Copyright (C) 2016-2017 Marcus A. Johansson
This is a snapshot of the HiQP source code as of after having finished my M.Sc. thesis. It is kept here as a showcase of my final M.Sc. work. For the real HiQP source code, please checkout the OrebroUniversity profile at GitHub.
The package is showcasing the use of the HiQP control framework, see [3], with the ABB YuMi robot in simulation via a collection of shellscripts. Dependencies include the hiqp packages (git@github.com:OrebroUniversity/hiqp.git), as well as the yumi_description package (git@github.com:OrebroUniversity/yumi.git) which is also available from APT. Early videos showing HiQP used for autonomous grasping and assisted teleoperation respectively can be found at: https://www.dropbox.com/s/2z4v4dn2u7w7sm0/yumi_autograsp.mp4?dl=0, https://www.dropbox.com/s/vmlgjishsw1d0v4/yumi_teleop.mp4?dl=0.
- hiqp_demos/launch/yumi_simulation.launch: launch file starting a Gazebo simulation of YuMi and the HiQP control framework
- hiqp_demos/config/yumi.yaml: ros_control parameter configuration for the HiQP control framework
- hiqp_demos/config/yumi_preload.yaml: this file offers the possibility to start the control framework with preloaded tasks for, e.g., limiting joint positions/velocities and obstacle avoidance
- hiqp_demos/scripts: various shell scripts containing exemplary servicealls to add/remove tasks and corresponding geometric primitives
Launch hiqp_demos/launch/yumi_simulation.launch to start a Gazebo simulation and the HiQP velocity controller. By uncommenting <rosparam file="$(find hiqp_demos)/config/yumi_preload.yaml" command="load"/> in the launch file, a set of tasks defined in hiqp_demos/config/yumi_preload.yaml can be pre-loaded at start-up. Due to imperfect joint velocity tracking in Gazebo, the model is prone to drift in velocity control mode. It is therefore recommended to disable gravity in Gazebo via hiqp_demos/scripts/utility/disable_gravity_gazebo.sh.
HiQP operates on geometric primitives, which can be attached to arbitrary frames, and uses them to formulate the following task classes:
- TDefJntConfig: specifies a desired joint configuration
- TDefJntLimits: specifies joint position/velocity limits
- TDefGeomProj: Expresses distance relationships between primitive geometries which are used for movement generation
- TDefGeomAlign: Expresses orientation relationshps between primitive geometries which are used for moment generation
Each task is assigned a priority. Tasks on lower hierarchy levels are fulfilled as good as possible (in a least-square-sense) in the null-space of higher ranked tasks, i. e., if redundancy is available.
In between examples, running tasks and associated primitive geometries can be removed with the convenience shell script in hiqp_demos/scripts/utility/clear_controller.sh. More detailed comments can be found in the corresponding shellscripts themselves.
- hiqp_demos/scripts/joint_configuration.sh: achieves a given joint configuration with defined dynamics.
- hiqp_demos/scripts/point_plane_projection.sh: moves an end-effector point anywhere on a given target plane.
- hiqp_demos/scripts/sphere_plane_avoidance.sh: demonstrates obstacle avoidance in HiQP. The end-effector point is moved towards the target plane as good as possible, such that the defined collision sphere is kept above the specified obstacle plane.
- hiqp_demos/scripts/line_line_alignment.sh: shows how to prallelly align two line primitives attached to the robot's kinematic chain.
- hiqp_demos/scripts/frame_frame_pose.sh: moves a frame attached to one of the robot's wrist to coincide with a predefined target frame.
- hiqp_demos/scripts/reach_to_grasp.sh: encapsulates a simplified version of the set-up used in [1, 2] to generate reach to grasp movements on-the-fly.
[1] ... R. Krug, T. Stoyanov, V. Tincani, H. Andreasson, R. Mosberger, G. Fantoni, and A. J. Lilienthal. The next step in robot commissioning: Autonomous picking and palletizing. IEEE RA-L, 1(1):546–553, 2016.
[2] ... T. Stoyanov, R. Krug, R. Muthusamy, and V. Kyrki. Grasp envelopes: Extracting constraints on gripper postures from online reconstructed 3D models. In Proc. IEEE/RSJ IROS, pp. 885–892, 2016.
[3] ... M. A. Johansson, Online whole-body control using hierarchical quadratic programming: implementation and evaluation of the HiQP control framework. MSc Thesis, 2016.