1. Introduction
In recent years, light detection and ranging (LiDAR) technology has been deployed in a myriad of applications. Over the years, the LiDAR system design has improved considerably, resulting in a design with remarkably low cost, size, weight, and power (SWaP) requirements. Being light and energy saving, the role of LiDAR in aerial and mobile platforms has increased to facilitate mapping and obstacle avoidance which were traditionally thought to be challenging. As stated by Liner. J, “A lean system is more desirable in the current social, economic, political, and global environments” [
1].
The classification of LiDAR instruments can be broad and subjective, depending upon the context of application. Nonetheless, this instrument is commonly classified using the three types of information-capturing functionality it offers namely spatial, spectral, and temporal. Spatial information-capturing is a fundamental functionality of every LiDAR instrument. This information is typically obtained using the time of flight (TOF) measurement. LiDAR systems that are able to gather spatial information are available in three varieties: one-dimensional (1D), two-dimensional (2D), three-dimensional (3D), with the 2D and 3D spatial information gathering achieved with the aid of optical deflecting systems. The spatial information is essential for constructing an accurate 3D map of the environment. However, spatial information alone is not sufficient for application requiring object detection. The second class of LiDAR instruments are capable of measuring the spectral information of a material such as the laser return intensity (LRI). LRI refers to the reflectance because of the interaction between the wavelength of the transmitted pulse from the LiDAR instrument and the targeted material. Since the LRI is characteristic to a specific material type, it is potentially useful for identification of surface properties of a target material. However, to avoid ambiguities in LRI readings, at least two wavelengths of laser are required. On top of that, some application requires temporal information gathering functionality in addition to spatial and spectral information. This can be achieved by using the repeated LiDAR technique. Repeated LiDAR is a process of collecting temporal data of a target environment over a finite period of time [
2]. Temporal information is essential for understanding dynamic processes such as plant growth and soil erosion [
3].
Numerous reviews on LiDAR technologies geared toward specific applications have been published in the past [
3,
4,
5], ([
6], p. 3), [
7,
8,
9,
10,
11,
12]. It was observed that the major portion of the literature is dedicated for theoretical analysis of LiDAR principle [
4,
5,
6,
7]. The understanding of LiDAR principle helps the designer to select an appropriate technique for his application of interest. Besides that, conceptual frameworks were introduced in the past to describe the LiDAR architectures. The LiDAR instrument is a composition of multiple subsystems such as power supply, range finding, control, and beam deflection. As mentioned earlier, the low SWaP specifications have significantly influenced the improvement of selected subsystems, especially the power supply and range finding. This serves as a guideline for the design and development of LiDAR systems at the component level.
Albeit the number of review articles published in the past, there is still a lack of review articles dedicated specifically to the scanning mechanisms of LiDAR instruments. Consequently, the LiDAR beam scanning literature continues to expand in different focus areas without appropriate organization. Being an integral component of the 2D and 3D LiDAR architecture, the scanning mechanism is an important feature because it influences the SWaP specifications of the overall design. Furthermore, there is a need for the designer to be aware of state-of-the-arts scanning mechanisms as the trends of LiDAR instruments in the market are changing rapidly. Hence, the aim of this paper is to provide a comprehensive survey of LiDAR scanning mechanisms by reviewing past research works as well as commercial datasheets. We propose a classification scheme for various LiDAR scanning mechanisms into four main classes: (i) opto-mechanical, (ii) electro-mechanical (iii) MEMS, (IV) solid-state based. Subsequently, the performance, SWaP specifications, as well as the technology readiness level for each class has been examined critically.
The study reveals that mechanical scanning has achieved high maturity level compared to the rest. Commercially available 1D TOF LiDAR modules in combination with mechanical stages have been the preferred combination for the development of 2D or 3D LiDAR systems. Next on the list, MEMS scanning has also achieved a remarkable technological readiness level for manufacturing. It is widely considered for applications requiring low SWaP requirements.
The rest of this paper is arranged as follows.
Section 2 introduces the LiDAR architecture and specifications relevant to beam scanning. In
Section 3, the classification of beam deflection mechanism is presented. In
Section 4, discussion on latest trends in LiDAR design is presented. Finally, there is a brief conclusion in the last section.
