Continual Area Sweeping

As mobile robots become increasingly autonomous over extended periods of time, opportunities arise for their use on repetitive tasks. We define and implement behaviors for a class of such tasks that we call continual area sweeping tasks. A continual area sweeping task is one in which a robot (or group of robots) must repeatedly visit all points in a fixed area, possibly with non-uniform frequency, as specified by a task-dependent performance criterion. Examples of problems that need continual area sweeping are trash removal in a large building and routine surveillance. We present a formulation for this problem and an initial algorithm to address it. The approach is analyzed analytically and is fully implemented and tested, both in simulation and on a physical robot.

Three movies referenced in the ICAR 2005 paper (for the ICRA 2006 movie, scroll down):

Configuration I - part 1
(4.2 MB MPEG)

In Configuration I - part 1, the robot starts with surveillance based on an equal probability of the ball appearing in any location. Then the balls are shown to the robot in two regions and the robot performs surveillance based on the new distribution of appearance of the balls.

Configuration I - part 2
(3.5 MB MPEG)

In Configuration I - part 2, the robot starts in the situation of the end of Configuration I - Part 1 (i.e. after learning), but since no balls are shown to the robot, it unlearns the distribution and starts to perform uniform surveillance. After that, balls are shown to the robot in new regions and the new distribution is learned by the robot.

Configuration II
(8 MB MPEG)

Configuration II demonstrates that the learning shown in Configuration I - part 1 generalizes to other positions of the walls. The robot is told the new locations of the walls, but uses identical control and learning code. In particular, it recalculates its own trajectories based on the wall locations and the locations at which the ball appears.

The movie referenced in the ICRA 2006 paper:

(4.2 MB MPEG)

In this configuration, one single robot starts sweeping the area. After a while another robot is added to the system. The new robot first takes responsibility for half of the field. After negotiation, its responsibility area reduces to less than one third of the field. When the added robot is removed, the other robot goes back to sweeping the whole area.

Full details of our approach are available in the following papers. The ICAR'05 paper presents the single-robot case and the ICRA'06 paper deals with the multirobot case.