Robotics

CurvACE gives robots a bug's eye view

View 9 Images
CurvACE reproduces the architecture of the eyes of insects and other arthropods
CurvACE installed in a robot
Visual range of CurvACE
CurvACE besides some dragonfly examples
Illustration of steps for the assembly of CurvACE parts
Size comparison between CurvACE and one-euro coin
Size comparison between CurvACE and a trilobite fossil
CurvACE reproduces the architecture of the eyes of insects and other arthropods
First prototype of a semi-cylindrical curved artificial compound eye
Detailed image of the processing units integrated in CurvACE
View gallery - 9 images

Robots are getting down to the size of insects, so it seems only natural that they should be getting insect eyes. A consortium of European researchers has developed the artificial Curved Artificial Compound Eye (CurvACE) which reproduces the architecture of the eyes of insects and other arthropods. The aim isn't just to provide machines with an unnerving bug-eyed stare, but to create a new class of sensors that exploit the wide field of vision and motion detecting properties of the compound eye.

The consortium, made up of researchers from CNRS, Aix-Marseille Université, EPFL at Lausanne, Fraunhofer Institute at Jena, and Université de Tuebingen, want to make a bit of a paradigm shift when it comes to camera design. Currently, most cameras are based on simple eyes. That is, the sort of eyes found in humans as well as those of other vertebrates and some molluscs. Essentially, it’s a box with a lens at one end and a retina at the other. This simple, yet elegant, arrangement has many optical advantages, but it isn't the only way to make an imaging device.

The major alternative used in nature is the compound eye. This is a dense mosaic made up of many tiny eyes. When you look at a dragonfly, for example, you’ll notice that the head is almost all eye. Or rather, a collection of eyes. In this arrangement, each facet of a compound eye is a fully functional eye. The major difference is that by making an eye up of segments, the entire organ has much lower resolution. To put it in everyday terms, if a person had compound eyes, they’d each have to be as big as the entire head to have the same resolution as regular eyes.

First prototype of a semi-cylindrical curved artificial compound eye

If the resolution is so bad, why compound eyes? The answer is that they have their own strengths. Compound eyes have a very large field of vision. The cross section of a compound eye is also thin, so it can wrap around an animal’s head without sacrificing the interior. And it’s extremely good at detecting motion.

How this motion detection works in insects and other arthropods is tricky because we don’t understand quite how vision works in, say, a fly. It has thousands of tiny eyes, but does it see thousands of tiny images? That’s hard to say because we don’t know what it sees, or if it even “sees” in any way that we would understand.

Leaving aside metaphysics, it’s probable that what a compound eye sees is a single, blurry, pixelated image. That may seem like a disadvantage, but such a low-resolution pixelated image highlights movement beautifully, making compound eyes very good for motion detection. Combined with the remarkably fast information processing found in a fly’s simple brain, this is what makes swatting the little blighters so frustrating.

Visual range of CurvACE

CurvACE isn’t the first attempt to exploit the architecture of the compound eye, but CurvACE aims to make a much deeper emulation combined with fast digital image processing. CurvACE imitates the structure of natural compound eyes by means of three planar layers of separately produced arrays less than one millimeter thick when assembled. These are made up of a microlens array, a neuromorphic photodetector array, and a flexible printed circuit board.

Fabricating the eye wasn't easy however. It required the accurate alignment of photoreceptive and optical components on a curved surface. The arrays were stacked, cut, and curved to produce a mechanically flexible imager and then equipped with an embedded and programmable low-power signal processor with high temporal resolution. “Embedded” in this sense means that, unlike conventional cameras, electronics can be layered into the compound camera itself because there’s no need to keep a clear line of vision between the lens and the photodetector except within each segment.

According to the consortium, the end result is a robotic eye smaller than a coin with a panoramic, undistorted, hemispherical field of view, a nearly infinite depth-of-field, no image blurring or off-axis aberrations, and image resolution identical to that of the fruit fly. Speaking of fruit flies, CurvACe processes images three times faster than the insect. in addition, the eye can detect motion in everything from a sunny day to moonlight.

CurvACE installed in a robot

The researchers see a number of applications for CurvACE, such as collision avoidance for both terrestrial and aerospace vehicles, distance estimation and vehicle landings. It could be installed in mobile robots, micro air vehicles, home automation and surveillance devices, medical instruments and smart clothing, as well as being used for experimental testing of insect vision theories.

There are currently four families of CurvACEs – cylindrical, active, spherical and tape compound eyes – and the consortium plans to use them to further advance the technology and to develop different prototypes for new applications.

The findings of the CurvACE project were published in PNAS.

The video below outlines CurvACE.

Source: CurvACE

View gallery - 9 images
  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
2 comments
Siegfried Gust
Watching the video is revealing. This needs a lot of improving if it's going to have any benefit versus conventional imaging sensors. Of all the potential applications that they mention, only the testing of insect vision theories seems practical. All the rest are just applications that might show promise if we didn't already have very capable mass produced imaging sensors.
Ian McIntosh
True, though this seems promising as a start. It will be interesting to see how much they can improve the resolution. I wonder could they incorporate a standard forward looking standard image sensor to produce a composite image?