Skip to main content
mdp_cfg.yaml is necessary for defining the MDP component settings. This includes parameters related to the high level commands given to the agent (like base velocity commands or end-effector commands), lower level actions provided to each of the DOFs (like relative change pose and binary gripper control), the set of observations fed to the policy network, termination conditions for each episode, reward design, and curriculum utilized in the training methodology. The following example depicts the MDP settings for training a Unitree Go2 that receives velocities commands and executes them based on a locomotion policy. 
actions:
    - joint_actions:
        type: JointPosition
        config:
            asset_name: robot
            joint_names: [".*"]
            scale: 0.5
            use_default_offset: True

commands: 
  - base_velocity:
      type: VelocityCommand
      config:
        asset_name: robot
        rel_standing_envs: 0.02
        rel_heading_envs: 1.0
        heading_control_stiffness: 0.5
        debug_vis: True
        heading_command: True
        resampling_time_range: [10.0, 10.0]
        ranges:
            lin_vel_x: [-1., 1.]
            lin_vel_y: [-1., 1.]
            ang_vel_z: [-1., 1.]
            heading: [-3.14, 3.14]

observations:
  - policy:
    - base_lin_vel: 
        type: base_lin_vel
    - base_ang_vel: 
        type: base_ang_vel
    - projected_gravity:
        type: projected_gravity
    - velocity_commands:
        type: velocity_commands
    - joint_pos_rel:
        type: joint_pos_rel
    - joint_vel_rel:
        type: joint_vel_rel
    - actions:
        type: actions
        config:
            params:
                action_name: joint_actions
    - height_scan:
        type: height_scan

terminations:
  - time_out:
      type: time_out
  - body_contact:
      type: body_contact
      config:
        params:
          threshold: 1.0
  - terrain_out_of_bounds:
      type: terrain_out_of_bounds
      config:
        params:
          distance_buffer: 3.0

events:
  - add_base_mass:
      type: add_base_mass
      config:
        params:
          mass_distribution_params: [-2.5, 2.5]
          operation: add
  - base_external_force_torque:
      type: base_external_force_torque
      config:
        params:
          force_range: [0.0, 0.0]
          torque_range: [-0.0, 0.0]
  - reset_base:
      type: reset_base
  - reset_robot_joints:
      type: reset_robot_joints
      
rewards:  
  - air_time:
      type: air_time
      config:
        weight: 5.0
        params:
          mode_time: 0.3
          velocity_threshold: 0.5
  - base_angular_velocity:
      type: base_angular_velocity
      config:
        weight: 5.0
        params:
          std: 2.0
  - base_linear_velocity:
      type: base_linear_velocity
      config:
        weight: 5.0
        params:
          std: 1.0
          ramp_rate: 0.5
          ramp_at_vel: 1.0
  - foot_clearance:
      type: foot_clearance
      config:
        weight: 0.5
        params:
          std: 0.05
          tanh_mult: 2.0
  - action_smoothness:
      type: action_smoothness
      config:
        weight: -1.0
  - air_time_variance:
      type: air_time_variance
      config:
        weight: -1.0
  - base_motion: 
      type: base_motion
      config:
        weight: -2.0
  - base_orientation:
      type: base_orientation
      config:
        weight: -3.0
  - foot_slip:
      type: foot_slip
      config:
        weight: -0.5
  - joint_pos:
      type: joint_pos
  - joint_torques:
      type: joint_torques
      config:
        weight: 1.0
If you want to take a look at the possible set of options for all MDP components available at your disposal, refer to this page.
⌘I