The story of radar-based perception is a deceptively simple one. The front-facing applications for radar-based perception are immediately evident:
But with cameras already positioned to view around the vehicle, what is the innovation in – or value of – radar-based perception in all directions?
In fact, through AI-based analysis of the vehicle’s surroundings, Arbe’s 360° Perception is critical for enabling safety in everyday driving scenarios.
The surround data captured by Arbe’s perception radars, which leverage AI to identify, classify, and track objects in 360°, is processed in real-time to create a full free space map around the vehicle, as well as an analysis of the evolving hazards sensed by the radars. From emergency braking to cyclist and pedestrian detection, and from lane change to highway autopilot and traffic jam assist, L2+ and higher applications have so far been mostly about driver comfort. By perfecting these applications with 360° perception in all environmental conditions, Arbe elevates them from a nice-to-have comfort solution to must-have safety features.
The use case of Autonomous Emergency Steering (AES) exemplifies the need for and benefits of 360° radar-based perception. When a vehicle must make an instantaneous decision regarding steering, it needs to be able to estimate the distance, speed, and orientation of surrounding objects (including, for example, the vehicle behind it) in long range, and to have a reliable Free Space Map in order to understand what action to take – steer into the next lane, guide the vehicle off to the shoulder, etc. In order to make the best and safest decision, it needs to understand what its options are in every direction. (By the way, these are the same capabilities required for merging onto a highway or navigating a busy intersection).
Aside from the fact that cameras do not directly measure speed and depth estimations for objects in the environment, they are also dependent upon vehicle headlights to be functional in the dark, and generally suffer from reduced functionality in adverse weather conditions. For these reasons, radar-based perception, which is unaffected by weather and lighting conditions, is a requirement for autonomous safety in all scenarios.
360° perception enables a vehicle to continuously track its surroundings and monitor emerging or immediate threats from once-perilous positions, including rear, side, and blind-spot scenarios. This is accomplished with multiple radars that overlap in the fringes, tracking objects of interest smoothly from one radar to the next and validating the whereabouts of the same object through two different perception algorithms. This “object sharing” capability reduces false alarms and provides higher resilience to occlusion scenarios.
There are additional benefits to object sharing: by employing multiple radars in tracking objects, Arbe lowers the tracking latency. Individual and independent sensors track objects over multiple frames in order to achieve an acceptable level of confidence and risk assessment. However, multiple radars working together are able to draw accurate conclusions – classification, trajectory, velocity, elevation, and more – and detect threats faster. Consider, for example, multipath rejection. Multipath refers to non-real objects or reflections that are sometimes identified by the radar due to the mirror effect or a return from another target. A single radar will require multiple frames in order to recognize that the object is not behaving as expected – for a car, a pedestrian, etc – over time, and thus reject the object as a multipath. But by comparing the object across multiple radars, the algorithms can recognize unusual patterns much more quickly, improving response time for the system as a whole and, by extension, improving safety.
Arbe’s 360° radar-based perception also includes a level of environmental understanding that resembles cognition, adjusting the radar parameters to optimize the needed performance not only for the driving scenario at hand but also the specific status of the vehicle and the particular task that is being performed. For example, if the car identifies the need for a left-hand turn, the side radar range detection must be optimized in order to safely execute the action. In cognition mode, Arbe’s radars, in coordination with the rest of the sensor suite, identify all of the characteristics of the situation at hand and automatically reconfigure to the necessary parameters, reaching the long-range detection necessary for the sensors to determine safe turning opportunities. Similarly, a large metal truck could obscure a smaller object or pedestrian due to the strong amplitude; in order to avoid saturation, Arbe’s radar either changes its mode, adjusts its range, or alters the threshold or sensitivity in response, allowing the system overall to continue to trust the environmental data being collected and correctly separate the two targets. Arbe’s radars are able to configure the radar’s parameters in real-time to safely account for real-world scenarios and challenges.
For truly safe autonomous technology, perception in 360° is non-negotiable. Arbe’s perception algorithms enable not only unprecedented understanding in 360° but also information sharing between multiple radars, supporting environmental comprehension that is both complete and coherent.
Ultra-high resolution radar adds unique depth and velocity information in real-time and in all weather conditions – a key functionality, and the reason imaging radar is being widely recognized as a required front-facing addition to camera sensors. Arbe’s 4D ultra-high-resolution radar technology is the first to be detailed enough to provide critical sensing data diversity, such as depth, relative velocity, object orientation, and long-range detection at levels optical sensors can’t match, and offering reliable redundancy in environmental conditions they can’t overcome. This is the breakthrough that will power the advancement from features like automatic emergency braking and adaptive cruise control to full autonomous path planning.
Connect to learn more