2. LiDAR Architecture Overview
The LiDAR architecture is defined as “the art of LiDAR instrumentation concerning LiDAR hardware and software” [
13,
14,
15,
16]. A fully functional LiDAR system is made of four major subsystems namely laser rangefinder, beam deflection, power management, and master controller units, as depicted in
Figure 1. These basic blocks are equally mandatory whereby a failure in any of these subsystems may lead to a loss in functionality of the LiDAR system. However, in the absence of the beam deflection subsystem, the LiDAR could still function as a 1D LiDAR, which is commonly known as a laser range finder (LRF).
The range finding unit is the core of the LiDAR system, within which are present the components required to generate, transmit, and receive short laser pulses. Laser diode, photo diode, transimpedance amplifier (TIA), and time to digital converter (TDC) are the commonly found components in the range finding unit. Optical components such as collimating lens and focusing lens are also employed in the range finding units to respectively reduce the divergence of the transmitted beam, and focus the received laser beam into the detector. The control unit handles basic signal processing, control signal generation, and communication with the host PC. The final subsystem is the power supply unit, which is responsible for producing necessary power required by the LiDAR system. The design of this unit is subjective, depending upon the voltage and current requirements of the beam sub units in the LiDAR system. Typically, direct current (DC) machines such as servo motors and stepper motors consume 5 to 12 V and 1.0 A, while the microcontroller requires 3.3 V to 5 V to operate. Hence, the power supply units incorporate regulators that are capable of handling multiple voltage requirements. Finally, the beam deflection unit which is an integral part of the LiDAR system is responsible for acquiring spatial information in 2-or 3-dimensional scans. Currently, there are numerous mechanisms to implement this unit. The following section is geared toward an extensive review on the beam deflection mechanisms for a LiDAR system.
2.1. LiDAR Specifications
The LiDAR scanner specifications are essential information for a developer to select an optimal product that best suits his application. It can be broken down to four levels as depicted in
Figure 2. The first specifications in the hierarchy include information related to ranging such as maximum and minimum detection range, resolution, accuracy, update frequency. Second specifications in the hierarchy are related to physical parameters such as size, weight, power consumption which can be found in a product manual. These specifications are important for applications involving mobile or aerial platforms where the size, weight, and power may be a constraint. Since LiDAR employs lasers, information related to that such as wavelength, power emitted, and class of laser is included in the specifications as a part of safety compliance. Some manufacturers even disclose specifications related to the optics of the LiDAR as shown in the bottom of
Figure 2. Example of optical specifications commonly found in product manual includes focal length of lens and beam divergence. The specifications described thus far only best describes a characteristics of 1D LIDAR. In the case of scanning LiDAR additional specifications related to motion of the beam deflection is required. This specification includes parameters like field of view (FOV), angular resolution, response time (T
resp) and number of scan points (N). The details of scanner specifications are discussed in the next section.
2.2. Beam Scanner Specifications
This section presents specifications related to motion of LiDAR beam. First the FOV is one of the primary specifications that falls under this list. FOV is the span of area that can be observed by a scanning LiDAR. Typically the unit for FOV is represented in degrees. For 2D LiDAR the FOV is only limited to horizontal plane, while for 3D scanner it involves both horizontal and vertical planes as depicted in
Figure 3. The computation of horizontal FOV (FOV
H) is shown in Equation (1). Similarly, the computation for vertical FOV (FOV
V) is shown in Equation (2).
Next, angular resolution is the second most important specification of scanning LiDAR. In contrast to the range resolution, angular resolution describes the smallest possible step that can be moved in axis of rotation of the laser beam as shown in
Figure 3. Based on the literature; there is no standard unit for representing angular resolution. Similar with the specifications of stepper motor, some manufacturers and researchers use degrees while others use total number of steps to represent the angular resolution. The relationship between these two units is described in Equations (3) and (4). The representation of angular resolution in terms of number of steps offer several advantages as it helps to derive further performance parameters such as line scan time (T
Line), angular axis rate (A), and total number of scan points (N). The angular resolution is mainly influenced by the motor and encoder employed in the design. High resolution metal disc (4 inch) encoders are capable of providing angular resolution up to 0.072° degrees. The resolution of the encoder disc is directly proportional to its diameter. Hence, there is a tradeoff between resolution and diameter of encoder disc when a low SWaP design is desired.
