Upload 8 files

skill_policies/h1_reaching/0000_best/h1_reaching_config.py

@@ -0,0 +1,312 @@
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# 1. Redistributions of source code must retain the above copyright notice, this
+# list of conditions and the following disclaimer.
+#
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+#
+# 3. Neither the name of the copyright holder nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Copyright (c) 2024 Beijing RobotEra TECHNOLOGY CO.,LTD. All rights reserved.
+
+
+from legged_gym.envs.base.legged_robot_config import LeggedRobotCfg, LeggedRobotCfgPPO
+
+
+class H1ReachingCfg(LeggedRobotCfg):
+    """
+    Configuration class for the H1 humanoid robot.
+    """
+    class human:
+        delay = 0.0 # delay in seconds
+        freq = 10
+        resample_on_env_reset = True
+        filename = 'CMU.npy'
+        rng = None  # None for no range, default None
+        
+    class env(LeggedRobotCfg.env):
+        # change the observation dim
+        num_actions = 19
+        frame_stack = 1
+        c_frame_stack = 3
+        command_dim = 14
+        num_single_obs = 3 * num_actions + 6 + command_dim # see `obs_buf = torch.cat(...)` for details
+        num_observations = int(frame_stack * num_single_obs)
+        single_num_privileged_obs = 3 * num_actions + 60
+        num_privileged_obs = int(c_frame_stack * single_num_privileged_obs)
+        
+        num_envs = 4096
+        episode_length_s = 24  # episode length in seconds
+        use_ref_actions = False
+
+    class safety:
+        # safety factors
+        pos_limit = 1.0
+        vel_limit = 1.0
+        torque_limit = 0.85
+
+    class asset(LeggedRobotCfg.asset):
+        file = '{LEGGED_GYM_ROOT_DIR}/resources/robots/h1/urdf/h1_wrist.urdf'
+
+        name = "h1"
+        foot_name = "ankle"
+        knee_name = "knee"
+        elbow_name = "elbow"
+        torso_name = "torso"
+        wrist_name = "wrist"
+
+        terminate_after_contacts_on = ['pelvis', 'torso', 'shoulder', 'elbow']
+        penalize_contacts_on = ["hip", 'knee', 'pelvis', 'torso', 'shoulder', 'elbow']
+        self_collisions = 0  # 1 to disable, 0 to enable...bitwise filter
+        flip_visual_attachments = False
+        replace_cylinder_with_capsule = False # replace collision cylinders with capsules, leads to faster/more stable simulation
+        fix_base_link = False
+        collapse_fixed_joints = False
+
+    class terrain(LeggedRobotCfg.terrain):
+        mesh_type = 'plane'
+        curriculum = False
+        # mesh_type = 'trimesh'
+        # curriculum = True
+        
+        # rough terrain only:
+        measure_heights = False
+        static_friction = 0.6
+        dynamic_friction = 0.6
+        terrain_length = 8.
+        terrain_width = 8.
+        num_rows = 20  # number of terrain rows (levels)
+        num_cols = 20  # number of terrain cols (types)
+        max_init_terrain_level = 10  # starting curriculum state
+        # plane; obstacles; uniform; slope_up; slope_down, stair_up, stair_down
+        terrain_proportions = [0.2, 0.2, 0.4, 0.1, 0.1, 0, 0]
+        restitution = 0.
