Robust Execution of Assembly Policies Using a Pose Invariant Task Representation

摘要

This paper discusses the robustness of executing robot tasks in contact with the environment. For example, in assembly, even the slightest error in the initial pose of the assembled object or grasp uncertainties can lead to large contact forces and, consequently, failure of the assembly operation. Force control can help to improve the robustness only to a certain extent. In this work, we propose using the position and orientation invariant task representation to increase the robustness of assembly and other tasks in continuous contact with the environment. We developed a variable compliance controller which constantly adapts the policy to environmental changes, such as positional and rotational displacements and deviations in the geometry of the assembled part. In addition, we combined ergodic control and vision processing to improve the detection of the assembled object's initial pose. The proposed framework has been experimentally validated in two challenging tasks; The first example is a mock-up of an assembly operation, where the object moves along a rigid wire, and the second is the insertion of a car light bayonet bulb into the housing.