Angular rate is another specification used to describe the frequency of LiDAR beam motion. For 2D LiDAR the time taken for one complete sweep in the FOVH is called the line scan time (T
Line). This parameter is a product of horizontal resolution (H
res) and response time (T
resp) as shown in Equation (5). Furthermore, the angular rate for horizontal axis (AH) is computed by taking the reciprocal of line scan time (T
Line) as shown in Equation (6). The unit for horizontal axis rate (A
H) is measured in Hertz. Unlike the computation of horizontal axis rate, the vertical axis (A
V) rate requires derivation of extra parameters before it can be computed. The total number of scan points (N) is one such parameter. It is a product of horizontal resolution (H
res) and vertical resolution (V
res) as shown in Equation (7). Later, the time taken for a complete frame (T
frame) is computed as shown in Equation (8). Lastly, the vertical axis rate (A
V) is computed by taking the reciprocal of frame time (T
frame). The angular rates can be also expressed in terms of degrees per second as shown in Equation (10).
where θ
1 is the minimum azimuth angle, θ
2 maximum azimuth angle.
where φ
1 is the minimum elevation angle and φ
2 maximum angle measured clockwise direction with respect to
y axis.
where dθ is the azimuthal angular resolution in degrees
where dφ is the elevation angle resolution in degrees
where H
res is the angular resolution in steps and T
resp is the response time of LiDAR to collect data from a point.
where T
Line is the time taken for 2D sweep
where H
res and V
res are the angular resolution in steps for horizontal and vertical planes respectively.
where N is the total number of scan points measured from the scene and T
resp is the response time of LiDAR to collect data from a point.
where T
frame is the time required for 1 complete area scan
where A is the axis rate in Herts and FOV is the Field of view in degrees.
4. Discussion
Electromechanical scanning methods were predominantly used for extending dimensionality of 2D LiDAR instruments as demonstrated by past researchers [
25,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38,
39,
40,
41,
42]. However, only a handful of researchers have considered extending the dimensionality of a 1D LiDAR using electromechanical scanning methods in the past. As stated earlier, the weight of the LiDAR is the main limiting factor that dictates the selection of appropriate scanning methods. In recent years, advancement in photonics industry has led to the emergence of new LiDAR products in the market such as SF Lidar (Ligthwave optoelectronics) [
82,
83,
84], LidarLite (Garmin) [
85,
86], Rplidar (Slamtec) [
87], Sweep (Scanse) [
44], Tera One (Terabee) [
43], and TFmini (Beneware) [
88]. Currently, the LiDAR instruments have achieved low size, weight, and power (SWaP) design compared to [
22]. Since weight is no longer an issue now, researchers began to reconsider using electromechanical scanning methods for designing a 3D LiDAR using a 1D LiDAR module. For instance, [
40] has demonstrated the conceptual design of a 2D LiDAR using a 1D LiDAR module and a servo motor. On top of that, custom made gear trains are utilized to extend the rotation of the servo motor to up to 360 degrees. As a step further, the design of a 3D scanner using a 1D LiDAR and stepper motor has been demonstrated by [
41]. Finally, there exist a number of commercially available 2D and 3D LiDAR sensors which employ electromechanical scanning method in their design. For instance, 2D LiDAR systems such as Sweep [
44], Terabee [
43], and Rplidar [
87] has adopted similar conceptual design demonstrated by [
41]. However, they were implemented using DC motor with encoder to achieve full azimuthal scan.
