A Learning Method for a Daily Assistive Robot for Opening and Closing Doors Based on Simple Instructions

Remote-controlled daily assistive robots that can help handicapped individuals are expected to be a key technology in supporting aging communities. We present a method to simplify user instruction via a human-machine interface for robotic manipulation of various types of doors. Using our system, the user can teach the robot to open an unknown door by just a few clicks on an image generated from the robot's perspective. After the teaching stage, the robot is able to manipulate the door based on just a single user click on the position of the door handle, and an instruction determining if the door is to be opened or closed. The contributions of this paper to the research area are to (i) present a method to detect door candidates effectively from the point cloud of the surrounding environment using hypotheses on the plane and the shape of doors, to (ii) introduce a door feature to learn and remember the manipulation model, and to (iii) implement the system on a daily assistive robot and evaluate it. We tested our system on doors with various shapes and manipulation models, obtaining results supporting the effectiveness of our approach.