Peter Schaldenbrand, James McCann, & Jean Oh
The Robotics Institute, Carnegie Mellon University
ArXiv
Colab Demo
GitHub
Twitter
Painting is an artistic process of rendering visual content that achieves the high-level communication goals of an artist that may change dynamically throughout the creative process. In this paper, we present a Framework and Robotics Initiative for Developing Arts (FRIDA) that enables humans to produce paintings on canvases by collaborating with a painter robot using simple inputs such as language descriptions or images. FRIDA introduces several technical innovations for computationally modeling a creative painting process. First, we develop a fully differentiable simulation environment for painting, adopting the idea of real to simulation to real (real2sim2real). We show that our proposed simulated painting environment is higher fidelity to reality than existing simulation environments used for robot painting. Second, to model the evolving dynamics of a creative process, we develop a planning approach that can continuously optimize the painting plan based on the evolving canvas with respect to the high-level goals. In contrast to existing approaches where the content generation process and action planning are performed independently and sequentially, FRIDA adapts to the stochastic nature of using paint and a brush by continually re-planning and re-assessing its semantic goals based on its visual perception of the painting progress. We describe the details on the technical approach as well as the system integration.
Planning, in FRIDA, is designed to achieve the high-level goals of the human user, which are specified with multi-modalities. FRIDA plans in a simulated environment created from real robot data and continually optimizes its plan based on visual perception to ensure the artist's intentions are realized.
We represent brush stroke shapes with 3 parameters: pressure, length, and bend.
Real strokes are painted from randomly sampled brush stroke parameters.
FRIDA models the relationship between the robot's movement and the appearance of the brush strokes more realistically than other differentiable simulators, such as DiffVG.
Strokes are rendered onto an image using only differentiable operations.
Loss functions compare the rendered, simulated painting to multiple modalities of input such as text, style images, and reference images.
Using stochastic gradient descent, FRIDA modifies randomly initialized brush stroke plans to minimize these loss functions in its simulated environment.
Once an initial plan is created, FRIDA begins to paint, periodically taking a picture of the current canvas and updating its plan to continue to decrease the loss functions.