The facts presented thus far has indicated that the pulsed TOF LiDAR design has gone through a progressive growth. Lower SWaP designs are seen in later LiDAR instruments, owing to the advancement of high-density circuit integration. The efforts for integrating electronic and optoelectronic circuits for LiDAR application dates back to the works of Pasi [
89]. In this work, Pasi has demonstrated an ASIC solution for a LiDAR receiver which consists of photodiodes, amplifiers, and time-to-digital converters (TDCs) implemented using the BiCMOS process as depicted in
Figure 13. The design was reported to hit the measurement accuracy of millimeters over a range of 4 to 34 m. In subsequent years, more research works in the integrated receiver design striving toward high speed and accuracy has been reported in [
90]. On the other hand, the development in the LiDAR industry is in step with research works and has gained exclusiveness in the LiDAR design. To date, VL6180 [
91], VL530L [
92], and ISL2950 [
93] are LiDAR sensors that have been successfully fabricated as standalone system on chip (SoC) designs. These integrated circuits (ICs) are capable of driving external laser diodes, receiving pulse from photodiode and computing TOF. They are even capable of communicating with a host microcontroller by means of inter-integrated circuit (I2C) protocol. The architectures of VL6180 [
91] and ISL29501 [
93] is shown in
Figure 14 and
Figure 15 respectively. The main difference between VL6180 and ISL29501 lies in the implementation of signal processing pipeline. VL6180 employs a microcontroller whereas the ISL29501 uses an on-chip DSP circuit to process the signals. Besides that, the ISL29501 provides the option of using external laser diodes and photodiodes if the application requires them. It is evident that the SoC design has influenced the traditional architecture used in previous LiDAR designs.
In the emergence of new LiDAR designs, the decision on whether to buy it or built it needs to be justified. The main objective of academic research in LiDAR designs is cost reduction. However, it is evident that LiDAR manufacturers are aware of the costing issues and have taken several measures to keep the overall price low. First, most manufacturers have made the decision to employ laser diodes in the near infrared (NIR) band. This is on the grounds that optoelectronic devices in NIR band has become inexpensive and easily available because of the advent of fiber optics communication industry. On top of that, NIR lasers are invisible and cause less harm to the eyes. Second, p-i-n photodiode is preferred over avalanche photodiode (APD) for its lower cost despite its drawbacks. Although APD theoretically has a higher gain than p-i-n diodes, the additional circuits for supplying high voltage add an overhead to the overall cost of the LiDAR instrument. For medium range operation, p-i-n photodiodes exhibit a satisfactory level of performance, making the use of APDs unnecessary. Finally, LiDAR manufacturers employ SoC signal processing technology to integrate all the required functions necessary for LiDAR operation.
Figure 16 shows the architecture of the LiDAR Lite V2 (Pulse light) [
86]. It can be observed that the SoC processing core employed in Lidarlite V2 [
86] is made of an 8 bit microcontroller to handle the control and communications; a sampling circuits to capture and down sample the logic state of external comparator; and a transmit signal generator to generate waveform patterns and a correlation processor to perform correlation operations of the incoming signal. This small form factor solution reduces the complexity and power consumption issues of the overall system. As a comparison, Kelden [
94] in his design spent a total of 125 Euros excluding manufacturing and miscellaneous costs. Despite this, the author claimed that they failed to produce a working prototype because of the use of cheap development procedure and lack of expertise. Meanwhile, commercially available LiDAR instruments such as the LiDAR Lite V3 [
85] costs the same amount as quoted by Kelden [
94] but with better performance. In conclusion, purchasing an off-the-shelf 1D LiDAR modules is a better alternative to building a module from scratch for applications that require medium operating range and centimeter level resolution.
The status of 2D and 3D LiDAR developments are slightly different compared to the 1D LiDAR counterpart. The development of 3D LiDAR instruments is an ongoing research. As the costs of commercially available 3D LiDAR instruments are still high, some studies from the literature have proposed alternative methods in developing a 3D LiDAR system. The first option is to mount a 2D LiDAR instrument onto a mechanical platform. Various platforms utilizing rotating (tilting), revolving, and oscillating motions were proposed. Among all the proposed methods, mounting a 2D LiDAR instrument onto a tilting unit has gained the most popularity in 3D LiDAR system development. Presently, the latest development in the field involves converting a 1D LiDAR instrument into a 3D LiDAR system. The conversion is done in two stages. In the first stage, the 1D LiDAR instrument is converted to a 2D LiDAR system by an actuator as illustrated in
Figure 17. Commercially available LiDAR instruments such as Sweep [
44] is constructed based on this conceptual design, utilizing a 1D LiDAR module developed by Garmin [
85].