+
+    class noise:
+        add_noise = True
+        noise_level = 0.6    # scales other values
+
+        class noise_scales:
+            dof_pos = 0.05
+            dof_vel = 0.5
+            ang_vel = 0.1
+            lin_vel = 0.05
+            quat = 0.03
+            height_measurements = 0.1
+
+    class init_state(LeggedRobotCfg.init_state):
+        pos = [0.0, 0.0, 1.0] # x,y,z [m]
+
+        default_joint_angles = { # = target angles [rad] when action = 0.0
+           'left_hip_yaw_joint' : 0. ,   
+           'left_hip_roll_joint' : 0,               
+           'left_hip_pitch_joint' : -0.4,         
+           'left_knee_joint' : 0.8,       
+           'left_ankle_joint' : -0.4,     
+           'right_hip_yaw_joint' : 0., 
+           'right_hip_roll_joint' : 0, 
+           'right_hip_pitch_joint' : -0.4,                                       
+           'right_knee_joint' : 0.8,                                             
+           'right_ankle_joint' : -0.4,                                     
+           'torso_joint' : 0., 
+           'left_shoulder_pitch_joint' : 0., 
+           'left_shoulder_roll_joint' : 0, 
+           'left_shoulder_yaw_joint' : 0.,
+           'left_elbow_joint'  : 0.,
+           'right_shoulder_pitch_joint' : 0.,
+           'right_shoulder_roll_joint' : 0.0,
+           'right_shoulder_yaw_joint' : 0.,
+           'right_elbow_joint' : 0.,
+        }
+
+    class control(LeggedRobotCfg.control):
+        control_type = 'P'
+        # PD Drive parameters:
+        stiffness = {'hip_yaw': 200,
+                     'hip_roll': 200,
+                     'hip_pitch': 200,
+                     'knee': 300,
+                     'ankle': 40,
+                     'torso': 300,
+                     'shoulder': 100,
+                     "elbow":100,
+                     }  # [N*m/rad]
+        damping = {  'hip_yaw': 5,
+                     'hip_roll': 5,
+                     'hip_pitch': 5,
+                     'knee': 6,
+                     'ankle': 2,
+                     'torso': 6,
+                     'shoulder': 2,
+                     "elbow":2,
+                     }  # [N*m/rad]  # [N*m*s/rad]
+        # action scale: target angle = actionScale * action + defaultAngle
+        action_scale = 0.25
+        # decimation: Number of control action updates @ sim DT per policy DT
+        decimation = 10  # 100hz
+
+    class sim(LeggedRobotCfg.sim):
+        dt = 0.001  # 1000 Hz
+        substeps = 1  # 2
+        up_axis = 1  # 0 is y, 1 is z
+
+        class physx(LeggedRobotCfg.sim.physx):
+            num_threads = 10
+            solver_type = 1  # 0: pgs, 1: tgs
+            num_position_iterations = 4
+            num_velocity_iterations = 0
+            contact_offset = 0.01  # [m]
+            rest_offset = 0.0   # [m]
+            bounce_threshold_velocity = 0.1  # [m/s]
+            max_depenetration_velocity = 1.0
+            max_gpu_contact_pairs = 2**23  # 2**24 -> needed for 8000 envs and more
+            default_buffer_size_multiplier = 5
+            # 0: never, 1: last sub-step, 2: all sub-steps (default=2)
+            contact_collection = 2
+
+    class domain_rand:
+        randomize_friction = True
+        friction_range = [0.1, 2.0]
+        randomize_base_mass = True
+        added_mass_range = [-5., 5.]
+        push_robots = True
+        push_interval_s = 4
+        max_push_vel_xy = 0.2
+        max_push_ang_vel = 0.4
+        dynamic_randomization = 0.02
+
+    class commands(LeggedRobotCfg.commands):
+        # Vers: lin_vel_x, lin_vel_y, ang_vel_yaw, heading (in heading mode ang_vel_yaw is recomputed from heading error)
+        num_commands = 4
+        resampling_time = 8.  # time before command are changed[s]
+        heading_command = True  # if true: compute ang vel command from heading error
+        curriculum = False # if true: curriculum update of commands
+
+        class ranges:
+            lin_vel_x = [-0, 0]  # min max [m/s]
+            lin_vel_y = [-0, 0]   # min max [m/s]
+            ang_vel_yaw = [-0, 0]    # min max [rad/s]
+            heading = [-0, 0]
+            # wrist pos command ranges
+            wrist_max_radius = 0.25
+            l_wrist_pos_x = [-0.10, 0.25]
+            l_wrist_pos_y = [-0.10, 0.25]
+            l_wrist_pos_z = [-0.25, 0.25]
+            r_wrist_pos_x = [-0.10, 0.25]
+            r_wrist_pos_y = [-0.25, 0.10]
+            r_wrist_pos_z = [-0.25, 0.25]
+
+    class rewards:
+        base_height_target = 0.89
+        min_dist = 0.2
+        max_dist = 0.5
+        # put some settings here for LLM parameter tuning
+        target_joint_pos_scale = 0.17    # rad
+        target_feet_height = 0.06       # m
+        cycle_time = 0.64                # sec
+        # if true negative total rewards are clipped at zero (avoids early termination problems)
+        only_positive_rewards = True
+        # tracking reward = exp(error*sigma)
+        tracking_sigma = 5
+        max_contact_force = 700  # forces above this value are penalized
+
+        class scales:
+            # TODO: 1. stand_still 2. joint_pos*2 3. add command input
+            # reference motion tracking
+            # joint_pos = 5
+            wrist_pos = 5
+            # feet_clearance = 0
+            # feet_contact_number = 0
+            # # gait
+            # feet_air_time = 0
+            # foot_slip = -0.05
+            feet_distance = 0.5
+            # knee_distance = 0.2
+            # # elbow_distance = 0.4
+            # # elbow_torso_distance = 0.4
+            # # contact
+            # feet_contact_forces = -0.01
+            # # vel tracking
+            # tracking_lin_vel = 0.
