LiDAR has gained support as a sensing technology that may make autonomous vehicles safe enough to commercialize. Could solid-state LiDAR be the next step? Here's a look at a couple of methods of implementing solid-state LiDAR and who's developing it.

Over the last several years, there's been a veritable revolution in LiDAR. In large part, this has been spurred by the ever-looming boom of autonomous vehicles, where LiDAR is increasingly used as a method of sensing obstacles. There's also been a resultant increase in affordable LiDAR, making it more accessible to a broad range of designers.

The next step in LiDAR evolution is, according to some, the utilization of solid-state LiDAR.

In this article, I'll introduce the basic concepts behind typical LiDAR, go over how solid-state LiDAR differs, and then take a look at who's working on developing solid-state LiDAR options.

 

The Problem with Rotating LiDAR Sensors

Most current-generation LiDAR technology bounces a pulsed laser off of a rotating mirror to distant targets and then records the time for the reflected pulse to return. This technique provides a full 360°-coverage of an area. Depending on the optics and resolution of the receiver, it can provide a scan sector beam width that varies from a one-channel, one-pixel high 0° field of view, to a multi-channel, 32-pixel, 45° field of view.

Of course, as we'll see in this article, a 45° field of view may be typical (for now) but it's certainly not the limit.

 

Image of a rotating LiDAR unit. Image courtesy of Renishaw.

 

With many high-precision moving parts, these mechanical LiDARs are expensive, sensitive to vibration, and difficult to miniaturize. Additionally, if mounted inconspicuously at the front of a car (rather than at the top, as Waymo originally showcased), the body of the car would eliminate as much as 50% of the LiDAR's field of view. 

To fully incorporate a unit into a car, automakers need a cheaper, more robust option.

One way of dealing with this issue is to incorporate multiple less expensive sensors. Multiple companies are now researching and developing solid-state LiDARs that are robust and inexpensive. While they have a limited field of view, their low price points allow vehicle makers to integrate several units into a car for a fraction of the cost of a single current-gen solid-state LiDAR.

 

Methods of Implementing Solid-State LiDAR

Solid-state LiDARs use a system, detector, and sometimes a MEMS mounted in a non-rotating housing to bathe a scene in pulsed-laser light and record the reflected pulses. There are several ways to accomplish this task:

 

Crossed Array Scanning

With this method, laser light is pulsed sequentially from a vertical array, while sequentially being detected by a horizontal array. This crossed configuration produces a resolution that is the product of the number of laser emitters and the number of photodiode detectors. The hardware keeps track of the laser position, the detector position, and the time for the light to return to generate a three-dimensional point cloud of the environment.

 

Image courtesy of LeddarTech

 

MEMS-Based Solid-State LiDAR

With a MEMS-based solution, a single powerful laser pulse is used along with a mirror to create the same effect as the laser array shown above. As the mirror is manipulated to create a scanning laser pulse, a sensor array detects the reflected light and generates the three-dimensional point cloud of the environment.

 

Image courtesy of LeddarTech

 

Innoviz's InnovizPro

The Israeli company Innoviz has a solid-state LiDAR called the InnovizPro that you can purchase and immediately place on your car. It is based on the MEMS technology and has an expected price of a few thousand dollars. 

Their upcoming InnovizOne, an automotive-grade LiDAR, is expected to be available in 2019 for just a few hundred dollars. An Innoviz press release first publicized at CES in January stated that high-volume customers could see prices as low as $100 per unit.

 

Image courtesy of Innoviz.tech

 

When interviewed via email, Innoviz CEO and Co-Founder, Omer Keilaf, stated that their LiDAR "is designed to deliver extremely rich, high-density output in the form of a high-resolution point cloud, enabling [them] to extrapolate advanced object detection capabilities. With extremely high data intensity, Innoviz's LiDAR is able to build additional layers of information and thereby offer superior levels of object detection, classification and tracking."

Innoviz provides a full software stack solution that includes object detection, classification, and tracking. The object classification and tracking are especially important for automotive applications where a changing environment poses serious threats to pedestrian and driver safety.

See the Innoviz LiDAR in action performing SLAM (Simultaneous Localization and Mapping) in the video below.

 


LeddarTech's LeddarVu, etc.

LeddarTech also has solid-state LiDAR solutions that are designed to be placed at multiple locations on an automobile to provide complete coverage. LeddarTech has multiple units available already for under $1000. 

