Cutting-edge technology! Get a glimpse of the 128-line LiDAR before its official release

Although the sensor solutions of various autonomous driving systems are different, most car companies and technology companies currently cannot do without LiDAR. This is why the entire market is eagerly waiting for this sensor to be mass-produced at a low price.

At the end of November, Velodyne announced a new laser radar product in North America, the 128-line laser radar VLS-128. Compared with the HDL-64, another high-line laser radar widely used by autonomous driving companies, the VLS-128 has been comprehensively upgraded in performance:

• First, the detection distance of VLS-128 can reach up to 300 meters, and the overall performance is 10 times that of HDL-64;

• Secondly, Velodyne has implemented automated calibration on this lidar and ASIC-ized key modules, which will greatly improve product production efficiency and reduce costs.

According to Velodyne officials, VLS-128 will be used in perception scenarios for Level 5 autonomous driving in the future. Due to its longer detection distance and higher resolution, the data generated can be used to label different objects and classify images even without camera assistance.

Velodyne China staff told Xinzhijia that VLS-128 is currently in the product debugging and verification stage for different specifications, and the final product specification parameters will be announced at CES in January next year.

Before CES, Velodyne brought the VLS-128 prototype to China. This week, VLS-128 just completed a small-scale demo. Below is the information and experience that New Intelligent Driving obtained at the demo site.

See the product first

The demo was held in an empty office in the Raycom Infotech Center, where Velodyne Beijing is located. A Velodyne VLS-128 prototype was placed on a small table in the center of the office. The entire space is about 30 meters x 80 meters, a total of about 2,400 square meters. According to the nominal parameters, the entire office should be within the "range" of this small device.

According to the product size measured on site, the VLS-128 prototype has a diameter of 17cm and a height of 13cm. As Velodyne CEO David Hall said, the size and weight of the VLS-128 are less than one-third of the HDL-64 .

1. Two important parameters: detection distance & angular resolution

According to data previously obtained by New Intelligent Driving, the VLS-128 prototype has a horizontal field of view of 360°, a horizontal angular resolution of 0.1°, a vertical field of view of 40° (15° up, 25° down), a minimum vertical angular resolution of 0.17°, and a detection distance of more than 200m. This set of data is basically consistent with the results verified by Velodyne China technicians on the day of the demonstration.

Velodyne told New Intelligent Driving that in terms of angular resolution:

• The horizontal direction is mainly related to the rotation speed of the laser transmitter. Taking 5 Hz as an example, the horizontal angular resolution is about 0.1°, and 20 Hz is about 0.4°. This value is easy to understand, because the faster the laser emission speed, the higher the scanning frame rate of the laser radar, and the fewer points can be hit in a fixed time. When the points are not densely hit, the horizontal angular resolution will be reduced accordingly.

• The vertical direction depends on the density of the laser radar beam emission. According to Velodyne's official introduction, the laser beams emitted by VLS-128 (and the previous VLP-16 and VLP-32) are dense in the middle and sparse on both sides at a vertical angle. This is to enhance the detection accuracy of the main sensing area. Based on this design, the central angular resolution of the 128-line laser radar can reach about 0.17°.

In fact, when it comes to the detection accuracy of LiDAR, angular resolution is an important indicator. In layman's terms, resolution can be understood as the ability to distinguish adjacent targets at a certain distance. The higher the resolution, the stronger the ability to recognize objects. In this way, imagine that in the vertical direction, the LiDAR will emit laser beams at different angles. The denser the beams, the greater the point cloud density in space, and the higher the corresponding angular resolution. In addition, the ranging error of VLS-128 within a short distance is within plus or minus 3 centimeters.

Judging from the detection effect demonstrated by the VLS-128 prototype, its detection effect on people walking in the house and other obstacles is already very obvious. Objects in the point cloud can be easily identified by the naked eye. If the algorithm directly performs image classification based on the point cloud in the later stage, it will bring convenience to data processing.

*Raw point cloud data output by VLS-128

2. Data volume

The VLS-128 prototype has two external cables, one for power supply and one for data transmission.