+            # tracking_ang_vel = 0.
+            # vel_mismatch_exp = 0.5  # lin_z; ang x,y
+            # low_speed = 0.2
+            # track_vel_hard = 0.5 * 2
+            # # base pos
+            default_joint_pos = 0.5
+            upper_body_pos = 0.5
+            orientation = 1.
+            # base_height = 0.2
+            # base_acc = 0.2
+            # energy
+            # action_smoothness = -0.002
+            torques = -1e-5
+            dof_vel = -5e-4
+            dof_acc = -1e-7
+            # collision = -0.2
+            #### humanplus ####
+            # lin_vel_z = -0.1
+            # ang_vel_xy = -0.1
+
+    class normalization:
+        class obs_scales:
+            lin_vel = 2.
+            ang_vel = 1.
+            dof_pos = 1.
+            dof_vel = 0.05
+            quat = 1.
+            height_measurements = 5.0
+        clip_observations = 18.
+        clip_actions = 18.
+
+
+class H1ReachingCfgPPO(LeggedRobotCfgPPO):
+    seed = 5
+    runner_class_name = 'OnPolicyRunner'   # DWLOnPolicyRunner
+
+    class policy:
+        init_noise_std = 1.0
+        actor_hidden_dims = [512, 256, 128]
+        critic_hidden_dims = [768, 256, 128]
+
+    class algorithm(LeggedRobotCfgPPO.algorithm):
+        entropy_coef = 0.001
+        learning_rate = 1e-5
+        num_learning_epochs = 2
+        gamma = 0.994
+        lam = 0.9
+        num_mini_batches = 4
+
+    class runner:
+        policy_class_name = 'ActorCritic'
+        algorithm_class_name = 'PPO'
+        num_steps_per_env = 60  # per iteration
+        max_iterations = 15001 # 3001  # number of policy updates
+
+        # logging
+        save_interval = 100  # check for potential saves every this many iterations
+        experiment_name = 'h1_reaching'
+        run_name = ''
+        # load and resume
+        resume = False
+        load_run = -1  # -1 = last run
+        checkpoint = -1  # -1 = last saved model
+        resume_path = None  # updated from load_run and ckpt

skill_policies/h1_reaching/0000_best/model_15001.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0097e7783bd73669aaa37f49cf2fcf2c3caf5328a473f12e264ae34b2e96e47f
+size 8502257

skill_policies/h1_squatting/0000_best/h1_squatting_config.py

@@ -0,0 +1,309 @@
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# 1. Redistributions of source code must retain the above copyright notice, this
+# list of conditions and the following disclaimer.
+#
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+#
+# 3. Neither the name of the copyright holder nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Copyright (c) 2024 Beijing RobotEra TECHNOLOGY CO.,LTD. All rights reserved.
+
+
+from legged_gym.envs.base.legged_robot_config import LeggedRobotCfg, LeggedRobotCfgPPO
+
+
+class H1SquattingCfg(LeggedRobotCfg):
+    """
+    Configuration class for the H1 humanoid robot.
