Automating Adaptive Execution Behaviors for Robot Manipulation

Abstract

Robotic manipulation in semi-structured and changing environments requires systems with: a) perception and reasoning capabilities able to capture and understand the state of the environment; b) planning and replanning capabilities at both symbolic and geometric levels; c) automatic and robust execution capabilities. To cope with these issues, this paper presents a framework with the following features. First, it uses perception and ontology-based reasoning procedures to obtain the Planning Description Domain Language files that describe the manipulation problem at task level. This is used in the planning stage as well as during task execution in order to adapt to new situations, if required. Second, the proposed framework is able to plan at both task and motion levels, intertwining them by incorporating geometric reasoning modules to determine some of the symbolic predicates needed to describe the states. Finally, the framework automatically generates the behavior trees required to execute the task. The proposal takes advantage of the ability of behavior trees to be edited during run time, allowing adaptation of the action plan or of the trajectories according to changes in the state of the environment. The approach allows for robot manipulation tasks to be automatically planned and robustly executed, contributing to achieve fully functional service robots.

Keywords

• Adaptation
• behaviour tree
• knowledge-based reasoning
• manipulation planning