Saved by the Sensor: Vehicle Awareness in the Self-Driving Age

First, they were inconceivable and then a far-fetched dream. Now self-driving cars are finally starting to make their way into the real world. In most robotic use cases today, an error can lead to a bad part, or perhaps an unswept floor. However, when driving a car, a computer must constantly make potentially life-threatening decisions. Because of this, the information fed to it has to be extremely accurate, meaning that several redundant sensor systems are often used to account for different driving situations.

Below is a list of the sensor systems used to maintain driverless-vehicle safety. In some cases, these can even apply to human-driven vehicles for tasks like emergency braking, blind-spot awareness, and adaptive cruise control to match up to the car in front of us. Significantly, they not only have to work once, or even a few times in a lab—they also must be automotive-grade and able to withstand the abuse of thousands of miles in varied conditions and still produce accurate results every single time.

Vision sensors — If you consider how we as humans observe the world, we collect an incredible amount of information via sight. Not only can we view and process the light waves reflecting onto our eyes, we can use the two of them together to observe distance. Vision sensors (or cameras, as they are also called) can be extremely useful in self-driving vehicles, able to take on such tasks as detecting pedestrians and even reading road signs. They can, however, be fooled in situations that human eyes would normally be able to handle, such as a brightly colored object against a bright sky or even painted objects that depict a different situation than reality.

LiDAR — Short for “light detection and ranging,” this sensing method measures the reflection of near-infrared light off of objects as it scans in 360°. Multiple beams are emitted at angles to each other, and the sensor array physically spins inside the device’s housing to produce a three-dimensional picture of its surroundings. Physically, the devices normally resemble a sort of oversized upside-down coffee cup, and can most famously be seen on top of Google’s Waymo cars, scanning an area as they drive.

While considered by some to be the gold standard in sensing technology for self-driving cars, price has been the major drawback. This type of sensor package originally cost a whopping $75,000 each when Google started experimenting with the technology nearly a decade ago. However, they have dropped to a tenth of that as of 2017.

Solid-state LiDAR — As the interest in LiDAR, which physically spins infrared lasers inside the sensor unit, has grown, companies are also developing solid-state versions with no moving parts. Importantly, these devices can be much less expensive than “traditional” automotive LiDAR units. Velodyne, for example, announced that it would be building these units at a price of hundreds of dollars each in 2018, and startup Innoviz has promised to make a sensor for only one hundred dollars, also in 2018.

While less expensive and smaller than their coffee-cup brethren, one disadvantage is that they cannot see in a full 360 degrees, meaning multiple sensors will be needed for a complete picture of the vehicle surroundings. This type of solid-state device can be implemented in other exciting ways as well, such as AEye’s version, which uses a vision system and on-sensor processing to focus the LiDAR beam in the needed area.