+    """
+    class human:
+        delay = 0.0 # delay in seconds
+        freq = 10
+        resample_on_env_reset = True
+        filename = 'CMU.npy'
+        rng = None  # None for no range, default None
+        
+    class env(LeggedRobotCfg.env):
+        # change the observation dim
+        num_actions = 19
+        frame_stack = 1
+        c_frame_stack = 3
+        command_dim = 1
+        num_single_obs = 3 * num_actions + 6 + command_dim # see `obs_buf = torch.cat(...)` for details
+        num_observations = int(frame_stack * num_single_obs)
+        single_num_privileged_obs = 3 * num_actions + 18 + 3
+        num_privileged_obs = int(c_frame_stack * single_num_privileged_obs)
+        
+        num_envs = 4096
+        episode_length_s = 24  # episode length in seconds
+        use_ref_actions = False
+
+    class safety:
+        # safety factors
+        pos_limit = 1.0
+        vel_limit = 1.0
+        torque_limit = 0.85
+
+    class asset(LeggedRobotCfg.asset):
+        file = '{LEGGED_GYM_ROOT_DIR}/resources/robots/h1/urdf/h1_wrist.urdf'
+
+        name = "h1"
+        foot_name = "ankle"
+        knee_name = "knee"
+        elbow_name = "elbow"
+        torso_name = "torso"
+        wrist_name = "wrist"
+
+        terminate_after_contacts_on = ['pelvis', 'torso', 'shoulder', 'elbow',
+                                       'hip', 'knee']
+        penalize_contacts_on = ["hip", 'knee', 'pelvis', 'torso', 'shoulder', 'elbow']
+        self_collisions = 0  # 1 to disable, 0 to enable...bitwise filter
+        flip_visual_attachments = False
+        replace_cylinder_with_capsule = False # replace collision cylinders with capsules, leads to faster/more stable simulation
+        fix_base_link = False
+        collapse_fixed_joints = False
+
+    class terrain(LeggedRobotCfg.terrain):
+        mesh_type = 'plane'
+        curriculum = False
+        # mesh_type = 'trimesh'
+        # curriculum = True
+        
+        # rough terrain only:
+        measure_heights = False
+        static_friction = 0.6
+        dynamic_friction = 0.6
+        terrain_length = 8.
+        terrain_width = 8.
+        num_rows = 20  # number of terrain rows (levels)
+        num_cols = 20  # number of terrain cols (types)
+        max_init_terrain_level = 10  # starting curriculum state
+        # plane; obstacles; uniform; slope_up; slope_down, stair_up, stair_down
+        terrain_proportions = [0.2, 0.2, 0.4, 0.1, 0.1, 0, 0]
+        restitution = 0.
+
+    class noise:
+        add_noise = True
+        noise_level = 0.6    # scales other values
+
+        class noise_scales:
+            dof_pos = 0.05
+            dof_vel = 0.5
+            ang_vel = 0.1
+            lin_vel = 0.05
+            quat = 0.03
+            height_measurements = 0.1
+
+    class init_state(LeggedRobotCfg.init_state):
+        pos = [0.0, 0.0, 1.0] # x,y,z [m]
+
+        default_joint_angles = { # = target angles [rad] when action = 0.0
+           'left_hip_yaw_joint' : 0. ,   
+           'left_hip_roll_joint' : 0,               
+           'left_hip_pitch_joint' : -0.4,         
+           'left_knee_joint' : 0.8,       
+           'left_ankle_joint' : -0.4,     
+           'right_hip_yaw_joint' : 0., 
+           'right_hip_roll_joint' : 0, 
+           'right_hip_pitch_joint' : -0.4,                                       
+           'right_knee_joint' : 0.8,                                             
+           'right_ankle_joint' : -0.4,                                     
+           'torso_joint' : 0., 
+           'left_shoulder_pitch_joint' : 0., 
+           'left_shoulder_roll_joint' : 0, 
+           'left_shoulder_yaw_joint' : 0.,
+           'left_elbow_joint'  : 0.,
+           'right_shoulder_pitch_joint' : 0.,
+           'right_shoulder_roll_joint' : 0.0,
+           'right_shoulder_yaw_joint' : 0.,
+           'right_elbow_joint' : 0.,
+        }
+
+    class control(LeggedRobotCfg.control):
+        control_type = 'P'
+        # PD Drive parameters:
+        stiffness = {'hip_yaw': 200,
+                     'hip_roll': 200,
+                     'hip_pitch': 200,
