The art scene is perhaps not the first industry that you imagine when thinking about automation. We imagined a robot that could adapt its paintings based on the interest of the crowd. The robot learns what techniques and compositions work by taking the attention from museum visitors as input for learning.

In a two-week project, we turned a CNC milling machine into an adaptive artist by preprogramming certain actions – taking blue or yellow paint, and drawing a rectangle or a circle – and using Q-learning to teach the robot about our personal taste. In 75 iterations, the robot learned that it should take paint before drawing, as well as our preference for variation in the canvas.

---