RNB-Control is a control framework to enable agile development of control algorithms for robot systems.

• ControlHub enables runtime replacement of the control algorithm and easy gain tuning with Web UI.
• A motion trajectory can be passed to the controller easily by TrajectoryInterface. You can pass a low-frequency trajectory because the TrajectoryInterface interpolates it to high frequency. On-line trajectory tracking is also available. The python example client provides a function that transforms a sequence of waypoints to a trajectory with fixed timesteps.

• Branching strategy: gitflow
• Coding style: PEP8
• Comment style: doxygen

• OS: Linux, Windows

• Linux - cmake
  • make a symbolic link to controlhub folder to your hardware control project.

  cd YOUR_CONTROL_PROJECT_SRC
  ln -s RELATIVE_PATH_TO_CONTROL_HUB ./control_hub

  • example (PandaControlRNB)

  cd ~/Projects/rnb-control/projects/PandaControlRNB/src/panda_control
  ln -s "../../../../control_hub" ./control_hub

• Windows - Eclipse (Assume rnb-control/projects is the workspace folder)
  • Right click the project and go to: New -> Folder -> Advanced -> Link to alternate location
  • Enter PROJECT_LOC\..\..\control_hub and Finish.
  • Right click the project and go to: Properties -> C/C++ General -> Paths and Symbols -> includes
  • For both GNU C and GNU C++ tab, add ${ProjDirPath}\..\..\control_hub.
  • Repeat above for all build configurations you use
• Windows - VisualStudio
  • Right click the project and Add -> New Filter -> control_hub
  • Drag and drop all contents of control_hub into the created filter

1. Create a ControlHub instance. The control mode(Joint/Task), joint degree of freedom, task degree of freedom, and control step time should be specified.
2. Call ControlHub::reset_controller(Q) when ever the hardware calls reset.
3. Inside the control loop, call ControlHub::check_nominal_reset() first. This will return true while and just after the controller is reset. Reset your nominal model too when this function returns true.
4. Call ControlHub::calculate_joint_control_torque() or ControlHub::calculate_task_control_torque() in the control loop, depending on the control mode. Check the function comments or doxygen document for the parameter information.

• To prevent needless computations, ControlHub has 4 flags for information usage. Each controller will have this values to specify if it uses certain information. For example, use_nominal_dynamics and use_nominal_kinematics: are False for PD controller, as it does not use nominal model. Thus, compute for these values selectively based on these flags.
  • use_real_dynamics: The controller use real robot's dynamic parameters (M, C)
  • use_nominal_dynamics: The controller use nominal dynamic parameters (M_n, C_n)
  • use_real_kinematics: The controller use nominal dynamic parameters (J, Jdot)
  • use_nominal_kinematics: The controller use nominal dynamic parameters (J_n, Jdot_n)

The TCP/IP based trajectory interface is included in the ControlHub, but if you need trajectory interface only, you can use the trajectory client separately. TrajectoryInterface includes Cspline interpolation and online tracking features.

1. Create a TrajectoryInterface instance. The control mode(Joint/Task), trajectory degree of freedom, and control step time should be specified.
2. Call TrajectoryInterface::update_position(Q) to give information about current robot pose.
3. Call TrajectoryInterface::reset_traj() when the hardware calls reset. Make sure at least one update_position is called before any reset call, sothat the TrajectoryInterface can be reset with recent robot pose information.
4. Call TrajectoryInterface::get_next_qc() in the control loop to get current target pose, velocity, acceleration.

• TrajectoryInterface::follow_traj is a flag that is set with the client's "follow" command and cleared with "stop" command. The trajectory is not tracked when this flag is not set.

• Follow instruction in projects/IndyControlRNB/README.md

• Follow instruction in projects/PandaControlRNB/README.md

• Follow instruction in projects/BulletSimControl/README.md

• Control WebUI is on http://{ROBOT-IP}:9990/ after launching robot or you can access WebUI through python (see scripts/WebClient Example)
• Selecting controller
  • Select : it changes current controller to chosen controller
  • Set Default : it allows current controller to be a default controller
• Gain tuning
  • Apply : it applies current state of gains
  • Save / Load : it saves current gains and loads saved gains
• Data plot/save
  • Apply : it regulates time step of plots
  • Restart Log / Pause Log : it starts and stops showing plots
  • Download Full : it save data from the moment you press this button to 30 secs ago. Before download saved data, pause log at the first
  • Assign values to custom_dat (JOINT_DOF) inside you controller, to plot your specific internal variables.
• When logging data, conflicts may occur because of high control frequency
  • solution : recommend to decrease below 2000Hz and control frequency can be regulated in "indyDeploy.json" (same path of TaskManager)
• When start logging, high calculation time of controller(NRIC_force) makes error and robot stops
  • solution : after changing another controller(PD, NRIC_PD, ...), restarting log and waiting a few seconds, then change to desired controller
• [NOTE] To plot data, assets folder should be copied to the program working directory.

• Open trajectory_clilent_python/trajectory_client_example.ipynb from jupyter notebook.
• Use move_joint_s_curve to move robot to target position.
  • To follow the original trajectory generated by Indy framework, call stop_tracking or enable auto_stop when calling move_joint_s_curve.
• Use move_joint_wp to pass multiple waypoints.
• move_joint_s_curve_online tests online trajectory following for adaptive motion. Check inside the function to implement your own adaptive robot motion.
• Follow instructions in
  • with IndySDK : trajectory_clilent_python/trajectory_client_example_Indy.ipynb
  • with Panda : trajectory_clilent_python/trajectory_client_example_Panda.ipynb

• As there are many restrictions in debugging with realtime HW systems, controllers/tester is provided to debug programmatic errors in the controller. This does not do dynamic or kinematic computation. It only pass specific trajectories to the controller to check if any numeric error is generated
• Follow instructions in controllers/tester/ReadMe.md for detailed usage.
• MAKE SURE YOU DON'T PUSH CHANGES ON THIS CODE.

• To generate doxygen document, run doxywizrd and open Doxyfile, click "Run doxygen" in "Run" tab.
• Doxygen documentation can be locally browsed from doxygen/html/index.html
• To host doxygen documents, run following command (port=5000 by default)

  python ./host_doxygen.py

• Now you can browse doxygen documents from other computers in the local network at <IP>:<PORT>