+                     'knee': 300,
+                     'ankle': 40,
+                     'torso': 300,
+                     'shoulder': 100,
+                     "elbow":100,
+                     }  # [N*m/rad]
+        damping = {  'hip_yaw': 5,
+                     'hip_roll': 5,
+                     'hip_pitch': 5,
+                     'knee': 6,
+                     'ankle': 2,
+                     'torso': 6,
+                     'shoulder': 2,
+                     "elbow":2,
+                     }  # [N*m/rad]  # [N*m*s/rad]
+        # action scale: target angle = actionScale * action + defaultAngle
+        action_scale = 0.25
+        # decimation: Number of control action updates @ sim DT per policy DT
+        decimation = 10  # 100hz
+
+    class sim(LeggedRobotCfg.sim):
+        dt = 0.001  # 1000 Hz
+        substeps = 1  # 2
+        up_axis = 1  # 0 is y, 1 is z
+
+        class physx(LeggedRobotCfg.sim.physx):
+            num_threads = 10
+            solver_type = 1  # 0: pgs, 1: tgs
+            num_position_iterations = 4
+            num_velocity_iterations = 0
+            contact_offset = 0.01  # [m]
+            rest_offset = 0.0   # [m]
+            bounce_threshold_velocity = 0.1  # [m/s]
+            max_depenetration_velocity = 1.0
+            max_gpu_contact_pairs = 2**23  # 2**24 -> needed for 8000 envs and more
+            default_buffer_size_multiplier = 5
+            # 0: never, 1: last sub-step, 2: all sub-steps (default=2)
+            contact_collection = 2
+
+    class domain_rand:
+        randomize_friction = True
+        friction_range = [0.1, 2.0]
+        randomize_base_mass = True
+        added_mass_range = [-5., 5.]
+        push_robots = True
+        push_interval_s = 4
+        max_push_vel_xy = 0.2
+        max_push_ang_vel = 0.4
+        dynamic_randomization = 0.02
+
+    class commands(LeggedRobotCfg.commands):
+        # Vers: lin_vel_x, lin_vel_y, ang_vel_yaw, heading (in heading mode ang_vel_yaw is recomputed from heading error)
+        num_commands = 4
+        resampling_time = 8.  # time before command are changed[s]
+        heading_command = True  # if true: compute ang vel command from heading error
+        curriculum = False # if true: curriculum update of commands
+
+        class ranges:
+            lin_vel_x = [-0, 0]  # min max [m/s]
+            lin_vel_y = [-0, 0]   # min max [m/s]
+            ang_vel_yaw = [-0, 0]    # min max [rad/s]
+            heading = [-0, 0]
+            # root height [m]
+            root_height_std = 0.1
+            min_root_height = 0.5
+            max_root_height = 1.1
+
+    class rewards:
+        base_height_target = 0.89
+        min_dist = 0.2
+        max_dist = 0.5
+        # put some settings here for LLM parameter tuning
+        target_joint_pos_scale = 0.17    # rad
+        target_feet_height = 0.06       # m
+        cycle_time = 0.64                # sec
+        # if true negative total rewards are clipped at zero (avoids early termination problems)
+        only_positive_rewards = True
+        # tracking reward = exp(error*sigma)
+        tracking_sigma = 5
+        max_contact_force = 700  # forces above this value are penalized
+
+        class scales:
+            # TODO: 1. stand_still 2. joint_pos*2 3. add command input
+            # reference motion tracking
+            # joint_pos = 5
+            squatting = 5
+            # feet_clearance = 0
+            # feet_contact_number = 0
+            # # gait
+            # feet_air_time = 0
+            # foot_slip = -0.05
+            feet_distance = 0.5
+            # knee_distance = 0.2
+            # # elbow_distance = 0.4
+            # # elbow_torso_distance = 0.4
+            # # contact
+            # feet_contact_forces = -0.01
+            # # vel tracking
+            # tracking_lin_vel = 0.
+            # tracking_ang_vel = 0.
+            # vel_mismatch_exp = 0.5  # lin_z; ang x,y
+            # low_speed = 0.2
+            # track_vel_hard = 0.5 * 2
+            # # base pos
+            default_joint_pos = 0.5
+            upper_body_pos = 0.5
+            orientation = 1.
