Will the sweeping robot abandon its original partner, laser radar?

It is no secret that robot vacuum cleaners use machine vision technology. Ecovacs finally released its first robot vacuum cleaner, the DG70, which is equipped with AIVI vision technology at the beginning of the year. Xiaomi's second-generation robot vacuum cleaner (1S) also uses visual navigation. Manufacturers are following suit and carefully adding this "new" technology, or "new" concept.

In the field of machine vision technology, there are many categories, including structured light and ToF, which have been widely used on mobile phones in recent years, and binocular vision, which is widely used on drones. What is the standard configuration of this wave of machine vision for sweeping robots?

In the field of machine vision, monocular vision and binocular vision are currently two types of positioning technologies that have attracted much attention. The huge differences between the two in structure and working principles have also led to different market application statuses.

Binocular vision positioning refers to the positioning of the three-dimensional spatial position of the target point through the principle of triangulation. The binocular vision positioning algorithm process includes camera calibration, binocular positioning, image processing, feature detection, stereo matching, three-dimensional measurement and posture measurement.

Binocular vision has high positioning accuracy, low entry threshold, easy 3D composition, and can also achieve passive light source positioning. Therefore, there are many companies in the industry that are engaged in binocular vision, including Tuyang Technology, Zongmu Technology, and Yunguan Technology.

Despite this, "in practical applications, binocular vision still has certain limitations," Zhou Kun, founder and CEO of Shenzhen Huanchuang Technology, told Leifeng.com.

As a master's tutor at Tsinghua University Shenzhen Graduate School, Zhou Kun has conducted in-depth research on visual human-computer interaction technology. Currently, Huanchuang Technology, where he serves as CEO, mainly provides monocular visual spatial positioning technology. According to Zhou Kun, he previously founded Taishan Online and conducted in-depth research on binocular vision technology. Zhou Kun believes that binocular vision has the following three main shortcomings:

• The module is large. The binocular vision positioning accuracy is affected by the baseline (the distance between the two cameras). The longer the baseline, the higher the measurement accuracy. Therefore, binocular vision is rarely used for positioning in products with high space requirements such as mobile phones.

• Higher cost. Compared with monocular vision technology, binocular vision technology is more expensive;

• Small effective viewing angle. Compared with monocular vision technology, the effective viewing angle of binocular vision is smaller.

Relatively speaking, Zhou Kun is more optimistic about the application prospects of monocular vision technology.

Monocular vision uses only one camera to complete the positioning work. There are two main methods of monocular vision positioning: positioning method based on single frame image and positioning method based on two or more frames. Among them, the most widely used methods are the feature point positioning (PnP principle) method based on single frame image positioning method and the auxiliary measurement method of adding IMU inertial sensor.

As SLAM technology has gradually matured in recent years, the application scenarios of visual technology have become more and more extensive, especially in indoor scenes, where monocular visual positioning technology is widely used.

In the field of sweeping robots, monocular vision technology is currently more widely used. The new sweeping robot products of Ecovacs and Xiaomi mentioned above have added a visible light camera for auxiliary positioning to better achieve obstacle avoidance. iRobot, which was one of the earliest foreign companies to apply visual navigation, also uses a monocular camera to continuously track images during movement, estimate the camera posture and relative position changes with objects, and thus build a map.

In consumer applications, sweeping robots are increasingly equipped with machine vision; at the same time, in industrial and medical applications, corresponding manufacturers are gradually beginning to experiment and explore.

Although monocular vision technology has certain advantages in size and cost, it is not as good as binocular vision in key positioning accuracy. In order to improve the positioning accuracy of monocular vision, auxiliary positioning technology has been introduced in practical applications.

The first commonly used auxiliary positioning technology is to add marking points to the object being measured in a controlled environment.

For industrial inspection, medical and other controlled environments, monocular vision is based on the PnP principle, and multiple markers are added to the measured object to improve positioning accuracy. For example, in surgical navigation in the medical field, specific markers can be added to the scalpel used by the doctor to greatly improve positioning accuracy.

The second auxiliary positioning technology is: introducing multi-sensor fusion technology and adding IMU inertial sensors.

By collecting the acceleration and angular velocity of the IMU inertial sensor (gyroscope, accelerometer, etc.), the motion mode of the object being measured is determined, and the image collected by the monocular camera is combined to achieve the final positioning through algorithms such as vision-based feature point matching and sparse optical flow. In addition, the introduction of IMU inertial sensors also makes it possible to perform 3D composition through monocular vision technology.

Monocular vision, assisted positioning..., machine vision is so popular, will the laser radar that is originally equipped with sweeping robots really be gradually replaced?

Before answering this key question, it is necessary to understand the "in-law" relationship between machine vision and lidar.

