The market for global autonomous vehicles is currently at over $54 billion and could reach more than ten times that amount by 2026. Drivers for this technology include regulatory pressure, increasing connectivity by passengers, high cost of vehicle ownership, and disinterest by an increasingly large segment of the population to own a vehicle outright. The principal technology enabler of autonomous vehicles is the interaction between predictive AI and imaging technology. The sensors create a 3D picture around the vehicle, and the AI interprets how the vehicle should respond. Though vehicle manufacturers have made significant progress in this area, the light-off point will come when a company reduces the amount of time between image and response to near-zero.
Technology Update
Because the imaging technology is a critical technology enabler, it has become the most significant limiting factor to widescale production of L3 and above autonomous vehicles. Development in sensor technology has historically combined video, RADAR, ultrasonic sensors, and LiDAR. This approach, however, has not been successful. Glare, focus issues, depth perception challenges, fog, and heavy rain have all inhibited the accuracy of existing sensor technology.
Thermal Imaging Sensors
Today, the leading sensor technology approach is thermal imaging. A thermal imaging sensor can detect longer wavelengths than those in the visual spectrum, images below the visual spectrum, making it adept at detecting humans, animals, and objects in heavy rain, fog, and at night. FLIR is leading the way in thermal imaging for autonomous vehicles.
Localizing Ground-Penetrating RADAR
Vibration caused by the vehicle can make LiDAR data unstable, and as such, the reliability of the technology may be limited. This is one of its primary drawbacks as a sensor for autonomous vehicles. Developed at MIT for military applications, localizing ground-penetrating RADAR (LGPR) was designed to solve the problem of sensing complex surfaces. The technology sends radio waves into the ground, creating a stable digital map of the subsurface, which does not shift or change and is not subject to weather variability. This map is the reference state for the sensor. With accuracy up to the levels of existing LiDAR, camera, and RADAR sensors, LGPR performs near parity with the best-performing systems without being negatively influenced by the weather. GSSI is continuing the development of this technology, currently licensed under a patent for prototype/demo-only uses.
Market Update
Navigant Research published a Q1 study citing three companies, Waymo, GM Cruise, and Ford Autonomous, ahead of the field in self-driving technology. The rankings prioritized companies that incorporate at least SAE L1 automation (driver support features) in their entire fleet, as well as OEMs that publicly demonstrated L4 autonomy and outlined their path to commercialization.
Waymo
To date, Waymo (formerly Google’s self-driving car project, now a company under Alphabet) is still using a combination of LiDAR, RADAR, and cameras to map the vehicle’s surroundings. The company has been willing to share its multimodal sensor data. This openness shows that Waymo is committed to transparency and working with public and private research entities work on improving the predictive accuracy of the software.
Waymo has also developed and marketed three standalone LiDAR sensors (short-, medium-, and long-range), and began selling them to customers. This step is a significant hedge on LiDAR as the winning sensor technology in the race to full autonomy, despite Waymo augmenting LiDAR with RADAR and cameras in its vehicle development.
GM Cruise
Like Waymo, Cruise (acquired by GM in 2016) uses a combination of LiDAR, RADAR, and cameras for its imaging strategy. Navigant Research rated Cruise just behind Waymo, and as it employs the same sensor approach, it is not surprising that the companies’ technologies are so close. Cruise scaled back its development fleet effort, choosing to continue testing in San Francisco and ensure the commercial launch goes smoothly.
Cruise acquired a LiDAR manufacturer in 2017, like Waymo seeming to hedge that the technology will win the race for preferred imaging technology. Cruise employs RADAR and cameras for redundancy in its imaging, surmising that at least one of the three will provide an accurate map in all conditions. This approach does little to inspire confidence that any of the three is the best choice, and illustrates the gap the winning OEM must fill with sensor technology.
Ford Autonomous
Ford, with Pittsburgh-based Argo AI, mainly uses LiDAR as its imaging strategy, with RADAR incorporated with the 3D map created by LiDAR. The company uncovered another potential failure mode beyond the known glare, fog, rain, snow phenomena...insect splatter on the LiDAR sensor.
This annoyance to drivers when occurring on the windshield could be catastrophic to LiDAR sensors, whose mission is to measure the distance to nearby objects as accurately as possible. Ford designed a two-pronged approach: a preventative measure and a corrective one. Air nozzles alter the flight paths of insects away from the LiDAR sensor, while high-pressure water sprays onto an affected lens, with velocity, duration, and impingement angle managed by AI.
Takeaway
For all the advancements in autonomous technology, the speed to market will only go as fast as the sensor technology will take them. The three companies closest to commercial viability, Waymo, GM Cruise, and Ford Autonomous, all employ LiDAR augmented by RADAR, and two with additional cameras.
Each of those sensor technologies has inherent limitations, and while engineers tirelessly aim to mitigate, weather and human behavior patterns are too unpredictable to be able to design against all circumstances the vehicle might encounter. Thermal imaging and localized ground-penetrating RADAR provide the best options for the winning sensor type. When OEMs can direct their automotive volumes at an optimal sensor technology, the price will plummet, unlocking the secret to widescale autonomous vehicle adoption.