Title: Rendering, Replicating, and Adapting Human Motions on a Cable Robot for Artistic Painting
Date: Monday, August 19
Time: 1:00 PM – 3:00 PM ET
Location: Klaus 1315 / google meet
Gerry Chen
Robotics Ph.D. Student
School of Interactive Computing
Georgia Institute of Technology
Committee
Dr. Frank Dellaert (Co-advisor) – IRIM, Interactive Computing, Georgia Institute of Technology
Dr. Seth Hutchinson (Co-advisor) – IRIM, Interactive Computing, Georgia Institute of Technology
Dr. Danfei Xu – IRIM, Interactive Computing, Georgia Institute of Technology
Dr. Jun Ueda – IRIM, Mechanical Engineering, Georgia Institute of Technology
Dr. Jean Oh – Robotics Institute, Carnegie Mellon University
Abstract
Artists have continually pushed their crafts to embody the furthest reaches of human capabilities, from delicate painting to athletic performances, highlighting the potential for robots to emulate these skills. This work aims to study the task of robot graffiti painting in three parts: rendering, replicating, and adapting human motions, ultimately contributing to the fields of robot art, cable robot control, motion planning, and generative modeling.
In this work, three parts to the problem of artistic painting guided by human motions are addressed: rendering a digital artwork in paint with a cable robot; replicating human input motions as closely as possible; and adapting human input motions to accommodate for differences in level of detail, style, and artistic medium. Through the difficult, interaction-rich task of robot art, modern challenges in human-robot collaboration can be studied. In particular, techniques for robot motion control through natural input interfaces drawn from human motions are developed. Rendering paint requires advances in state estimation and control techniques for fast, fluid motions on a cable robot. Replicating human motions bridges the input motions and robot kino-dynamic capabilities, requiring advances in optimal trajectory retiming techniques. Finally, adapting goes beyond rote replication by augmenting input motions to better fit the composition, style, and medium intended by the robot-artist team, requiring embodiment-specific painting motion generation and sketch retargeting. Put together, the thesis forms a cohesive body of work producing human-robot paintings and making novel contributions to the fields of robot art, human-robot collaboration, and cable robot control.