WoLF: Whole-body Locomotion Framework for quadruped robots

This repo contains a collection of different robots and sensors used in WoLF.

Setup

See the documentation here.

How to add a new robot

To add a new robot you can start by copying the structure of an existing one. The robot descriptions are defined as ROS packages, so you will need to add it to your ROS workspace. The required files in the package are the following:

• params/robot_name_params.yaml it contains the controller parameters, use this file to tune a robot.
• robots/robot_name.urdf.xacro it is the robot URDF description describing the kinematics and dynamics of the robot.
• robots/robot_name.srdf.xacro it is the robot SRDF description, use this file to define the kinematic chains and default postures.

You can add plotjuggler and rviz visualization files in the respective folders plotjuggler and rviz. The folder urdfs can be used to store extra xacro files useful for the robot description. And last but not least, don't forget the robot's meshes in the meshes folder, unless your robot is a stealth one :)

To launch the new robot with the WoLF controller:

roslaunch wolf_controller wolf_controller_bringup.launch robot_name:=new_robot

Parameters description

• default_duty_factor defines the ratio between the stance and total cycle time: T_stance / T_cycle.
• default_swing_frequency defines the frequency of the swing: 1 / T_swing.
• default_contact_threshold defines the force magnitude that is used to detect if the robot is contact with the enviroment or not.
• default_step_reflex_contact_threshold if the step reflex is active, it represents the magnitude of the contact force that triggers a step reflex, by default it is calculated as default_contact_threshold * 1.5.
• default_step_reflex_max_retraction if the step reflex is active, this is the maximum step retraction length, by default it is calculated as max_step_height/2.0.
• default_base_linear_velocity_x linear velocity along robot's x axis.
• default_base_linear_velocity_y linear velocity along robot's y axis.
• default_base_linear_velocity_z linear velocity along robot's z axis.
• default_base_linear_velocity defines a common velocity value along xyz axis, it overrides the previous linear velocities.
• default_base_angular_velocity_roll angular velocity along robot's roll axis.
• default_base_angular_velocity_pitch angular velocity along robot's pitch axis.
• default_base_angular_velocity_yaw angular velocity along robot's yaw axis.
• default_base_angular_velocity defines a common velocity value along roll pitch and yaw axis, it overrides the previous angular velocities.
• default_step_height defines the step height.
• default_cutoff_freq_gyroscope cut off frequency for the gyroscope filter.
• default_cutoff_freq_qdot cut off frequency for the joint velocities filter.
• default_friction_cones_mu friction cones mu.
• max_step_height defines the maximum step height allowed.
• max_step_length defines the maximum step length allowed.
• max_base_roll defines the maximum rotation along the robot's roll axis.
• min_base_roll defines the minimum rotation along the robot's roll axis.
• max_base_pitch defines the maximum rotation along the robot's pitch axis.
• min_base_pitch defines the minimum rotation along the robot's pitch axis.
• estimation_position_type [ground_truth (simulation only) | estimated_z] .
• estimation_orientation_type [ground_truth (simulation only) | imu_magnetometer | imu_gyroscope] .
• use_contact_sensors [true | false] use the hardware_interface::ContactSwitchSensorInterface if available.
• compute_odom [true | false] compute the odometry i.e. create the odom -> base_footprint transformation.
• activate_step_reflex [true | false] activate the step reflex.
• activate_push_recovery [true | false] activate the push recovery.
• initial_pose_simulation Initial robot's pose in simulation.
• activate_com_z [true | false] activate the control of the robot's height through the CoM task, otherwise use the waist task. By default it is true.
• activate_postural [true | false] activate the postural task (the postural task is in charge of keeping the robot close to the standup pose), by default it is false.
• activate_angular_momentum [true | false] activate the angular momentum task, by default it is true.
• activate_joint_position_limits [true | false] activate the joint position limits constraint (the values are loaded from the URDF), by default it is false.
• regularization regularization value for the solver, by default it is set to 1e-3 for the accelerations and 1e-6 for the forces.
• min_forces_weight contact forces minimization task weight (higher values will reduce the contact forces produced by the solver), by default it is 0.0, i.e. not active.
• min_qddot_weight joint accelerations minimization task weight (higher values will reduce the joint accelerations produced by the solver), by default it is 0.0, i.e. not active.