The LeddarVu platform utilizes "patented signal processing and algorithms" to create solid-state LiDAR sensors with versatility.

 

The Vu8, a solid-state LiDAR module developed on the LeddarVu platform. Image courtesy of LeddarTech

 

LeddarTech's software implementation also allows for object detection and classification.

Additionally, their website offers a significant number of resources to help users to better understand the technology. For example, LeddarTech's whitepaper on their optical time-of-flight sensing tech (PDF) explains their use of software-implemented algorithms in processing sensor data. This educational approach can help engineers better understand the technology in the LeddarTech LiDAR units.

 

  Leddar Optical time of Flight Whitepaper  


Quanergy's S3

Quanergy claims to have "the world's first affordable solid-state LiDAR sensor" in their S3 model, which was unveiled way back in January of 2016. While the original LiDAR modules from half a decade ago would typically cost tens of thousands of dollars, the S3's price point was initially announced to be $250. 
 

The S3 LiDAR sensor. Image courtesy of Quanergy.

 

The S3 can fit in the palm of one hand and its lasers have a 120 ° arc. Quanergy touts the fact that the S3 has no moving parts which they assert will improve reliability. This is just one of many ways companies like Quanergy are hoping to ease friction between LiDAR design and implementation in the automotive industry, which requires high levels of reliability for the sake of safety.

 

Conclusion

Inexpensive, solid-state LiDAR units are on the market now from multiple manufacturers. As time progresses, their cost will continue to drop and their capabilities will continue to improve. 

How long will it be before this technology is modified for home use? From electric wheelchairs and drones to motorcycles and other mobile platforms, there are plenty of applications—aside from autonomous cars— that could benefit from environmental awareness and mapping.

Have you worked with LiDAR before? Would you use solid-state LiDAR in a project if you could? Let us know in the comments below.

 

Comments

7 Comments


  • MisterBill2 2018-02-23

    What is missing, I think, is a system using a camera to cover some field of view, with a flashed laser, and a means to examine the individual pixels to get time of flight information. The same camera could also see the images and thus determine what it needed to measure the distance to. Because it would only take one sensor for both purposes it should cost a bit less than two sensors. Also it would have perfect correlation between image and distance. Plus, aiming would be simple, just by viewing the image. Of course examining pixels in real time instead of in a scanned sequence would be quite a trick, It would take more than clever software, I am certain.

    • User8192 2018-02-23

      It depends on what you mean by “aiming” the sensor.  Automobile manufacturers today can’t even get the headlights right in the factory.  I find the headlights need to be adjusted even on brand new cars after one takes delivery.

  • ronsoy2 2018-02-23

    Be sure your car insurance has good coverage. And you are in a BIG heavy vehicle!

  • Still quite skeptical about any truly autonomous vehicle, I might meet you half way. LiDAR mounted on all four corners might serves as sensors which in turn notify the human driver as to trouble approaching from any direction. But steering, accelerating or braking will almost always require a human decision. Artificial intelligence might involve a voice warning of trouble approaching, from any direction, offer appropriate suggestions, but not too much more. Well maybe something like this chick possible.

  • User8192 2018-02-23

    All this effort is in an attempt to eliminate human drivers.  Meanwhile, the electronics content and prices of cars have been escalating into the stratosphere, to the point where the average shmoe won’t be able to buy any of it.  Sometimes keeping it simple and accepting responsibility for one’s actions is the best solution.

  • rpapagolos 2018-02-27

    I think the ultimate solution would require some fixed points along roads that the system could reference for distance and other information maybe a 3D map of the road ahead. The system could eliminate some fixed objects (building, intersection, curb sides, utility poles and the like) if the system could get this information without scanning in real time it would only need to scan for changes to the static environment.
    At first glance this may seam impossible but remember all the roads traffic lights etc didn’t exist when cars first hit the road we managed to build all of the infrastructure we now have. Adding fixed points of reference would take time money and commitment but is is deferentially doable.

  • drc_567 2018-03-02

    An extremely beneficial feature, related to the goal of real time analysis, would be the ability to detect any object that has motion, with respect to the vehicle reference frame. A computation of this nature would require the time difference of any two subsequent points for a given object ... implying that individual objects could be designated somehow. A system of this nature would mitigate collisions between the vehicle and animals, people, or other vehicles.