2A2-Q13 Dressing assistance with failure recovery actions based on cloth states' recognition(Robots for Home/Office Application)

Abstract

The goal of this work is the development of a robotic system for dressing assistance, which is one of the toughest tasks for carers. Because of clothes' flexibility, manipulation of clothes easily fails. For example, a robot cannot pull up pants if the pants catch the toe of a care-receiver. In our system, the current state of the clothes is recognized using optical flows, which illustrates the expansion and contraction of the clothes. The system compare the current flows to the ones in a preformed database, which stores many visual features at success/failure states, to decide if the robot should redo the motion with another trajectory. We tested our system in an experiment to put pants on persons by a humanoid robot. These results suggested the effectiveness of our approach.