Before the rise of SLAM navigation and positioning technology, LiDAR had long been the original equipment of sweeping robots. With the rise of visual navigation technology, machine vision has attracted much attention and has become a navigation technology different from LiDAR.

" In fact, the short-range lidar currently used in the robotics field is also made with monocular vision sensors, " Zhou Kun told Leifeng.com.

In the field of robotics, both the vSLAM machine vision and LiDAR that we are talking about nowadays are actually machine vision positioning technologies. There is little difference between the visible light camera in the vSLAM solution and the infrared camera in the LiDAR solution. However, since LiDAR uses active light source positioning technology, its positioning is more accurate; the monocular visible light camera uses passive light source for positioning and navigation, and its accuracy is lower. Therefore, it is currently recognized in the industry that LiDAR is more suitable for positioning and navigation.

From this perspective, the machine vision and lidar used in sweeping robots are more like a "relative" relationship. The often-mentioned machine vision uses visible light cameras, while lidar uses infrared cameras.

What is different is that machine vision (visible light camera) has been mentioned repeatedly. In addition to its positioning and navigation functions, it is more about its ability to achieve 3D composition and the many application expansions brought about by future image recognition.

Will LiDAR be replaced by machine vision?

Especially in the application of sweeping robots, is this technology iteration or technology integration?

At present, it still belongs to the latter. Including the two domestic robot vacuum cleaner sales giants mentioned above, they still use vision to avoid obstacles and assist the entire navigation system.

Currently, 2D vision is still achieved using visible light monocular cameras. However, due to the lack of active light source, the positioning accuracy is poor, and the quality of the constructed maps and trajectory routes is relatively poor. The use of ToF, monocular structured light, and binocular vision can easily achieve 3D modeling and can be used for obstacle avoidance. However, when using SLAM algorithms, the efficiency will be much lower due to the narrow horizontal field of view (less than 90 degrees).

It can be seen that it will take some years for machine vision to completely replace LiDAR. Zhou Kun also believes that the development direction of sweeping robots is more of a fusion trend: standard LiDAR is used for SLAM algorithm and navigation positioning, while machine vision is used for obstacle avoidance, 3D vision, deep learning and other corresponding functions.

This also means that the demand for laser radar in the sweeping robot market, which is worth millions of units, is still considerable. "Our overall business focus this year is in the sweeping robot field, which is expected to account for more than 90%. Among them, Camsense X1 (laser radar) is our main product this year," Zhou Kun told Leifeng.com about Huanchuang Technology's product plan this year.

Huanchuang Technology started developing high-precision positioning algorithms in 2015. The algorithm mentioned here is Camsense HPP (High Precision Positioning), an algorithm based on infrared optical spatial positioning and recognition. According to official information, Camsense HPP can achieve sub-millimeter spatial positioning resolution within a range of 5m x 5m, which is suitable for short-distance precise positioning in indoor environments such as precision equipment measurement and medical surgery navigation.

This algorithm is also the core technology of Huanchuang Technology.

In addition, Zhou Kun told Leifeng.com that compared with the laser radars currently on the market, Huanchuang Technology has made many improvements to the light source, lens, and image sensor. On the one hand, it has improved the performance, and on the other hand, it has avoided the patent risks in traditional triangulation methods.

"Compared to Asia Optics' use of industrial-grade linear image sensors for LiDAR, Huanchuang Technology uses customized area array image sensors." The improvements brought about include:

Compared with industrial-grade image sensors, customized area array image sensors are less expensive;

It uses an area array image sensor, which has a larger vertical field of view than a linear image sensor.

According to Zhou Kun, "The vertical field of view of current industrial-grade linear image sensors is generally 0.15 degrees, while the vertical field of view of our customized infrared sensor lidar can reach 40 degrees. This also makes it easier to calibrate and focus the light source during the production process."

Laser radar is the current focus of Huanchuang Technology, and it is also an important product after it shifted to the fields of robotics, industrial inspection, and medical applications. Prior to this, Huanchuang Technology's monocular vision technology had also been used in the fields of television, VR, etc.

Shenzhen Huanchuang Technology Co., Ltd. was formally established in 2014. At the beginning, Huanchuang Technology positioned itself as a traditional visual sensor and first applied it to somatosensory games in the TV field at that time. "At that time, except for TCL, almost all domestic TV manufacturers were our customers. We provided them with monocular cameras to support somatosensory games in home entertainment."

Why give up the TV market?

Zhou Kun said frankly, "We are also well aware that there is a ceiling in this field. Later, we further expanded our market to the VR field which was popular at that time." In 2016, Huanchuang Technology won the bid of iQiyi, providing spatial positioning system for its VR equipment, and successfully completed the development of technical products. However, as the domestic VR market ultimately failed to develop as expected, this product ultimately failed to enter mass production.