+            # base_height = 0.2
+            # base_acc = 0.2
+            # energy
+            # action_smoothness = -0.002
+            torques = -1e-5
+            dof_vel = -5e-4
+            dof_acc = -1e-7
+            # collision = -0.2
+            #### humanplus ####
+            # lin_vel_z = -0.1
+            # ang_vel_xy = -0.1
+
+    class normalization:
+        class obs_scales:
+            lin_vel = 2.
+            ang_vel = 1.
+            dof_pos = 1.
+            dof_vel = 0.05
+            quat = 1.
+            height_measurements = 5.0
+        clip_observations = 18.
+        clip_actions = 18.
+
+
+class H1SquattingCfgPPO(LeggedRobotCfgPPO):
+    seed = 5
+    runner_class_name = 'OnPolicyRunner'   # DWLOnPolicyRunner
+
+    class policy:
+        init_noise_std = 1.0
+        actor_hidden_dims = [512, 256, 128]
+        critic_hidden_dims = [768, 256, 128]
+
+    class algorithm(LeggedRobotCfgPPO.algorithm):
+        entropy_coef = 0.001
+        learning_rate = 1e-5
+        num_learning_epochs = 2
+        gamma = 0.994
+        lam = 0.9
+        num_mini_batches = 4
+
+    class runner:
+        policy_class_name = 'ActorCritic'
+        algorithm_class_name = 'PPO'
+        num_steps_per_env = 60  # per iteration
+        max_iterations = 15001 # 3001  # number of policy updates
+
+        # logging
+        save_interval = 100  # check for potential saves every this many iterations
+        experiment_name = 'h1_squatting'
+        run_name = ''
+        # load and resume
+        resume = False
+        load_run = -1  # -1 = last run
+        checkpoint = -1  # -1 = last saved model
+        resume_path = None  # updated from load_run and ckpt

skill_policies/h1_squatting/0000_best/model_15001.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab45fa5311536001fa5a077d1ddb441a4bab425bfa83271c91d888368d94e368
+size 7344113

skill_policies/h1_stepping/0000_best/h1_stepping_config.py

@@ -0,0 +1,307 @@
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# 1. Redistributions of source code must retain the above copyright notice, this
+# list of conditions and the following disclaimer.
+#
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+#
+# 3. Neither the name of the copyright holder nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Copyright (c) 2024 Beijing RobotEra TECHNOLOGY CO.,LTD. All rights reserved.
+
+
+from legged_gym.envs.base.legged_robot_config import LeggedRobotCfg, LeggedRobotCfgPPO
+
+
+class H1SteppingCfg(LeggedRobotCfg):
+    """
+    Configuration class for the H1 humanoid robot.
+    """
+    class human:
+        delay = 0.0 # delay in seconds
+        freq = 10
+        resample_on_env_reset = True
+        filename = 'CMU.npy'
+        rng = None  # None for no range, default None
+        
+    class env(LeggedRobotCfg.env):
+        # change the observation dim
+        num_actions = 19
+        frame_stack = 1
+        c_frame_stack = 3
+        command_dim = 4
+        num_single_obs = 3 * num_actions + 6 + command_dim # see `obs_buf = torch.cat(...)` for details
+        num_observations = int(frame_stack * num_single_obs)
+        single_num_privileged_obs = 3 * num_actions + 18 + 12
+        num_privileged_obs = int(c_frame_stack * single_num_privileged_obs)
+        
+        num_envs = 4096
+        episode_length_s = 24  # episode length in seconds
+        use_ref_actions = False
+
+    class safety:
+        # safety factors
+        pos_limit = 1.0
+        vel_limit = 1.0
+        torque_limit = 0.85
+
+    class asset(LeggedRobotCfg.asset):
+        file = '{LEGGED_GYM_ROOT_DIR}/resources/robots/h1/urdf/h1_wrist.urdf'
+
+        name = "h1"
+        foot_name = "ankle"
+        knee_name = "knee"
+        elbow_name = "elbow"
+        torso_name = "torso"
+        wrist_name = "wrist"
+
+        terminate_after_contacts_on = ['pelvis', 'torso', 'shoulder', 'elbow',
+                                       'hip', 'knee']
+        penalize_contacts_on = ["hip", 'knee', 'pelvis', 'torso', 'shoulder', 'elbow']
+        self_collisions = 0  # 1 to disable, 0 to enable...bitwise filter
+        flip_visual_attachments = False
+        replace_cylinder_with_capsule = False # replace collision cylinders with capsules, leads to faster/more stable simulation
+        fix_base_link = False
+        collapse_fixed_joints = False
+
+    class terrain(LeggedRobotCfg.terrain):
+        mesh_type = 'plane'
+        curriculum = False
+        # mesh_type = 'trimesh'
+        # curriculum = True
+        
+        # rough terrain only:
+        measure_heights = False
+        static_friction = 0.6
+        dynamic_friction = 0.6
+        terrain_length = 8.