Figure A1 shows the exploded view of the Sweep LiDAR for greater insights on the hardware implementation. The main drawback of this method is that the scanning performance is influenced by the module itself. The current LiDAR module requires 40 milliseconds to perform an individual acquisition, limiting the update rate. Nevertheless, at a cost of
$250 with the capability of 40-m detection range and centimeter resolution, the Sweep LiDAR is a very attractive proposition for obstacle detection application in robotics. The following year after the publication date of [
44], the developers of Sweep has initiated an open source project called sweep DIY scanner kit. The objective of this project is to convert the existing Sweep 2D LiDAR into a 3D LiDAR system as illustrated in
Figure 18, thus completing the conversion from 1D to 3D. The conversion is done by mounting the Sweep onto a stepper motor.
Figure A2 shows the exploded view of a DIY 3D scanner kit. Despite its inherent weakness of lower scanning rate, the prototype can produce a 3D map of the environment within a few minutes [
45]. The application for this type of LiDAR system includes collision detection for unmanned vehicles platform, surveying application such as confined space measurement as well as for research and education.
Although the LiDAR systems mentioned above have reach a satisfactory level of improvements, it still lacks in specifications to be employed in ADAS application. The requirements of LiDAR systems for ADAS application have been discussed earlier by [
70,
81] and are presented in
Table 6. There are three specifications: range, resolution, and safety, which depends upon the LiDAR transceiver unit. Most of the commercial LiDAR operating below 100 m are based on TOF principle. However, for applications require sensing beyond few hundred meters continuous wave (CW) LiDAR are commonly used. The main reason TOF is preferred over CW LiDAR is due to eye safety. A short high-power impulse is safe and does not damage the human eye. However, such a technique is no longer suitable for detecting distance above 100 m unless it includes highly sensitive detectors such as photomultiplier tube (PMT). As an alternative, CW LiDAR operating on 1550 nm is being considered. Besides that, with CW LiDAR design, an improved resolution and power dissipation can be achieved. The rest of the specifications listed in
Table 6 is highly influenced by the scanning mechanism. It is evident that the LiDAR instruments available in the market are unable to achieve a high-speed scanning exceeding 25 frame per second (fps) under the budget of
$200. Hence, mechanical scanning is clearly not the best candidate for this application. The MEMS scanning appears to be a suitable candidate for ADAS because it is small, lightweight, low power, and cheap. However, the main drawback in the size of the mirror violates the horizontal FOV requirements. Increasing the mirror size will not solve the problem because it will reduce its abilities to work in high speed [
81]. Clearly, for this application an innovative solution such as OPA LiDAR or slow light LiDAR are required.
5. Conclusions
In conclusion, a comprehensive review on state-of-the-art LiDAR scanning mechanisms has been reported. The sources of the literature are based on past research works and commercial products. Mechanical scanning mechanism is the most matured technology as less research is being reported lately. It is evident that (After many iterations and improvements), current commercially available 1D TOF LiDAR instrument has achieved remarkable progress in terms of design (i.e., size, weight, and power) and performance (i.e., range, resolution, and accuracy). This has turned the user’s attention to construct self-made 3D LiDAR systems. Among all the existing options, the conversion from a 1D LiDAR to a 3D LiDAR system is currently the best option in terms of cost. However, this method suffers from low scanning rates, which makes it unsuitable for applications requiring high speed operations. Despite its weaknesses, the low cost of this 3D LiDAR system makes it an attractive solution for applications such as robotics, surveying, agriculture, and education.
Besides mechanical scanning, MEMS scanning has gained huge attention as a novel solution for applications with low SWaP requirements. It is by far the more matured technology compared to solid-state scanning. Currently a great emphasis is given in increasing the robustness of the MEMS scanners so that it can meet the requirements for use in ADAS applications, with existing solutions already sufficient for use in UAV implementations.
Finally, there is a huge expectation for solid-state LiDAR systems to fill in the gap in ADAS application. The two-competing technology under these categories are OPA and PCW. Currently, the readiness level of this technology is still low compared to MEMS scanners. Since solid-state scanning is believed to have superior robustness, FOV, and scanning rate potential, both academics and industries are putting in a lot of effort to further develop this technology.