To bring safe and affordable autonomous vehicles to market, companies are making trade-offs across a variety of dimensions including sensor performance, practicality of deployment and price point.
A number of AV leaders have chosen lidar for its high-resolution performance capabilities despite it typically being expensive and impractical. Some companies have overlooked innovations in radar that make it highly suitable for market adoption. Thanks to the advent of radar sensors coupled with cutting-edge machine learning algorithms, AVs outfitted with radar sensors can finally provide active sensing capabilities on a par with their lidar counterparts, and sometimes even better. Here's why.
1. Software-enhanced radar now offers spatial resolution matched to consumer AV applications.
As momentum in AV development started to pick up two decades ago, researchers focused primarily on lidar because it offered unmatched spatial resolution inherently given its shorter wavelength. Lidar's ability to build a high-definition map of the environment allowed researchers to develop a new set of perception, localization and navigation algorithms that paved the road to where the AV tech stack is today.
In the pursuit of AV breakthroughs using lidar, many researchers have since lost sight of radar's potential. Traditional automotive radar has poor spatial resolution, limited sensitivity and a narrow field of view. But it is very cost effective given its history of deployment in automotive since the early 2000s.
Low-cost radar deployments coupled with sophisticated software can allow AVs to sense obstacles, pedestrians and environmental information at a spatial resolution required for safe navigation.
2. Longer-wavelength radio waves can "see" where lidar is blind.
Radio frequency energy has an inherent advantage over the optical wavelengths used in lidar systems. The longer-wavelength radio waves can penetrate mediums that might otherwise scatter or absorb the higher-frequency energy of lidar. The situations in which lidar performance degrades often overlap with obscuring a camera's vision, or a human's vision. This results in sensors with a coupled failure mode.
Radar sensors perform well in many adverse weather conditions,. Having a sensor that does not suffer from the same environmental failure modes creates an independent means of sensing to enhance safety. This represents a potential massive safety win for radar over lidar.
3. Software-enhanced radar is practical to deploy and cost effective.
For decades, the most effective way to improve radar resolution was to add more antennas and channel hardware, which increased the cost, size and power consumption of radar sensors, rendering them unfit for consumer applications.
A sensor enhanced with modern machine learning algorithms and signal processing techniques can, on the other hand, achieve resolution improvements at a price point far more accessible to commercial and consumer automakers.