In terms of data transmission volume, Velodyne China technicians told Xin Zhijia that compared with the previous 64-line lidar with 2 million points per second and 5-6M data volume, the point cloud number of VLS-128 has roughly doubled to about 3-4 million points per second, and the data volume is about 1M per second.

The data of VLS-128 will be continuously sent out in small packets with a high frequency, with an average interval of 3-5 microseconds. However, some autonomous driving practitioners on site also told Xin Zhijia that in the actual data processing process, developers are usually accustomed to performing overall data processing after completing a circle of scanning, which will cause a certain degree of delay in detection and output of detection results. This delay needs to be compensated by software optimization.

* On-site staff showed the network card parameter information for point cloud data transmission

3. Software and post-processing

When asked about the performance improvement of VLS-128 in dealing with the problem of strong light, Weng Wei, Executive General Manager of Velodyne Asia Pacific, mentioned some of Velodyne's layout in the combination of software and hardware. He pointed out that the problem of strong light has always been a problem of combining software and hardware for LiDAR. Based on Velodyne's previous products, VLS-128 has made certain optimizations in both hardware and firmware.

As for the software algorithms that support the hardware, Velodyne's VP of autonomous driving John Eggert also revealed that there is a team within the company that is conducting related research and development, but the product has not yet been launched.

The team revealed that at CES 2018, Velodyne will release a data table of the reflection intensity of different objects detected by its point cloud data to assist developers in implementing classification algorithms.

4. Facing mass production: new factories and ASICization

When VLS-128 was released, CEO David Hall mentioned in his blog that he spent three days assembling an HDL-64 by hand, and his thumb and index finger were numb for a month after that. However, VLS-128 does not use a lot of manual processes, but "adopts a new automated calibration technology, which greatly improves production efficiency."

Velodyne staff told Xin Zhijia on site that the new product will be produced at the company's San Jose factory. This production line was once used to produce VLP-32C products and can perform automated parameter calibration.

In addition, the launch control unit inside the VLS-128 is currently close to the ASIC level, and the light source and detector have been modularized, but the overall solution still uses a mechanical rotation. According to the rough data given by Weng Wei, compared with the previous 64-line product, the 128-line lidar has reduced more than 200 internal components. By 2019, when automotive-grade products are launched, the latter's internal components will only account for 10% of the existing 64-line products.

In terms of other performance, John Eggert introduced that Velodyne's lidar products can currently cope well with rainy weather, but severe fog or haze pollution is still a challenge. As an industrial product, the VLS-128 consumes about 40W of power and has an IP67 dust and water resistance rating.

Market plan: To prioritize finding incremental value in the shared travel sector

Last month, the ride-sharing company Uber just signed a large order of 24,000 SUVs with Volvo, which will join Uber's huge army of autonomous ride-sharing vehicles. According to Weng Wei, Velodyne is currently negotiating with Uber on the customization of VLS-128 products, and Uber will also become the first delivery partner of Velodyne's VLS series. According to the plan, in addition to positioning, obstacle avoidance and other needs, Uber also hopes that VLS-128 can be used as the main sensor for updating regional real-time high-precision map data to obtain better added value beyond the sharing economy.

In terms of product pricing, Weng Wei hopes that the price of VLS-128 and HDL-64 will be as close as possible, but he also revealed that Uber plans to gradually purchase VLS-128 products over a three-year period, and the product price agreed upon by both parties has dropped to the range of several thousand to ten thousand US dollars.

However, the current product size of the VLS series cannot meet the embedded needs of mass-produced models. Velodyne officials revealed that at next year's CES, its solid-state LiDAR product Velarray for automotive standards will also be released for the first time. This product eliminates mechanical rotating parts, has a detection distance of 200m+, a horizontal detection angle of 120 degrees, and a vertical resolution that is better than VLS-128, expected to reach 0.1°. Weng Wei also said that Velarray specifications will start from 32 lines, gradually transition to 16 lines, 8 lines, and even 4-line obstacle avoidance-based LiDAR for low-speed autonomous driving vehicles.

*Solid-state LiDAR product Velarray

"In the future, the technologies of VLS-128 and Velarray will be iterated approximately four times within two years, and the final automotive-grade products will be released in the fourth quarter of 2019." [End]