Task-level teleoperated manipulation for the HRP-2Kai humanoid robot

This paper presents the strategy used by the team AIST-NEDO at the DARPA Robotics Challenge to deal with the designated manipulation tasks by means of a task-level teleoperation of the HRP-2Kai humanoid robot, considering a disaster-hit scenario that is inherently non-structured and a limited communication between the user and the robot. The strategy, based on the information provided by a laser rangefinder and a set of cameras installed at the head and at both hands, consisted in the alignment of 3D models representing the desired manipulation targets with a measured point cloud, in order to provide a reference frame to describe the manipulation motion required for each task. Each motion was carefully planned in advance by assuming minimum information of the object representing the manipulation target. In order to exemplify the before mentioned approach, two representative tasks of the DARPA Robotics Challenge are described, as well as the corresponding results obtained during the competition.