+        terrain_width = 8.
+        num_rows = 20  # number of terrain rows (levels)
+        num_cols = 20  # number of terrain cols (types)
+        max_init_terrain_level = 10  # starting curriculum state
+        # plane; obstacles; uniform; slope_up; slope_down, stair_up, stair_down
+        terrain_proportions = [0.2, 0.2, 0.4, 0.1, 0.1, 0, 0]
+        restitution = 0.
+
+    class noise:
+        add_noise = True
+        noise_level = 0.6    # scales other values
+
+        class noise_scales:
+            dof_pos = 0.05
+            dof_vel = 0.5
+            ang_vel = 0.1
+            lin_vel = 0.05
+            quat = 0.03
+            height_measurements = 0.1
+
+    class init_state(LeggedRobotCfg.init_state):
+        pos = [0.0, 0.0, 1.0] # x,y,z [m]
+
+        default_joint_angles = { # = target angles [rad] when action = 0.0
+           'left_hip_yaw_joint' : 0. ,   
+           'left_hip_roll_joint' : 0,               
+           'left_hip_pitch_joint' : -0.4,         
+           'left_knee_joint' : 0.8,       
+           'left_ankle_joint' : -0.4,     
+           'right_hip_yaw_joint' : 0., 
+           'right_hip_roll_joint' : 0, 
+           'right_hip_pitch_joint' : -0.4,                                       
+           'right_knee_joint' : 0.8,                                             
+           'right_ankle_joint' : -0.4,                                     
+           'torso_joint' : 0., 
+           'left_shoulder_pitch_joint' : 0., 
+           'left_shoulder_roll_joint' : 0, 
+           'left_shoulder_yaw_joint' : 0.,
+           'left_elbow_joint'  : 0.,
+           'right_shoulder_pitch_joint' : 0.,
+           'right_shoulder_roll_joint' : 0.0,
+           'right_shoulder_yaw_joint' : 0.,
+           'right_elbow_joint' : 0.,
+        }
+
+    class control(LeggedRobotCfg.control):
+        control_type = 'P'
+        # PD Drive parameters:
+        stiffness = {'hip_yaw': 200,
+                     'hip_roll': 200,
+                     'hip_pitch': 200,
+                     'knee': 300,
+                     'ankle': 40,
+                     'torso': 300,
+                     'shoulder': 100,
+                     "elbow":100,
+                     }  # [N*m/rad]
+        damping = {  'hip_yaw': 5,
+                     'hip_roll': 5,
+                     'hip_pitch': 5,
+                     'knee': 6,
+                     'ankle': 2,
+                     'torso': 6,
+                     'shoulder': 2,
+                     "elbow":2,
+                     }  # [N*m/rad]  # [N*m*s/rad]
+        # action scale: target angle = actionScale * action + defaultAngle
+        action_scale = 0.25
+        # decimation: Number of control action updates @ sim DT per policy DT
+        decimation = 10  # 100hz
+
+    class sim(LeggedRobotCfg.sim):
+        dt = 0.001  # 1000 Hz
+        substeps = 1  # 2
+        up_axis = 1  # 0 is y, 1 is z
+
+        class physx(LeggedRobotCfg.sim.physx):
+            num_threads = 10
+            solver_type = 1  # 0: pgs, 1: tgs
+            num_position_iterations = 4
+            num_velocity_iterations = 0
+            contact_offset = 0.01  # [m]
+            rest_offset = 0.0   # [m]
+            bounce_threshold_velocity = 0.1  # [m/s]
+            max_depenetration_velocity = 1.0
+            max_gpu_contact_pairs = 2**23  # 2**24 -> needed for 8000 envs and more
+            default_buffer_size_multiplier = 5
+            # 0: never, 1: last sub-step, 2: all sub-steps (default=2)
+            contact_collection = 2
+
+    class domain_rand:
+        randomize_friction = True
+        friction_range = [0.1, 2.0]
+        randomize_base_mass = True
+        added_mass_range = [-5., 5.]
