error_vel_xy and error_vel_yaw too large

[154332]

################################################################################
Learning iteration 18459/20000

                   Computation: 3673 steps/s (collection: 25.992s, learning 0.767s)
         Mean action noise std: 1.28
      Mean value_function loss: 0.1706
           Mean surrogate loss: -0.0124
             Mean entropy loss: 19.1296
                   Mean reward: 97.40
           Mean episode length: 1000.00

Episode_Reward/track_lin_vel_xy_exp: 1.8623
Episode_Reward/track_ang_vel_z_exp: 1.2540
Episode_Reward/lin_vel_z_l2: -0.0239
Episode_Reward/ang_vel_xy_l2: -0.3513
Episode_Reward/base_height_l2: -0.0542
Episode_Reward/joint_torques_l2: -0.0372
Episode_Reward/joint_acc_l2: -0.5073
Episode_Reward/joint_pos_limits: -0.0516
Episode_Reward/joint_power: -0.0079
Episode_Reward/stand_still: -0.0542
Episode_Reward/joint_mirror: -0.0430
Episode_Reward/action_rate_l2: -0.7033
Episode_Reward/undesired_contacts: -0.1797
Episode_Reward/contact_forces: -0.0050
Episode_Reward/feet_air_time_variance: -0.1401
Episode_Reward/feet_gait: 0.4330
Episode_Reward/feet_contact_without_cmd: 0.0005
Episode_Reward/feet_slide: -0.1981
Episode_Reward/feet_height_body: -0.1200
Episode_Reward/upward: 3.8633
Curriculum/terrain_levels: 3.2721
Metrics/base_velocity/error_vel_xy: 0.8048
Metrics/base_velocity/error_vel_yaw: 0.7937
Episode_Termination/time_out: 1.0000
Episode_Termination/terrain_out_of_bounds: 0.0000

Hello, I would like to ask a question. Why, even though I have trained it nearly 20,000 times, the parameters error_vel_xy and error_vel_yaw are still quite large, while theoretically they should be very small. I really can't figure out the reason. I hope you can help me. Thank you.

[RandomOakForest]

Thank you for posting this. We need more details on what you are training. I'll move this post to our Discussions for follow up.