PRIMITIVE BASED HIERARCHICAL PLANNING FOR HUMANOID ROBOTS

The key for robust and reliable usage of humanoid robots in various and unstructured environments relies on the planning and control capabilities. Moreover, different levels of autonomy are required, depending on the scenario and the task set to be accomplished. In general higher level planning is executed at first, and each action (at lower levels) is subsequently planned. Nevertheless, in case of specific tasks where human decision is still preferred or in case of robot misbehavior, remote control of the robot is needed. In this thesis a hierarchical planning and control framework for humanoid robots is presented. The purpose of the proposed algorithms is to allow transitions from gross to fine motor skills and to act at different levels, such as joint, cartesian, primitive and task levels. Primitive and task level planning algorithms based on control primitives have also been implemented for both mobile manipulators and humanoid robots. Several examples of motion primitives are proposed for both robotics platforms. Furthermore, a Motion Description Language is used for both the formal validation of the obtained plans and for the generation of consistent complex plans starting from previously defined motion primitives. In case of failure of one of the planning levels, a human can remotely intervene to avoid undesired robots behaviors substituting the planner through the proposed Pilot Interface. This is a Graphical User Interface that acts as a natural connection between the different planning level and the human that in case of necessity can decide the correct level of autonomy of the robot (and hence the level of planning) to accomplish a task. Finally, both the planning algorithms and the Pilot Interface have been experimentally validated and applied on different robots and scenarios.