We consider how the tedious chore of folding clothes can be performed by a robot. At the core of our approach is the definition of a cloth model that allows us to reason about the geometry rather than the physics of the cloth in significant parts of the state space. We present an algorithm that, given the geometry of the cloth, computes how many grippers are needed and what the motion of these grippers are to achieve a final configuration specified as a sequence of
—folds that can be achieved while staying in the subset of the state space to which the geometric model applies. G-folds are easy to specify and are sufficiently rich to capture most common cloth folding procedures.We consider folds involving single and stacked layers of material and describe experiments folding towels, shirts, sweaters, and slacks with a Willow Garage PR2 robot. Experiments based on the planner had success rates varying between 5/9 and 9/9 for different clothing articles.