A mobile manipulator robot that brings objects to assist people

Abstract

Purpose: The help a mobile manipulator robot provides to an individual can be beneficial if the robot can understand the needs of the person and adapt its actions to the behaviour of the user. This is especially true for elderly or disabled persons. When the user or the robot wishes to begin the action of the robot bringing an object to the user, first the robot must choose the object to bring; this is achieved when the robot uses the context of the situation and the dialog between the user to select the object. Then the robot must plan a trajectory for itself and its arms. This trajectory must avoid collision, while being safe and understandable by the human. The robot must sense motion of the user and adapt its initial trajectory based on the behaviour of the user. Finally, the last phase of exchanging the object is carried out through physical interaction between the robot and user. In this paper, we propose a complete example of such a task in a particular context. Method: As described in two research projects, PHRIENDS and SAPHARI, that examine the comfort and safety of interactive tasks using robots, the architecture of the robot, from mechanical design to software, must be prepared to allow the robot to address several issues simultaneously at all levels 1. To accomplish the assistance task during interaction with the user, the robot must be equipped with its own reasoning capability and be able to adapt to human needs and abilities 2. Symbolic and mental models are used to define the task to be achieved 3,4. This task is then refined using planners and scene models to define the grasping skills and trajectory of the robot and the associated control laws 5. The choice of the grasp technique greatly influences the task. A path is computed to define the motions of the robot, to move the object using Rapidly-Exploring Random Trees (RRTs) and stochastic optimization. Another research project, ASSIST, showed the necessity of taking into account HRI constraints that are needed to generate a trajectory for which the speed is adapted to the human attitude. Different controllers are used along the trajectory to adapt the robot's motions to the human movements and behaviours 6. During the execution phase, the robot rapidly and periodically checks and rechecks the moves and re-plans the task, to improve ultimate the solution. If a better trajectory is found, it is sent to the controller, which switches to the new trajectory. Results & Discussion: Such a system is realized with a PR2 robot (Figure 2). The robot assists the individual by providing objects on request, or by proposing objects to bring, that the robot believes may be useful for the human user. The initial results show the validity of this approach, and the usefulness of each element needed to achieve the task.