In January a group of artists and students from the MIT visited the Computer Graphics Group (Prof. Deussen) and evaluated the e-David painting robot (see www.e-david.org) for different methods of painterly abstraction. With our painting robot we want to mimic human artists and find out to which extent painting processes can be performed by machines. Part of our project is research on computational creativity: we want to research and develop methods in which the machine starts to develop its own painting strategies and styles. The machine should be able to learn from past paintings when producing new ones. Another important aspect is quantification of abstraction. We want to develop styles in which we can predict and maintain a wanted degree of abstraction, technically (number of strokes) and also perceptually.
The MIT students want to use the machine in a semi-automatic way and combine human and machine painting. Ideally the human would only direct the system by sketching and giving hints for the style to be used. Then the machine does the often cumbersome painting work. At the end the human finishes the artwork. Benjamin Tritt (http://bentritt.com) is a well-know American Artist that combines classical painting styles with modern motives. He is interested in machine painting since large-scale artworks are very time-consuming and robots might be very helpful in creating intermediate painting layers.
Benjamin Tritt (right) and Michael DiBenigno discuss with Thomas Lindemeier (left)
e-David experimented with different paint palettes:
e-David is now on Google+.
2013/2014 - Winter
ARTE showed a television report about e-David in their show "Tracks".
2013 - Autuum
New paintings in acrylic color:
2013 - July 17
Our vimeo video will be shown at the robotfilmfestival in San Francisco on July 20-21, 2013.
2013 - July 8
Our video has been added to the 'Vimeo Staff picks'. This step helped enormously to spread the video. We're are very happy about receiving so much positive feedback across the web.
2013 - June
The german television channel ZDFinfo showed a video feature about e-David in their show "Elektrischer Reporter".
2013 - Spring
We've continued working with acrylic colors and have developed a new algorithm, that just works with black and white color. Transparent layers of black and white are painted on a gray background.
For details we use thin brushes. The detail regions are detected using computer vision techniques.
For such a picture e-David paints about 10 hours (10.000 brush strokes).
A painting of the famous 'Matterhorn', a mountain located in the swiss alps.
2013 - February
Patrick Tresset joined our project and is helping us until October. We are looking forward to work with him. He also works on a painting robot named call, that draws portaits of humans.
<iframe src="https://www.youtube.com/embed/bbdQbyff_Sk" width="560" height="315"></iframe>
Read this article about the work of Patrick Tresset:
We extended the set of black ink painting styles with a new one. This style performs better on long flowing objects like hair or tree branches. The previous style tends to perform better on other forms like short grass.
Since there are two different styles which perform different on different objects, we tried to combine the two painting styles in a drawing.
After much work we've accomplished to create a feedback guided painting style, which uses acryl color and four types of different brushes. The complete painting process is done by eDavid itself. The changing of color and brush is all done by eDavid. It is possible to use 24 different colors in the painting process.
The next stept towards a feedback oriented painting process is to use colors that are not fully opaque. We used black ink that was diluted by water. An algorithm simulated the brush strokes and the darkening of the ink and thus was able to construct a series of brush strokes that represents the object. The inputs for this form of paintings were grey scale photographs. The resulting paintings consist of 10-20.000 Brush strokes of all the same length but different orientation. The painting time is about 3.000 strokes per hour, the computing time is 1.000 strokes per minute.
We achieve our own aesthetics since no human painter would be able (and willing) to paint brushes of this orientation and uniform length. Nevertheless, the object to be rendered is clearly visible.
First time we work with color. We used three different kind of Brushpens (a special form of Pen with a brush) with colors Yellow, Magenta, Cyan. The image is renered by using differenet brush strokes widths in a set of parallel lines.
First time we work with brushes. The width of a stroke is controlled by the distance of the tool from the canvas. We encounter problems with rippled paper that arises during painting. We measure the paper surface before painting but for technical reasons are not able to do this during the painting. This is why we have to use very solid and heavy paper. However, nice results appear by painitng strokes of varying sizes - such paintings are hard to do manually.
First results are obtained by rendering portraits (and other images) using a single long line. The line is produced in a way that dark areas of the image are crossed more often than light ones. The drawing consists of about 500.000 individual line segments. We drew by using a fine liner.
Roboter setup with painting table, tool holder, tools with damping is finished. A first demo program creates sketches of faces. On of our models: our former university president.