stage of development, but people's exploration of smart devices never stopped.

Traditional smart terminals such as computers and mobile phones are not entirely designed for AI. With the booming AI industry in recent years, people have increasingly realized that the increasingly mature AI market lacks a physical carrier similar to a mobile phone.

Speaking of AI, Baidu Duer has had AI genes since its inception. From smart speakers, smart screens, to smart study machines and buddy machines, based on conversational AI voice interaction, its different types of smart hardware have covered over 46 million household users.

And this time, Baidu Duer's release of AI glasses also brings a new signal: from household scenarios to wearable scenarios, the further expansion of the product line means that Baidu Duer has also begun to explore the creation of more innovative and interactive smart devices and seek more scenario integration applications.

However, while AI glasses bring greater imagination, they also mean greater challenges.

Currently, in addition to "smart glasses with AR functions" and "AI glasses without a display screen," there is also an increasingly popular category, "audio glasses," in the industry. But the latter is functionally closer to traditional headphones, focusing mainly on the audio experience, and they do not have the complex AI processing capabilities of Baidu Duer products, so they are not true "AI glasses" in the real sense.

Taking Meta's Orion and Ray-Ban Meta as typical cases, the former adopts the most cutting-edge AR solution, but this amazing futuristic product is currently limited to the prototype stage due to issues such as the industrial chain and technology costs. In contrast, the latter still has the appearance of ordinary glasses, and after technological updates and iterations, although it does not have display functions, it can take photos, translate, and engage in dialogue, and sales have surpassed 1 million.

Therefore, the emergence of Ray-Ban Meta is more like a product that balances technology and demand. This just proves the growth law of disruptive products or industries: it does not happen overnight but requires time and market polishing.

The same evolutionary thinking is also reflected in Baidu Duer.

After observing the current state of the industry and consumers' demand for portable smart assistants, Baidu Duer's products adopt a "no-display AI glasses" design. The benefits of this approach are obvious:

On the basis of maintaining user habits, people can touch a new form of interaction without bearing the additional weight of display capabilities; manufacturers are also exempted from the high pressure of technological costs and can popularize a product in the mass market, continuously innovating through user feedback and technological iterations.

Therefore, from the perspective of configuration specifications alone, this new product does not have the main purpose of "showing off hardware" but focuses on the application of AI technology on hardware to enable users to better utilize AI.

02

To make AI glasses, we must first make good AI

For a true AI hardware product, especially an emerging category like AI glasses, the core is not just the hardware alone but the deep integration of software and hardware. Specifically, the hardware provides basic computing and sensing capabilities, while software, especially large AI models, endows these hardware with more "intelligence."

For Baidu Duer's AI glasses, the most direct point is that they are equipped with a large Chinese model.

In this way, it not only solves many shortcomings of previous smart devices in Chinese processing, such as inaccurate semantic understanding and complex Chinese contexts. The DuerOS operating system reconstructed based on the large model also brings users a smarter and more natural interactive experience.

In addition, as the key to enhancing the user experience, end-to-end optimization involves the entire process from data collection, transmission, processing to presentation. Therefore, manufacturers often need to carefully weigh which operations are suitable for completion on end devices (such as smart glasses, smart speakers, etc.) and which are more suitable for processing in the cloud based on the specific needs of tasks, such as real-time requirements and computational load.

From the perspective of optimal resource allocation, in order to improve efficiency in different application scenarios, Baidu Duer's end-to-end design provides a more refined solution: it not only meets the real-time requirements of data processing but can also flexibly adjust the reasonable allocation of computational resources.

Essentially, this is out of consideration for product adaptability and practicality.

It is worth mentioning that in terms of functional applications, unlike directly calling large model APIs, Baidu Duer's AI glasses adopt a "software and hardware integration" product concept during development. How should this be understood?

