30% is technological innovation, 30% is product reliability, and the remaining 40% is mass production. "Christopher said

"After solving the mass production problem, do you feel more relaxed?" the reporter asked. Christopher pursed his lips and said: "At this stage, the annual mass production of hundreds of thousands of units has been solved, and there will be new things to solve in the next stage. After all, the problems encountered in the mass production of 500,000 units and 1 million units are again It will be different." Recently, an article "Shenzhen Doctors on the Assembly Line" has been widely circulated on the Internet, opening up the story of scientific research doctors who need to go to the front line of manufacturing to start a business. After half a year, the reporter interviewed one of the protagonists in the article - Christopher, the founder of Camsense, to hear his story in the process of mass production of lidar.

Following the trend of "service robots", aiming at lidar

In recent years, driven by the rise of the lazy economy, the help of consumption upgrade trends, and the drive of smart home outlets, my country's sweeping robots have ushered in a developing market that has begun to take shape. Christopher, who is deeply involved in the field of computer vision and spatial positioning, has also expressed his optimism about the sweeping robot market on various occasions.

According to the data from CMM, the sales volume and sales of China's sweeping robots doubled from 2013 to 2017, and maintained rapid growth, achieving a doubling of development within 5 years. However, compared with Europe and the United States, the penetration rate of China's sweeping robots is still low, and it still faces a broad market in the future.

The sweeping robot has undergone three technological changes, and is divided into random and planned according to the cleaning path. The user experience of the first generation of random walking robot vacuum cleaners was extremely bad, and the market potential was not stimulated until the emergence of intelligent navigation robot vacuum cleaners.

Christopher believes that the upgrading and upgrading of sweeping robots is imperative, and will eventually develop from random blind-collision sweeping robots to intelligent, navigation sweeping robots, while standard lidar sweeping robots will become the mainstream of the market.

It's not hard to understand how lidar, which makes measurements using geometric principles, works. The light source emits a laser beam, and the beam returns to the image sensor after hitting the object. The radar module can calculate the distance between the transmitter and the object by using the pre-calibrated geometric parameters based on the imaging of the transmitted and received signals on the image sensor. .

During the entire measurement process, with its high range resolution and angular resolution, lidar can feed back rich information such as distance, angle, and reflection intensity, and then generate multi-dimensional images to assist the system in understanding the detection target in detail.

About 50 lidar startups have raised more than $1 billion in investment in the past three years, including $420 million in 2018, according to data compiled by Reuters. It is not difficult to see that lidar has become a fertile ground for practitioners. In the process of exploring business scenarios, Camsense also targeted this field early.

After two years of exploration and preparation, Camsense's lidar product line was mass-produced and shipped last year, and has become the exclusive radar supplier for many robot customers at home and abroad. "Our overall business focus this year is in the field of sweeping robots, which is expected to account for more than 90%. Among them, Camsense X1 is our main product this year." Christopher said.

Compared with the expensive market purchase price of most sweeper manufacturers, through technological innovation and improvement, Camsense's lidar cost has been greatly reduced.

"Like other electronic products, lidars have low and high seasons. We generally produce to order. The shipments in the low season are about 30,000 units, and the peak season will reach 100,000 units or even higher." Christopher said.

Open the market, evaluation and reliability are indispensable

Under the shackles of high cost, lidar was once considered to be only capable of showing off, and could not be mass-produced. Its cost mainly consists of material cost and production cost.

At present, most of the mainstream robotic lidars on the market use industrial-grade linear image sensors, which have a high sensor frame rate, but the high price and cost limit the popularity of lidars in the field of robotics. In addition, its vertical field of view is only about 0.15 to 0.16°, which requires extremely high assembly accuracy and stability, and large-scale mass production is challenged.

"During use, product drop, external vibration, and slight deformation caused by high and low temperature will affect the stability of the product and increase the after-sales cost virtually." Christopher pointed out.

In addition to sensors, another source of high cost is chips.

The lidar works in a rotating scanning manner, one light source sweeps a circle, and a 360-degree horizontal annular line of sight is obtained. The more light sources, the denser the line of sight, and the higher the resolution of the live map. Increasing the number of lidar lines in the traditional way will multiply the product volume and increase the difficulty of mass production. In order to reduce the volume of lidar, special chips with strong computing power and high efficiency are required.

For cost considerations and to avoid the "neck stuck" phenomenon, Camsense uses a custom area array consumer-grade sensor and a self-developed LiDAR ASIC dedicated chip Camsense C1. The cost of its area array consumer-grade sensor has reached the consumer-grade level, and the vertical field of view can reach 30-40 degrees. At the same time, it has built-in image processing algorithms with independent intellectual property rights, which can make the frame rate reach more than 2000Hz.

In recent years, many sensors are taking this route, that is, using ordinary consumer-grade components + unique algorithms, so the cost is reduced all at once.

"We have been preparing to make our own chip since the beginning of 2017, and it was successfully developed and launched into the market at the end of 2018." Christopher said that the frame rate of the lidar used by other brands can actually reach 4000Hz, but it is limited by the back-end computing. Capability and efficiency, only reached 1800Hz.

