How technology will bestow self-driving vehicles with sight
Nowadays, our vehicles are far from “blind.” In fact, they are connected to and have the capability to process an astounding amount of data. GPS systems provide us with the best routes to take based on time of the day and real-time traffic and weather conditions. Sensors provide split-second information to automatic braking systems, alert us to necessary maintenance, transmit emergency help alerts and even wake us up if we are about to nod off at the wheel.
Soon, however, our cars will literally be able to see, interpret the world around them, and become their own chauffeurs.
Many vehicles are already equipped with different a range of systems that virtually augment drivers’ perceptions (as in rear-view cameras or proximity alert systems) or provide information directly to the vehicle from sensors and even radar systems. However, the true undisputed protagonist of artificial vision is the “lidar,” originally a portmanteau of “light” and “radar,” but now often considered an acronym for “Light Detection and Ranging.” The lidar will be essential to self-driving vehicles.
The lidar was originally developed in the 1960s, right after the advent of the laser, and first employed in 1971 by the Apollo 15 Mission to map the surface of the Moon. However, the technology has continued to evolve in precision, reach, and speed, also integrating 3D capabilities.
State-of-the-art lidars will provide self-driving car artificial intelligence systems with high-precision 3D maps, as well as a terrific real-time system for detecting obstacles and tracking moving objects: not only cars, but also bicycles, pedestrians, animals, etc.
Technically, a lidar creates a continuous, real-time and constantly updated 360-degree 3D map of the area around it by firing off millions of laser beams per second and measuring how long it takes each beam to return to the sensor.
It’s as if our eyes could see in all directions, all the time. Lidars also provide an unprecedented accuracy: +/- 2 cm and can detect virtually anything within their range, and under just about any light condition, up to 120 meters away. Moreover, unlike cameras, lidars are cannot be tricked by shadows or bright light.
You’ve probably seen a lidar before. It’s the bulky box mounted atop test self-driving vehicles rooves that spins continuously around, which introduces another issue: size. Nobody wants to own – let alone drive around with – a car crowned by a spinning turret.
A range of different companies and start-ups are currently working on developing smaller, more compact solid-state lidars that can be integrated directly into car body parts, such as its bumpers. However, in this case, cars would require four or more such systems to preserve 360-degree vision, a financial burden for any automotive maker.
Accuracy at a Cost
Indeed, this technology comes at a price, literally. Lidar sensors are currently priced about US$75,000 and are mainly used in aircraft and speed-detection devices, as well as for high-precision mapping.
And although some lidar producers are claiming to soon be able to cut the system costs by as much as 90%, that still represents a significant amount, even for the price of even an upscale car. Furthermore, this requires a substantial investment in research, against a supply demand that is still extremely low.
A further issue is represented by the fact that different lidar systems, generate different images, so the artificial intelligence platform in each car would always require the same type of system to ensure continued and efficient machine learning, but this may well be overcome by a future standard once lidars are ready to be mass-produced.
So, despite the many challenges, including the claim by some automotive producers that a range of other sensors will be more than sufficient to create safe and dependable artificial vision systems, it is highly likely that the true success of self-driving cars will depend on the lidar.
For further information: