To that end, a new product called EyeDAR, a low-power millimeter-wave radar sensor roughly the size of an orange, could provide radar-equipped AVs with critical inputs about surrounding traffic, extending and enhancing the vehicles’ sensing accuracy.
Placed at key points such as streetlights and intersections, these low-profile, inexpensive sensors could ensure AVs never fail to pick up on emergent obstacles, even when they are not within proper range for the vehicles’ onboard sensors or when visibility is severely limited.
“Current automotive sensor systems like cameras and lidar struggle with poor visibility such as you would encounter due to rain or fog or in low-lighting conditions,” said Kun Woo Cho, a postdoctoral researcher at Rice University who leads the EyeDAR research project. “Radar, on the other hand, operates reliably in all weather and lighting conditions and can even see through obstacles.”
Radar systems transmit signals in a given direction, and when that signal encounters an obstacle in its path, part of it reflects back to the source, carrying information about the obstacle. However, only a small fraction of the radar signal emitted reflects back, and most of it actually bounces away from the source device.
Scattered Signals
In the context of self-driving vehicles, this means a large fraction of the radar signal their sensing stack emits scatters away from the vehicle, leaving them with an incomplete view of their surroundings. Pedestrians emerging from behind large vehicles, cars creeping forward at intersections or cyclists approaching at odd angles can easily go unnoticed.
Thanks to its placement on roadside infrastructure such as traffic lights, stop signs or streetlights, EyeDAR can capture radar reflections that would otherwise be lost. The device’s unique structure allows it to determine the direction of reflected signals and report that information back to self-driving vehicles.
“It is like adding another set of eyes for automotive radar systems,” said Cho, who specializes in metamaterial antenna design.
EyeDAR boasts a simple, elegant design inspired by a highly efficient real-world sensor: The human eye. The device consists of two main components: A 3D-printed Luneberg lens made from resin which functions similarly to the lens of the eye, focusing incoming signals from any direction onto a focal point on the opposite surface; and an antenna array surrounding the lens on the back end which functions like a retina, detecting the signal and determining its direction.
A low-power millimeter-wave radar sensor about the size of an orange could provide radar-equipped AVs with critical inputs about surrounding traffic, which could extend and enhance the vehicles’ sensing accuracy
Source: Rice University
Whereas conventional radar systems rely on large antenna arrays and complex algorithms to estimate angles, EyeDAR’s physical design does most of the computation work typically required for direction finding – one of the most power- and data-intensive tasks in radar processing.
“Our lens consists of over 8,000 uniquely shaped, extremely small elements with a varying refractive index,” Cho said.
Antenna Array
Through an intentional distribution of these elements, the lens structure interacts with incoming radar signals in a smart way, routing them to the right spot on the antenna array. The approach has proved fruitful: In testing, EyeDAR was able to resolve target directions more than 200 times faster than traditional radar designs.
Moreover, EyeDAR communicates what it sees without transmitting new signals. Instead, the sensor alternates between absorbing incoming radar waves and reflecting them back to the source radar in a form it can interpret as a sequence of 0s and 1s.
“Like blinking Morse code,” Cho said. “EyeDAR is a talking sensor – it is a first instance of integrating radar sensing and communication functionality in a single design.”
This combination of sensing and communication in a compact, inexpensive and low-power architecture makes it feasible to deploy large numbers of sensors across roadways.
In the case of self-driving cars, the system promises to be especially useful in dense, high-traffic urban settings. However, the potential application space is much wider: EyeDAR could integrate into robots, drones and wearable platforms. Networks of these sensors could also share information with one another, allowing each device to see well beyond its own range of sight.
Cho said she has interest in what the system represents from a computing standpoint. As autonomous systems increasingly interact directly with people, Cho argues intelligent physical design will have to complement artificial intelligence.
“EyeDAR is an example of what I like to call ‘analog computing,’” Cho said. “Over the past two decades, people have been focusing on the digital and software side of computation, and the analog, hardware side has been lagging behind. I want to explore this overlooked analog design space.”