User:Kkrib/sandbox/eDavid (robot)

eDavid (robot) is a painting robot developed at the Universität Konstanz (University of Konstanz) that calculates brushstrokes from an input image and paints the image on a canvas. The project began in 2011 as a one-armed welding robot modified to be able to paint. The robot’s arm can interchange between different brushes and pens, and is equipped with a distance sensor to measure exactly how far the arm is from the canvas. eDavid features two main painting methods: predefined stroke candidates and dynamically generated strokes.

Development
eDavid was developed by the Department of Computer and Information Science at the University of Konstanz. The goal for the project was to build a robot that mimics the manual painting process: painting in iterations until the desired image is achieved. This differs from earlier painting robots because instead of computing the perfect set of strokes and painting once, the robot paints in iterative steps, moving the optimization from the computer to the canvas. eDavid implements features from earlier works, including AARON, a robot developed by Harold Cohens to paint abstract art. However unlike AARON, eDavid focuses on representing an input image accurately rather than generating abstract art.

Painting Process
eDavid takes in an input photograph and generates a set of brushstrokes which vary in length and width. The algorithm implements a visual feedback loop. The loop follows these steps: This process mimics the human painting process, just as a painter would lighten and darken a painting to satisfaction rather than paint a perfect painting in one attempt.
 * paints a set of computer generated strokes
 * checks to see if the painted strokes accurately represent the input image
 * generates the next set of strokes to more accurately match the input image

Stroke Generation Methods
eDavid utilizes two different stroke generation methods: Predefined strokes, and Dynamically generated strokes.

Predefined strokes
This method chooses the new strokes from a set of 3 different lengths, 3 different widths, and 60 different directions, resulting in 540 different possible strokes. At each location in the image all of the 540 possible stroke candidates are “tried-out”, and eDavid chooses the stroke that matches the image with the highest quality. During each iteration of the feedback loop, eDavid paints 300 predefined strokes. The resulting brushstrokes are short in comparison to dynamically generated strokes.

Dynamically generated strokes
This method uses the image gradient to create longer brushstroke paths rather than many shorter strokes which are used in the predefined strokes method. The image gradient is how the color or intensity of an image is changing at any particular location. For example, if at a particular pixel in an image, the image is transitioning between a lighter color to a darker color, the gradient at that location will indicate which way the image is brightening/darkening and how fast it is brightening/darkening. eDavid uses the image’s gradient to generate a collection of brushstrokes perpendicular to the gradient’s direction.

eDavid takes all of the dynamically generated strokes, and chooses paths that do not overlap. The chosen brushstrokes, usually longer and smoother, result in darker painting than a painting painted using predefined strokes.

Comparison of styles
Predefined strokes result in very clean paintings, however the shorter preset brushstrokes seem more robotic than paintings created by dynamically generated strokes. One way to improve upon this painting style is to include longer and more varied brushstrokes, but this could slow down the process due to the larger number of predefined brushstrokes to choose from.

In dynamically generated strokes, areas of the painting with high detail are often too dark. However, dynamically generated strokes look more complex and artistic than predefined strokes due to the uniquely created brushstrokes. One way to improve upon the dynamically generated strokes is to prevent the creation of dark spots by filtering very fine lines in high detail areas.

Future Work
eDavid is currently only capable of producing grayscale images, and the researchers at the University of Konstanz are looking to implement color paintings in the future. The current limitations of color painting include the inability to mix oil colors properly due to the inconsistent viscosity of different colors and the stroke technique of oil color brushes is much more complex.