eDavid the robot painter excels in numerous styles
July 17, 2013
The line between art and technology isn't just being blurred, it's being erased altogether. Painting or sketching from photographs and life, for example, is a technique that is now being mastered by robots. The latest, called eDavid, combines a camera, computer vision software, and a standard welding robot arm to skillfully recreate (in a variety of styles no less) any image you feed its software. It seems that even art, a cornerstone of human ingenuity since the dawn of man, isn't safe from a robot takeover.
Though some of the sketches of eDavid (Drawing Apparatus for Vivid Image Display) look a bit like an image run through Photoshop filters, or printed on an old dot-matrix printer, the results will send shivers down the spines of traditional artists.
"(We're) working in computer graphics, so we are computer scientists. But we love art and have some projects that are at the border between science and art," says Oliver Deussen of the Computer Graphics and Media Design lab at the University of Konstanz in Baden-Württemberg. "With our feedback mechanism we try to mimic the way in which a human painter creates a drawing, a sequence of applying paint strokes to the canvas and then comparing if the color distribution approximates what should be on the canvas."
Essentially, a camera pointed at the canvas captures an image of the work in progress and regularly compares it to a goal image. The software then chooses what paint color and brush strokes (which the team calls stroxels) are needed, and where, incorporating a technique called Line Integral Convolution. This method tracks important image features like smooth outlines. By constantly comparing the two images, the software can make up for inaccuracies in brush strokes and unpredictable paint mixing that occurs on the canvas.
Most of the work was done in black and white, but the team has already completed at least one color piece and plans to do more in the future. The system also ditches the complex process of mixing paint on a palette. Instead, the robot arm dips the brush into one of 24 paint wells and washes the brush when needed.
The visual feedback loop determines to what extent the original image is approximated, which in turn affects how the paint or ink is distributed. This gives rise to different styles. Some sketches were done by simply dabbing the canvas in a pointillistic style, while the paintings employ a series of short brush strokes, giving an impressionistic style.
It usually takes around ten hours to complete a painting like the ones shown in the photo gallery. That's because eDavid is programmed to create images in the style of Rembrandt, using many translucent layers to build up the detail.
"To a certain extent aesthetics can be described by means of mathematics and created by algorithms. This is fascinating for us," adds Deussen. Putting the technological aspects aside, eDavid wouldn't look out of place as an art installation in and of itself. In fact, the group is currently working with artist Patrick Tresset on an exhibit about robot paintings. Tresset created a portrait-drawing robot using inexpensive Robotis brand hobby robot servos.
It's sort of mind-blowing to think that if robots are ever a part of our everyday lives, this kind of software could turn them all into talented artists. For example, students at the Learning Algorithms and Systems Laboratory at the Ecole Polytechnique Federale de Lausanne (EPFL) programmed their resident HOAP-3 robot to sketch portraits from images it took with its cameras. It's a trick that has also been implemented in AIST's HRP-2 robot and other industrial robot arms.
Working with Deussen on the project are Thomas Lindemeier, Sören Pirk, and Mark Tautzenberger of the Computer Graphics and Media Design lab.
You can see eDavid in the video below.
- Around The Home
- Digital Cameras
- Good Thinking
- Health and Wellbeing
- Holiday Destinations
- Home Entertainment
- Inventors and Remarkable People
- Mobile Technology
- Urban Transport
- Wearable Electronics