gains:

      Kp_leg: {haa: xxx, hfe: xxx, kfe:xxx}
      Kd_leg: {haa: xxx, hfe: xxx, kfe: xxx}

      xx_foot:
          Kp: {x: xxx, y: xxx, z: xxx, roll: 0, pitch: 0, yaw: 0}
          Kd: {x: xxx, y: xxx, z: xxx, roll: 0, pitch: 0, yaw: 0}
          type: [force|acceleration]
          weight: xxx

      waist:
          Kp: {x: 0.0, y: 0.0, z: xxx, roll: xxx, pitch: xxx, yaw: xxx}
          Kd: {x: 0.0, y: 0.0, z: xxx, roll: xxx, pitch: xxx, yaw: xxx}
          type: [force|acceleration]
          weight: xxx

      CoM:
          Kp: {x: xxx, y: xxx z: xxx}
          Kd: {x: xxx, y: xxx, z: xxx}
          weight: xxx

      angular_momentum:
          weight: xxx

      postural:
          weight: xxx

• Kp_leg and Kd_leg define the impedance gains for the the hip abduction/adduction (haa), hip flexion/extension (hfe) and knee flexion/extension (kfe) joints in each leg. These gains are used for different things:
  • to zero the joints during the init phase.
  • to perform a controlled shutdown in case of anomaly (using only the damping value of the impedance).
  • to set the gains for the postural task.
• The default tasks in WoLF are the following:
  • xx_foot these tasks allow the robot to track the swing trajectories with its legs. Since we are working with a quadruped we usually have 4 of them defined as left front (lf), right front (rf), left hind (lh) and right hind (rh).
  • waist this task is used to control the attitude of the robot's base/trunk and its height if activate_com_z is false.
  • CoM this task is used to stabilize the Center of Mass (CoM) of the robot with respect to its support polygon along x and y, and control the CoM height if activate_com_z is true.
  • postural this task is used to impose a preferred joint configutation to the robot, by default the stand up posture is used. This task can be activated with activate_postural set to true.
  • angular_momentum this task is used to counteract angular momentum variations generated on the robot's base/trunk by disturbances. This task can be activated with activate_angular_momentum set to true.
  • xx_arm if the robot has an arm mounted on top, it is possible to add a task to control the arm end-effector position and orientation.
• tasks' gains:
  • Kp and Kd define the proportional and derivative gains for each task.
  • type [force|acceleration] by default the tasks' gains are defined as force gains meaning that Kp and Kd work as Cartesian impedance gains. Please check Appendix A in the following paper for a better explanation of Cartesian impedance.
  • weight relative task weight. A higher value means that the selected task has a higher priority compared to the others and therefore the solver will try to minimize this task first.

How to reset the robot pose and posture

To reset the robot pose in gazebo you can simply use the gazebo's shortcut ctrl-r, this command should set the robot pose to the origin. Then, to reset the robot's posture run the following:

rosrun wolf_description_utils go0

now your robot should be ready to stand up again!

How to define the standup and standdown posture

To define the stand up and stand down posture in the SRDF file, you can use the joint_state_publisher by launching the following command in your terminal:

roslaunch wolf_description_utils standalone.launch robot_name:=your_robot

Disclaimer

Some robots have been tuned better than others, please don't be sad if your favorite robot doesn't work perfectly. At the moment, the robots we feel more confident with are:

• Spot
• Go1

Feel free to tune and share your parameters with us if you feel they improve the control of a specific robot :)

Extra robot descriptions are available here.

Legal notes

wolf_description_utils and sensors_description are licensed under a license GNU General Public License v3.0. For the robot descriptions all rights belong to their respective owners.