On January 4, 2019, Hesai Technology announced the release of PandarGT 3.0, a 2D scanning solid-state LiDAR with 60° x 20° FOV and an extended range of 300 meters at 10% reflectivity. Its unique, dynamically-adjustable beam distribution can enhance the vertical resolution up to 0.07°, enabling pedestrian recognition from 150 meters away. Equipped with Hesai’s patented interference rejection function, PandarGT 3.0 averts possible interference from any other LiDAR.
2019: The Much-Belated Year One of Solid-State LiDAR
Solid-state LiDARs have been the holy grail of the autonomous driving and ADAS industry. They bear the hope of low cost, high reliability, and fine resolution. Year after year LiDAR companies renew their presentations and delivery dates of solid-state products, yet the Waymo test fleets are still crowned with their signature mechanical LiDARs, and worldwide pilot programs of robo-taxis and autonomous delivery vehicles opt for mechanical LiDARs without exception.
What are the real advantages of being solid-state? What are the challenges facing the players in the field? How does PandarGT 3.0 – the third generation of Hesai's solid-state LiDAR – face these challenges?
Solid State: Core Advantages and Pending Challenges
LiDAR as a 3D scanner has been widely used in autonomous driving and robotics. It is often the crucial and yet most expensive sensor on an autonomous car, priced at over tens of thousands of dollars per unit.
Solid-state LiDAR, by definition, has no moving parts (or "rotors", as in mechanically rotating LiDARs). Typical all-solid-state solutions include Optical Phased Array and Flash.
In recent years, the industry has incorporated all non-rotating LiDARs into the solid-state group. LiDARs with tiny moving parts are not strictly solid-state, but they demonstrate many traits shared by solid-state ones, such as high resolution and limited horizontal FOV (forward-facing instead of 360°). Hence, the name "solid-state" applies.
Core Advantage 1 – Ultra-High Resolution
The most obvious advantage – mechanical LiDARs’ fixed vertical resolution (0.11° for VLS-128 from Velodyne) cannot compare with continuous-scanning solid-state LiDARs' 0.03°, reached at low frame rates such as 5 Hz. The result is picture-like point clouds.
Core Advantage 2 - Efficient Volume Production (Manufacturability)
Mechanical LiDARs by convention have one pair of laser emitter and receiver per channel. As the number of channels increases, so does the time needed for assembly and calibration. By contrast, most solid-state LiDAR solutions contain only one or a few pairs of emitters and receivers. This structural distinction cuts down the workload of inter-channel calibration and raises production efficiency by a large margin.
The solid-state products on the market today may have possessed the above two advantages, but are all struggling to trade off among the other elements – measurement range, reliability and cost. This is the reason for the current market void and the frequently postponed delivery dates.
Here is a look into these challenging elements:
Long Range Measurement?
Mechanical LiDARs are practically reaching their physical limit: Hesai’s recently released Pandar40P has a stunning range of 200 meters at 10% reflectivity. By contrast, the solid-state ones are still underdeveloped. Due to flood illumination, the range of flash LiDARs is limited by physics to within 50 meters. Other types of solid-state LiDARs are curbed by the core component’s immaturity – the optical steering mechanism cannot reach the desired optical efficiency.
Public information shows that most solid-state sensors on the market are struggling to detect 30 to 65 meters at 10% reflectivity. This partly explains why so many live demos take place indoors – 30 m @10% products can be quite impressive in a meeting room but expose their weaknesses when taken on a drive.
This brings another paradox: does a 30-meter-range LiDAR need high resolution? Solid-state LiDARs are most valued for their range. The vertical resolution of a 32-channel mechanical LiDAR is 1.2 m at 200 meters away; solid-state technology can bring down this resolution to 0.1 m, depicting a pedestrian with 15 lines in the point cloud instead of merely two lines. For detecting objects at the 30-meter range though, a 32-channel sensor suffices and a solid-state one would be overkill.
It is said that solid-state LiDARs have no moving parts and are therefore more likely to become automotive grade. This notion is repeated frequently enough for people to believe it, though it remains unproven. At present, the only certified automotive-grade LiDAR in the world is a mechanically rotating one from Valeo; furthermore, hundreds of automotive parts involve a rotary motion and have no problem reaching automotive grade.
On the other hand, the scanners used by most solid-state LiDARs still face uncertainty in production processes and reliability. Until a mature scanner supplier comes to rescue, most solid-state LiDAR developers will be stuck in the demo phase. Unstable performance and inability to manufacture in volume has characterized the solid-state field over the past three years. It is one thing to demo, and another to turn this brand-new design into a product as reliable as traditional mechanical LiDARs.