In order to create a product with high market acceptance, light parity is not enough, and reliability is also crucial. Camsense has designed very strict R&D testing standards to ensure product stability in thousands of experiments.

For a simple example, when a sweeper is working outdoors, from the perspective of the visual sensor, the camera is extremely susceptible to interference from background light or strong light, causing the signal to be flooded by sunlight, and the ranging function completely fails. Therefore, Camsense's test includes a high-intensity anti-glare interference test to ensure that the product can work normally at noon outdoors in summer.

Other tests include long service life tests, harsh high and low temperature storage tests, rigorous vibration, drop tests, and high-strength dust resistance tests.

Before a product goes on the market, it will go through several stages such as R&D, testing, trial production, and mass production. It is reported that Camsense has its own trial production workshop, and all products will be trial-produced in the trial production workshop before mass production.

"Since the establishment of our own trial production workshop in 2018, we have trial-produced nearly thousands of products before and after. These products are only used for trial production before large-scale mass production, and will not be sold as final products, and will eventually be scrapped." Christopher said .

After talking about R&D, testing and trial production, let’s talk about the challenging mass production link.

Mass production is not easy, strict control of supplier selection and production process

Standing at the forefront of technical theory and putting the theory into production is not an easy hurdle to cross. Along the way, Camsense has also experienced various tests on the road to mass production of lidar.

The biggest challenge during this period was the standard establishment process.

At present, the lidar industry has not yet formed a unified standard in the industry. Most customers do not know which threshold ranges should be set for the service life, impact resistance, high temperature and low temperature resistance and other parameters of the lidar products they need. Generally, they wait until the products are shipped and customers will provide feedback after encountering problems. .

Once and again, the cost of production and commissioning is particularly large.

"Because of the lack of industry standards in the early stage, there is no unified judgment rule for what is qualified and reliable lidar. Fortunately, we have contact with many customers, and in the early stage, we mainly focus on industry brand customers and large solution companies. The richer experience will give us this and that kind of requirements,” Christopher pointed out.

"At that time, our approach was to collect the standards of each customer, take the highest requirements as the basis, and then raise them on this basis. After repeated testing and iteration of actual mass production, we finally formed our own corporate standards. At present, we Many of the clients we serve have asked us to provide their own inspection standards as their inspection standards.”

The second major barrier is the supply chain management problem that is prevalent in the field of hardware manufacturing. Most of the physical manufacturing industry has a process of ramping up production capacity, especially at the stage of starting from 0.

Christopher told reporters that when the order volume is relatively small, large-scale foundries simply do not look down on these orders. In the early stage, they can only rely on their accumulated contacts in the industry to find suppliers to provide them.

Speaking of which, Christopher has brought the rigor of scientific research into full play in the selection of suppliers and the control of production processes.

In the past year, in order to find the most suitable supplier, Christopher traveled almost all over the Pearl River Delta, communicated with the heads of the various factories, went to the factory to see the production line, and did not dare to slack off.

Christopher said that after many improvements, Camsense has concluded a complete set of management and control processes. It has a strict supplier import process and high-standard quality assurance agreement, follows strict IQC inspection standards, and even controls the accuracy of screws.

In the process of process control, Camsense has formulated detailed operation instructions, and set up strict factory inspections and penalties for illegal operations.

For example, in a glue dispensing process, it is necessary to check the amount and position of glue; screws at different positions of the radar need to use electric screwdrivers with different torques, and workers need to wear electrostatic wristbands and electrostatic clothing.

According to reports, all Camsense products will undergo high temperature, low temperature shock test, vibration shock test and aging operation test. A series of processes can ensure high pass-through rate and low market repair rate.

"Through technological innovation, including sensors and chips, we have reduced our costs by 40%; at the same time, our strict quality control system has increased our production costs by 10%." Christopher said with a smile, in 2019, Camsense's annual shipments will reach hundreds of thousands.

Unsurprisingly, the increase in production capacity can lead to a downward trend in prices, and the cost of research and development will also be gradually amortized. There are many people entering the game of lidar, and many people leaving the game. It has a long way to go. Christopher said that there are still many new problems to be solved in the future.

Christopher is a master tutor of Tsinghua University Shenzhen Graduate School, a high-level talent in Shenzhen, a leading talent in Nanshan District, CEO of Shenzhen Camsense Co., Ltd., undergraduate and master of Tsinghua University, under the tutelage of 973 Chief Scientist, Changjiang Scholar Professor Dai Qionghai. More than ten years of experience in product development and technology management in IT and human-computer interaction technology industries. He has worked in Bell Labs and China Mobile successively, and has conducted in-depth research on visual human-computer interaction technology. As the main inventor, he has more than 20 international and national patents, and has successively won the second prize of Shandong Province Science and Technology Progress Award and Shenzhen Science and Technology Progress Award. Municipal Science and Technology Patent Award. In early 2014, he founded Camsense as CEO, leading the company to research and industrialize computer vision spatial positioning and position tracking technology. At present, the products have been widely used in TV, VR and robotics fields.