Generally speaking, direct calls are of course simpler, and integration and deployment are faster. However, due to the reliance on external interfaces and third-party capabilities, devices are greatly limited in terms of functional innovation and personalized customization. Moreover, the lack of deep integration of software and hardware can easily lead to performance issues such as slow response speeds, low data processing efficiency, and unstable devices, making it more difficult to achieve a truly intelligent interactive experience.

The so-called "software and hardware integration" is actually collaborative design between software and hardware. Typical examples are ecological companies like Apple and Xiaomi, which have the ability to simultaneously integrate software and hardware resources to achieve close cooperation between hardware design and the operating system.

Reflected in AI glasses, on the one hand, the design of the hardware can fully consider the operational requirements of the software, and on the other hand, the software can also be optimized in combination with specific hardware architectures. For example, for components such as the camera, microphone, and sensors of AI glasses, the integration design and optimization of visual recognition algorithms and interactive experiences can improve the overall performance of the device.

This approach is ingenious.

For users, the integrated design facilitates product quality control and optimization from the system level, making the multimodal interactive experience more natural and smooth. For Baidu Duer, software and hardware integration can also provide greater innovative extension opportunities for products, enabling the continuous integration of more advanced technologies and achieving self-research breakthroughs.

On this basis, the next step to consider is how Baidu Duer's rich application ecosystem, combined with the integration of Baidu Maps, search, and encyclopedic capabilities, can be better reflected on this AI glasses?

For example, as a first-person perspective device, with the input of rich information such as vision, sound, and location, with the help of the reasoning, understanding, and generation capabilities of large AI models, there are undoubtedly huge application scenarios and imagination space. The abilities such as "asking while walking" and "object recognition encyclopedia" demonstrated by Baidu Duer at this conference are precisely like this.

From seeing the world to understanding the world is the true value of AI glasses.

03

Be optimistic, maybe the future is already here

In fact, for those who have been paying close attention to the technology sector for a long time, AI glasses are not exactly a new concept. However, for a long time, due to factors such as technology and market acceptance, they have indeed mostly remained at the conceptual level.

With the continuous maturity of technology and the continuous expansion of application scenarios, AI technology is gradually moving from the "proof of concept" stage in the first half to the "large-scale application" stage in the second half. The logic behind this is to make technology accessible to everyone, which is becoming the epitome and new starting point of industry consensus.

Changes in AI glasses are also happening gradually.

The BIS Research smart glasses market research report precisely validates this point. In 2023, global smart glasses sales were 1.01 million units, and shipments in the first quarter of 2024 increased by 217% year-on-year. It is estimated that by 2029, the global smart glasses market will be close to 106.8 billion yuan.

Taking the most basic hardware specifications as an example, many people were not optimistic about AI glasses in the past, and one of the most intuitive reasons was that they were too heavy and uncomfortable to wear. Although Baidu Duer AI glasses have not yet been officially launched, based on the current disclosure, it will indeed be a smart wearable device worth looking forward to.

First of all, it weighs only 45 grams, which is lighter than the industry's mainstream level, and the temple size is also smaller, making it basically the same as ordinary glasses for daily wear.

In terms of camera capabilities, it is equipped with a 16-megapixel ultra-wide-angle lens that supports shooting up to 4656*3496 pixels, surpassing many similar products, including Ray-Ban Meta.

There is also the issue of battery life that everyone cares about. According to reports, Baidu Duer AI glasses can support continuous music listening for over 5 hours, far exceeding the industry average of 3 hours, and can be fully charged in just 30 minutes.

As a device that can be worn all day long in front of your eyes, it can listen to your call at any time. Compared to taking out a mobile phone or turning on a computer, AI glasses have become a more targeted first terminal for multimodal perception technology.

Changes in human-computer interaction forms and continuous technological accumulation in the software and hardware fields have made people gradually believe in the future described by technology companies. When AI glasses cross the thresholds of wearing comfort and practicality, they may indeed become a more suitable carrier for AI than mobile phones and even have the potential to be the primary entry point for human-computer interaction.

In this regard, Baidu Duer has indeed taken an important step forward.