Compared with mechanical LiDARs, solid-state sensors benefit from a simpler mechanical and optical structure and fewer components. However, this is no guarantee of lower cost: in order to improve performance, LiDAR companies tend to use expensive lasers (one fiber laser costs as much as thousands of dollars) and scanning mechanisms, raising the system cost even higher. Adding that solid-state LiDARs cannot scan 360° individually, the cost of using multiple LiDARs and implementing a point-cloud stitching algorithm further burdens a self-driving car’s sensor budget.
PandarGT 3.0 - Hesai's Third-Generation Solid-State LiDAR
Since the release of PandarGT (GT stands for “gu tai/固态”, the word for “solid-state” in Chinese) in December 2017, Hesai’s solid-state team has gone through trials and errors to finally present this breakthrough version – PandarGT 3.0. A trail of engineering nuts was cracked in its making – electronically, mechanically, andoptically. The road test results speak for themselves.
In summary, two things matter the most in the development of this product.
1. Self-developed High-Speed Scanning Mirror System
We believe in looking forward two steps in advance – pioneering technology is developed and implemented in parallel with the promotion of existing products. The research and design of scanning micromirrors dated back to the time when Hesai started developing mechanical LiDARs. The structural design, manufacturing processes, mirror drive module and angle feedback system have all evolved dozens of times. The result is a scanning mirror with significantly enhanced optical performance, stable operation from -40℃ to +120℃ and impressive shock resistance. This best-in-class operating temperature range facilitates the automotive application of solid-state LiDARs.
Besides the ultra-high optical efficiency that extends the LiDAR’s measurement range, the mirror is also characterized by a scanning frequency of 1500 lines per second. For GT 3.0, this frequency translates into 300 channels at 5 Hz and 150 channels at 10 Hz.
2. Self-developed Fiber Laser
To achieve its 300 m @10% measurement, GT 3.0 uses a 1550 nm fiber laser. This wavelength enjoys a far higher power limit than 905 nm for Class 1 eye safety, and is thus more suitable for long-range measurement. Still, fiber lasers may pose problems for automotive use – typical drawbacks include a narrow temperature range and a large form factor. Current solutions usually separate the fiber laser from the LiDARs’ detection module, putting the laser in a cooler place inside the vehicle.
However, these solutions bring about certain troubles:
- The LiDAR system as a whole is still cumbersome.
- Due to arrangement constraints, the optical fiber and signal wires connecting the detection module and the laser/control module travel a long distance within the car body. This increases cabling complexity and aggravates the system’s signal integrity.
- Separated fiber joints are prone to dust intrusion, which damages the laser without warning.
To solve those problems, Hesai developed its own fiber laser and controller. This laser module has 1 MHz pulse frequency and 1 nm pulse width, its dimensions smaller than all the commercially available counterparts that have an average power of over 2 W. Furthermore, piece cost is reduced to hundreds of dollars for volume production, and the -40℃ to 110℃ temperature range satisfies stringent automotive requirements.
Tailored Features for Self-Driving Applications
1. Dynamically-Adjustable FOV
GT 3.0 is an intelligent LiDAR: its vertical FOV can be dynamically adjusted between 5° and 20°, adapting to local and highway scenarios. In addition, users can set the channels to be either uniformly distributed or progressively converging toward the center. In the converging mode, vertical resolution at the FOV center reaches 0.035° at a frame rate of 5 Hz and 0.07° at 10 Hz. Even the vertical FOV as a whole has a configurable pitch of ±5°, enabling flexible measurement of objects above and below the FOV.
2. Full Interference Rejection
GT 3.0 shares the full interference rejection feature with Pandar40P, Hesai’s newly released mechanical LiDAR. When multiple LiDARs operate in close range, Hesai LiDARs will be unperturbed. This feature is detailed in Critical Features of HESAI Pandar40P.
Outdoor Live Demo at CES
2019 is the first time Hesai is attending CES, the world’s largest tech show for consumer electronics. We will be exhibiting Pandar64 (mechanical LiDAR) in the LVCC North Hall and holding outdoor live demos of PandarGT 3.0 (solid-state LiDAR) in the Platinum Lot. The measurement range, resolution and dynamically-adjustable FOV of GT 3.0 will be demonstrated in real time. Please visit Hesai’s official website (www.hesaitech.com) to make an appointment.