QoLT Systems: Active Home

Project Lead:  Chris Atkeson, Carnegie Mellon

Our current goal in the Active Home project is to develop a robotic assistant for household tasks involving manipulation. The Active Home project is one of our QoLT Systems, whose purpose is to facilitate development of core technology within and across thrusts. A fundamental research barrier we addressed this year is grasping in cluttered environments, like a kitchen or living room. Most previous research has looked into manipulation and planning while making several unrealistic assumptions like specifying where and how to grasp the object of interest or where the robot should move before starting to manipulate the object. It is unrealistic for a human user interacting with the robot to specify all of this information before the robot can even move. The main area of focus of our recent work is to build algorithms to allow the robot to autonomously move and grasp objects.

We divide the above problem into two separate components: 1) grasp set generation and 2) motion planning. For grasp set generation, the objective is to generate a set of grasps given a target object in a known environment under the constraints that the grasps are stable and that the robot’s hand does not collide with obstacles in the environment when grasping. For motion planning, the goal is to generate a trajectory for the robot to move to the object and complete one of the grasps in the set, again without collision.  Both components require complex geometric reasoning on the part of the robot, however, if the robot is to be practical, this reasoning must be done quickly and efficiently.