According to Leifeng.com, until now, the domestic VR market has been lukewarm, and there have been no well-known devices such as Microsoft HoloLens and HTC Oculus.

However, during this period, Huanchuang Technology developed its M series products, Camsense M1 and Camsense M2. Zhou Kun told Leifeng.com (official account: Leifeng.com) that until now, although these two products for the TV and VR fields are no longer iterated and updated, they are still being shipped steadily. So far, the product shipments have reached 150,000 sets.

For technology companies like us, it is important to choose the field in which the technology will be applied. After several years of technology accumulation, in 2017, we officially switched our business areas to robotics and industrial inspection.

Subsequently, Huanchuang Technology won the bids from COMAC and Huachuang Group, and developed the Camsense M Pro and Camsense S series products for industrial inspection.

"During the aircraft manufacturing process, there are many workstations required, such as riveting. The riveting process is achieved by a robotic arm, which requires precise guidance and positioning during the movement process. The Camsense S customized solution we provide to COMAC is to provide navigation and guidance for the robotic arm." Previously, this type of high-end manufacturing industry was mostly monopolized by foreign countries. The benchmark of Huanchuang Technology's Camsense S is Canada's NDI company. Camsense S can already achieve a positioning accuracy of 0.2mm.

As mentioned above, in 2017, after entering the second phase, Huanchuang Technology began to expand its business into the field of robotics. At the same time, the field of robotics has gradually explored how to better apply machine vision to make up for the shortcomings of existing robots in obstacle avoidance, and tried more image processing fields.

Zhou Kun believes that the current research and development focus of the robotics field, especially the sweeping robot field, is as follows:

• Reduce the cost of SLAM navigation and positioning solutions. "Lidar is still the best configuration for SLAM navigation and positioning. How to make the Lidar SLAM solution more cost-effective and how to develop better algorithms are currently a major issue in the industry."
  

• Solve the common obstacle avoidance and object recognition problems. Many robot manufacturers have begun to introduce machine vision to provide corresponding solutions.
  

• SLAM + obstacle avoidance fusion solution. That is, machine vision + lidar fusion solution, "This is the focus of the current industry and an important research and development direction of our company."

Since its establishment in 2014, Huanchuang Technology has entered its fifth year.

In terms of technology, we still focus on monocular vision. After jointly establishing a high-precision positioning laboratory with Tsinghua University, we established an optical laboratory in February this year to provide a professional R&D and testing site for optical high-precision positioning systems. So far, we have more than 20 R&D personnel.

In the market, it has evolved from the initial fields of television and VR to the current fields of industry, medical care, and robotics.

In addition, Zhou Kun also particularly emphasized the issue of mass production.

For consumer-grade LiDAR, mass production is a major challenge. Huanchuang Technology has its own trial production workshop, where all products are trial-produced before mass production. Since the establishment of our own trial production workshop in 2018, we have trial-produced 5,000 products. These products are only used for trial production before large-scale mass production and will not be sold as final products.

According to Zhou Kun, “Huanchuang Technology will achieve profitability this year.”

In the medical field, Huanchuang Technology has already had interested manufacturers. This year's main product is the Camsense X1 laser radar, with a target shipment of more than 500,000 units; the target market is sweeping robots, which are expected to account for more than 90% of the overall business this year.

Zhou Kun told Leifeng.com that Huanchuang Technology's laser radar currently has three types of customers:

The first category is the whole machine manufacturers of sweeping robots, such as 360, which have the ability to develop SLAM algorithms themselves;

The second category is solution manufacturers, who develop SLAM algorithms based on Huanchuang Technology's LiDAR and then provide complete SLAM solutions to device manufacturers;

The third category is ODM manufacturers, who develop SLAM algorithms based on Huanchuang Technology's lidar, and then develop complete machine solutions for mass production.

"We currently have more than ten cooperative manufacturers in the three categories."

In the fifth year of Huanchuang Technology, LiDAR has become the focus and sweeping robots have become a betting area. As for the current situation of sweeping robots in China, machine vision has just emerged and LiDAR is still the original equipment, which is still a key area that the industry continues to pay attention to.

Apple released new financial report, and was "dragged down" by the Chinese market again

Alibaba’s ten-year AI boss started by “liberating” porn appraisers

The China Securities Regulatory Commission launched an investigation into LeTV and Jia Yueting; Momo responded to the removal of the Tantan App; Luo Yonghao announced the listing of Xiaoye e-cigarettes

Luo Yonghao admitted defeat

Apple fights against phone addiction; former DJI employee fined $200,000 for leaking company source code; Musk reaches settlement with SEC: lawyers must approve tweets