Crawling Hand Project

Instructions

• This repo is based on MuJoCo 2.3.6
• Create a new Conda env with Python>= 3.8
• Install mujoco by pip installl mujoco

Visulization

• In mujoco conda env, python -m mujoco.viewer --mjcf=five_finger_hand_ssss.xml
• With ROS, roslaunch urdf_tutorial display.launch model:=URDF_finger_ssss.urdf

Explanation of code

• gait_test/GA_crawling_full_test_new.py, the optimal structures achieved by optimizing the link lengths and finger placements for grasping one, two, or three objects, respectively.
• *_replay_*.py replay the crawling in simulation.

Notes

• Under folder singer_finger/, URDF_finger_xxxx.urdf is the urdf file for a single finger.
• For six fingers, see descriptions/six_fingers_llll.xml, which makes 6 copies of the single finger.
• To use pytorch_kinematics, modify the function build_chain_from_mjcf in mjcf.py by m = mujoco.MjModel.from_xml_path(data) (line 75)
• Load the model from .xml by

xml_path = 'descriptions/five_finger_hand_ssss.xml'
chain = pk.build_chain_from_mjcf(xml_path, 'hand')

Real robot control (with ROS)

• install ROS2. I use ros-iron
• build from source of dynamixel with ROS2.
• make sure that we can connect the control all motors in Dynamixel Wizard. Scan and enable torque control to test each motor.
• source ~/research/lasa/open_source/crawling_robot/dynamixel_new/install/setup.bash
• launch the robot ros2 launch pantilt_bot_description pantilt_bot.launch.py. The usb_port, baud_rate, id in ros2_control.xacro should be the same as Dynamixel Wizard
• to switch controllers, ros2 control switch_controllers --activate joint_state_broadcaster --activate joint_trajectory_controller --deactivate velocity_controller
• run control_interface_real_robot.py for an example.
• For velocity control ros2 control switch_controllers --deactivate joint_trajectory_controller

Dynamixel control (without ROS)

• Current-based position control https://github.com/leap-hand/LEAP_Hand_API 8ms for reading position only and sending position commands
• https://github.com/AlanSunHR/DynamixelController has full access to all parameters of motors.

KDL

conda install -c conda-forge python-orocos-kdl for PyKDL from https://github.com/orocos/orocos_kinematics_dynamics
pip install urdf-parser-py
conda install -c conda-forge matplotlib
conda install -c conda-forge quaternion conda install numpy scipy numba

Grasping stability in MuJoCo

The following parameters are important:

    <option>
        <flag gravity="enable" multiccd="enable" />
    </option>
 <option cone="elliptic" impratio="10"/>

we also need to set the friction coefficients

Steps to connect motors

• Give access to USB port /dev/ttyUSB0 and /dev/ttyUSB0 by sudo chmod -R 777 /dev/ttyUSB0
• Open Dynamixel Wizard and scan motors
• Set the correct USB port number and baudrate
• run detach_mechanism/key_command_tester.py to control the attachment/detachment mechanism