2020 Brings New Crop of LiDAR Sensors for Autonomous Vehicles
LiDAR has been a staple sensing technology in the development of autonomous vehicles. What's new in automotive LiDAR sensors?
Light detection and ranging, otherwise known as LiDAR, is a powerful, remote sensor system that uses light in the form of laser pulses to measure distances. In autonomous vehicles, rotating LiDAR systems create 3D point maps of the surrounding environments and as a result can detect other vehicles, pedestrians, cyclists, and other obstructions.
Previous LiDAR Systems
We previously looked at LiDAR systems implemented by Google in their autonomous vehicles under Waymo, which first utilized Velodyne’s 64-channel LiDAR sensor. At the time, the devices came with a hefty price tag of $60,000 per sensor. Since then, the journey to find an affordable LiDAR solution has been ongoing for engineers and researchers at different companies and universities.
Since Google’s announcement of their own autonomous vehicle, many developments have been made in LiDAR technology. Image from Waymo
Affordability isn’t the only concern for LiDAR. Over the last several years, safety issues have been raised regarding potential attacks such as saturating and spoofing, which could degrade sensor systems' performance and reduce their reliability.
Since our last look at LiDAR in autonomous vehicles, momentum has picked up across the industry to implement LiDAR and improve upon the existing technology.
Bosch’s Complete Sensor Portfolio for Automated Driving
Bosch recently announced its plans to produce long-range LiDAR sensors for the first LiDAR system suitable for automated driving. Their latest sensor will be used in conjunction with cameras and RADAR sensors to detect both long and short ranges for on the highway and in cities.
To help bring down the cost of traditionally expensive sensors, Bosch will exploit economies of scale to make them affordable and accessible in mass markets.
Additionally, Bosch has been able to enhance its systems with improved artificial intelligence which can classify various objects and trigger warnings or emergency braking maneuvers.
At CES 2020, Baraja announced an update to their LiDAR line, Spectrum-Scan. The Spectrum-Scan platform offers detail at long ranges of objects on dark highways over 200 meters away. Their sensors also offer variable resolution and full API control for trigger instant, inertialess changes in resolution and scan pattern.
The newest generation of Spectrum-Scan claims "inherent interference immunity" via encoded light. This is an additional layer of safety on top of their method of splitting light through a prism for more data points for comparison (e.g., light angle, wavelength, etc.). All of these elements must match up before a sensor's measurements are processed for vehicle decision-making.
A detailed pointcloud of a city street as seen by Baraja's Spectrum-Scan LiDAR sensor. Image used courtesy of Baraja
Baraja asserts that their LiDAR is industrialized and designed for manufacturing and scalability to meet the needs of the autonomous vehicle industry. Their approach allows for the customizability of LiDAR configurations, depending on the needs of a manufacturer.
Velabit by Velodyne
Velodyne’s latest LiDAR venture, the Velabit, was announced in January. The tiny automotive-grade sensor is smaller than a deck of cards, measuring 2.4" x 2.4" x 1.38", with a range of 100 meters and a 60-degree horizontal and 10-degree vertical field of view.
It is designed to be integrated within a vehicle’s body or windshield and produces a robust directional image during the day or night, along with competitive range and resolution for faster object identification and longer braking distances at highway speeds.
The Velodyne Velabit LiDAR sensor. Image used courtesy of Velodyne
The Velabit was notably announced to be sold for $100 when it releases in mid-2020.
Alternatives to LiDAR
While many companies have invested their time and money in LiDAR developments, there have been a handful of criticisms from engineers and researchers. Tesla’s CEO, Elon Musk, has been quoted several times as being critical of LiDAR devices, claiming that they are “expensive sensors that are unnecessary.” In 2019 at an event called Tesla Autonomy Day, he rather famously also asserted that "Anyone relying on LiDAR is doomed." For their part, Tesla cars rely on several alternatives to LiDAR which include GPS, maps, cameras, and other kinds of sensors.
Research done at Cornell University seems to support Musk’s sentiments that LiDAR has many alternatives. The researchers found that data captured from stereo cameras was nearly as precise as LiDAR with the only gap being in the data analysis. Using stereo cameras—or, as the authors refer to it, Pseudo-LiDAR—can offer an affordable alternative to LiDAR with similar results.
An example qualitative comparison between the sensing abilities of LiDAR, Pseudo-LiDAR, and AVOD (aggregate view object detection). Image used courtesy of Cornell University via arXiv
With improvements in camera technology, the quantum leap, and promising results of stereo cameras, the authors believe there could be a revival of image-based 3D object recognition that can close in on the image/LiDAR gap.
While the criticisms of LiDAR might question its purpose in the industry, there's little argument that LiDAR is useful as part of a comprehensive autonomous vehicle solution, leveraging its capabilities and potentially curtailing its limitations.
It is apparent that producers and manufacturers of LiDAR are motivated to provide reliable solutions that can offer performance at long ranges and in dark environments. Long-range performance allows for autonomous vehicles to be viable on highways, a milestone many companies are racing to achieve.
Along with improved sensors, focusing on improving artificial intelligence for classification and object recognition has seemed to bring LiDAR to new heights