Specification and utilization of manipulation skills to facilitate programming of bimanual manipulation tasks

The proposed method of robot manipulation planning uses a hierarchical approach for the decomposition of manipulation skills. Manipulation skill is a pattern of activity which describes an ability that achieves or maintains a particular goal. Manipulation skills are compositions of basic robot skills to reach some predefined goals. They constitute an interface between low-level constraint-based task specification and high level symbolic task planning. The task of the robot is approximated by a set of parameterized manipulation skills. The parameters are generally related to the task variations, such as: type of a motion, grasping rule, an initial and final configuration. Basic skill is an action that implements a sensory-motor coupling using a single low-level controller.

Grasp synthesis for a multi-fingered robotic hand under object pose uncertainty

In the proposed method of grasp planning, both, object pose uncertainty and object dynamics are considered. These two factors greatly affect success or failure in a real-world robotic grasping and should be considered simultaneously. We use a probabilistic distribution model for the object pose error and incorporate object dynamics along with three metrics into the grasp quality evaluation. The proposed approach is based on simulation of grasping process assuming that the 3D model and dynamics model of the object are known. The key advantage of grasp synthesis using such a simulation is the ability to remove grasps that might seem analytically promising but would fail if executed in reality.

Grasp implementation

Planning of the manipulator trajectory in an unstructured environment is done by taking into account the depth sensor data. Grasp planning is performed by taking into account the expected external forces acting on the grasped object. This methodology was tested on an example of removing the jar cap by two-handed robot.

Experimental systems utilizing position-force and impedance control

For the purpose of manipulation task execution, a task specification and decomposition method was developed and utilized. Manipulation tasks were executed by two different types of robotic systems. The first one was a two arm system composed of two modified industrial manipulators with joint position controllers and additional force/torque sensors mounted in the wrists. Inertial measurement units (IMU) were mounted next to the force/torque sensors. Eye in hand cameras and standalone cameras supplemented the system. The second system was the Velma robot composed of two redundant 7DOF torque controlled arms, active torque controlled torso and position controlled 2DOF head containing a stereo-pair and an RGBD camera. In the case of the first system the virtual effectors controlling the manipulators exhibited four elementary behaviours: idle behaviour, contact with the environment, pure position control, and position control with contact expected. The virtual effectors controlling the manipulators of the second system provided the following behaviours: idle behaviour, extended impedance control on top of the hierarchical torque controller. To improve the quality of force/torque measurements a method taking into account the inertial forces obtained from the IMU readings was developed. Several robot programming frameworks (MRROC++, ROS, OROCOS, Discode) were used to implement and verify the approach.