Closed-Loop Global Motion Planning for Reactive, Collision-Free Execution of Learned Tasks

Robotic manipulators have the potential to assist people with a variety of routine tasks in homes and workplaces. One key challenge for autonomously completing tasks in environments where people live and work is that objects in the environment are often not static-objects relevant to a task may move independently or be moved by people. A second key challenge is that many tasks must be performed in a manner that enforces constraints that humans are aware of from context and intuition: for example, the constraint that a glass of water must be kept approximately level when being carried in order to avoid spilling the water. A third key challenge is that homes and workspaces are often cluttered with (possibly moving) obstacles that must be avoided to safely perform a task. We present a motion planning approach that addresses each of these challenges by using a learned task model and executing in a closed-loop manner, enabling the robot to be responsive to the motion of task-relevant objects and obstacles that should be avoided.
Citation:
Chris Bowen and Ron Alterovitz, "Closed-loop Global Motion Planning for Reactive, Collision-free Execution of Learned Tasks," ACM Transactions on Human-Robot Interaction, vol. 7, no. 1, pp. 10:1-10:16, May 2018.
For more information:
robotics.cs.un...