Learning to slide a magnetic card through a card reader

摘要: This paper describes a set of experiments in which an upper-torso humanoid robot learned to slide card through reader. The small size and the flexibility presented number manipulation challenges for robot. First, because most is occluded by reader robot's hand during sliding process, visual feedback useless this task. Second, bends easily, it difficult distinguish between bending hitting obstacle order correct trajectory. To solve these proposes method constraint detection that uses only proprioceptive data. dynamic joint torque thresholds are calibrated using movements free space. experimental results show method, can detect when movement constrained modify trajectory real time, makes solving task possible.