How AI Radar Perception Can Transform Autonomous Transportation
In the race to design a safe autonomous vehicle, the key is building reliable perception capabilities. Each of the three primary AV sensors (camera, lidar, radar) has its own strengths and weaknesses.
April 15, 2021
As the race to build a safe autonomous vehicle has sped up over the past decade, I’ve been fascinated by the technology OEMs select to power their vehicles’ computers and the sensors that feed these mechanical brains information about the environment around them.
What I’ve seen are big bets on artificial intelligence (AI) and software, and since joining the board of Oculii in January, I’ve become a big believer in the power of both to significantly enhance the performance of future radars to safely steer the vehicles of the future.
Why AI Software Solutions Are the Future of Autonomous Vehicles
In the race to design a safe autonomous vehicle, the key is building reliable perception capabilities. Each of the three primary AV sensors (camera, lidar, radar) has its own strengths and weaknesses:
Cameras can sense the context and rich visual information in the environment. However, they do not directly measure 3D information and perform poorly in low-light environments, including fog and snow.
Lidar provides high 3D spatial resolution, but it is still unproven in harsh environments and too expensive to be widely deployed in mass-manufactured passenger vehicles.
Radar has the ability to perform well in poor weather conditions, and is cost-effective, low-power and market-proven from a robustness and reliability perspective. But traditional radar has poor spatial resolution, limited sensitivity and narrow field of view.
Over the past decade, AI software has been used to perform sensor fusion and provide a sum that is greater than that of the individual parts, leveraging the strengths of each sensor while mitigating the weaknesses.
In particular, AI software has transformed the capabilities of camera hardware, enabling significantly more accurate classification, detection and scene segmentation. But regardless of how advanced camera-based AI has become, these sensors are still fundamentally passive, optical 2D sensors that can be easily impaired by environmental factors – limiting the operational design domain of an autonomous vehicle that has to work in all weather conditions.
Industry-Voices-bug (002)_89
Unlocking the Transformative Potential of Radar with AIWhile radar is the only sensor capable of operating in poor weather conditions, traditional radar is not capable of providing the spatial resolution or performance that automakers require for fully autonomous operation. This is because traditional radars are not “intelligent” – the radar repetitively sends out the same constant signal, and resolution is poor as a result.
Just as AI software revolutionized what the automotive industry was able to achieve with standard camera hardware, I believe AI software is about to revolutionize what can be achieved with standard radar hardware.
Today innovative software exists that solves what has traditionally been a hardware problem by using an adaptive, phase-modulated waveform that changes in real time with the environment, to dynamically increase the resolution of any radar hardware platform by up to 100X.
Karl-Thomas Neumann_Headshot (002)
By enabling existing radar hardware to be significantly higher resolution, this new software makes AVs a more viable commercial reality. Radar technology is significantly more affordable than alternative 3D sensors, and it is market proven.
The future of AV perception is already here. It is no longer out of reach to use reliable, low-cost sensors with intelligent AI software to enable autonomous functions in vehicles, and the auto industry needs this kind of thinking.
Karl-Thomas (KT) Neumann (pictured above, left) is a member of the Board of Directors of Oculii, a high resolution radar company enabling the next generation of autonomous systems. He is a former CEO of automaker Opel and Tier 1 supplier Continental AG.
About the Author
You May Also Like