In this paper, we present a novel heuristic-based approach to flatten wrinkled garments by means of autonomous robotics. We have designed a heu-ristic-based strategy to flatten crumpled cloth by eliminating visually detected wrinkles. In order to explore and validate visually guided clothing manipulation, we have developed a hand-eye interactive learning system that incorporates a clothing simulator to close the effector-garment-visual sensing interaction loop. We also propose a criterion to evaluate the various approaches used to flatten cloth. In this paper, our heuristic-based method is applied to virtual cloth in our simulator and the resulting flattening performance is compared to that obtained by manual flattening methods. These experiments demonstrate that the effec-tiveness and efficiency of our heuristic-based garment flattening methods ap-proach that of manual flattening. 1 Introduction With the rapid development of robotic manipulation technology, low-cost range sensing devices and state-of-the-art computer vision algorithms, the potential now exists to devise an autonomous robotic system capable of performing clothing manip-ulation for laundry service tasks. Interacting with clothing requires dexterous manip-ulation and hand-eye coordination capabilities that come naturally to humans but represent a challenge to current autonomous robotic systems. Such robot capabilities have matured to some extent as demonstrated by [2], [3]; however, several limitations still remains that must be circumvented before autonomous robotic laundry systems become viable. Specifically, current research and development in perception and manipulation tasks for robotic laundry systems consist of isolating clothes from a heap [4], [5], [7], performing clothes classification [6] and then folding the clothes [2], [3], [7]. Re-searchers usually assume that garments are in a tractable state, but do not measure the state of the garment being flattened prior to applying any folding procedure. There-fore, garment or cloth flattening has not been fully considered in current robotics developments. In this paper, we devise and demonstrate a novel heuristic-based