Collaborative Action Planning for Humanoid Robots Exchanging a Small Object

Abstract: This thesis focuses on the collaborative action planning for humanoid robot agents alone with each other or with a human agent. The action performed by these agents consists of exchanging a small deformable monochrome object using one arm. This situation occurs for instance when robots are employed to help elderly or disabled persons at home or retirement houses get an object difficult to reach, tidy rooms,etc. The used robots are NAOs made by Aldebaran Robotics. The thesis covers details about behavior trees which are a formal method used to plan the exchange and synchronize the NAOs, the Robot Operating System ROS which gives a global structure for the computer code, and NAOs with their software environment. This work also focuses on the image processing algorithms (HSV-lters, Connected-Components Labeling and Union-Find) that were used to detect an object and get its contour in the NAO's camera image, and the geometric applications that allow to get the 3D-position of a quasi-spherical object, knowing its dimensions and its projection on a camera focal plane.