+        push_robots = True
+        push_interval_s = 4
+        max_push_vel_xy = 0.2
+        max_push_ang_vel = 0.4
+        dynamic_randomization = 0.02
+
+    class commands(LeggedRobotCfg.commands):
+        # Vers: lin_vel_x, lin_vel_y, ang_vel_yaw, heading (in heading mode ang_vel_yaw is recomputed from heading error)
+        num_commands = 4
+        resampling_time = 8.  # time before command are changed[s]
+        heading_command = True  # if true: compute ang vel command from heading error
+        curriculum = False # if true: curriculum update of commands
+
+        class ranges:
+            lin_vel_x = [-0, 0]  # min max [m/s]
+            lin_vel_y = [-0, 0]   # min max [m/s]
+            ang_vel_yaw = [-0, 0]    # min max [rad/s]
+            heading = [-0, 0]
+            # feet pos command ranges
+            feet_max_radius = 0.25
+
+    class rewards:
+        base_height_target = 0.89
+        min_dist = 0.2
+        max_dist = 0.5
+        # put some settings here for LLM parameter tuning
+        target_joint_pos_scale = 0.17    # rad
+        target_feet_height = 0.06       # m
+        cycle_time = 0.64                # sec
+        # if true negative total rewards are clipped at zero (avoids early termination problems)
+        only_positive_rewards = True
+        # tracking reward = exp(error*sigma)
+        tracking_sigma = 5
+        max_contact_force = 700  # forces above this value are penalized
+
+        class scales:
+            # TODO: 1. stand_still 2. joint_pos*2 3. add command input
+            # reference motion tracking
+            # joint_pos = 5
+            feet_pos = 5
+            # feet_clearance = 0
+            # feet_contact_number = 0
+            # # gait
+            # feet_air_time = 0
+            # foot_slip = -0.05
+            # feet_distance = 0.5
+            # knee_distance = 0.2
+            # # elbow_distance = 0.4
+            # # elbow_torso_distance = 0.4
+            # # contact
+            # feet_contact_forces = -0.01
+            # # vel tracking
+            # tracking_lin_vel = 0.
+            # tracking_ang_vel = 0.
+            # vel_mismatch_exp = 0.5  # lin_z; ang x,y
+            # low_speed = 0.2
+            # track_vel_hard = 0.5 * 2
+            # # base pos
+            # default_joint_pos = 0.5
+            upper_body_pos = 0.5
+            orientation = 1.
+            # base_height = 0.2
+            # base_acc = 0.2
+            # energy
+            # action_smoothness = -0.002
+            torques = -1e-5
+            dof_vel = -5e-4
+            dof_acc = -1e-7
+            # collision = -0.2
+            #### humanplus ####
+            # lin_vel_z = -0.1
+            # ang_vel_xy = -0.1
+
+    class normalization:
+        class obs_scales:
+            lin_vel = 2.
+            ang_vel = 1.
+            dof_pos = 1.
+            dof_vel = 0.05
+            quat = 1.
+            height_measurements = 5.0
+        clip_observations = 18.
+        clip_actions = 18.
+
+
+class H1SteppingCfgPPO(LeggedRobotCfgPPO):
+    seed = 5
+    runner_class_name = 'OnPolicyRunner'   # DWLOnPolicyRunner
+
+    class policy:
+        init_noise_std = 1.0
+        actor_hidden_dims = [512, 256, 128]
+        critic_hidden_dims = [768, 256, 128]
+
+    class algorithm(LeggedRobotCfgPPO.algorithm):
+        entropy_coef = 0.001
+        learning_rate = 1e-5
+        num_learning_epochs = 2
+        gamma = 0.994
+        lam = 0.9
+        num_mini_batches = 4
+
+    class runner:
+        policy_class_name = 'ActorCritic'
+        algorithm_class_name = 'PPO'
+        num_steps_per_env = 60  # per iteration
+        max_iterations = 15001 # 3001  # number of policy updates
+
+        # logging
+        save_interval = 100  # check for potential saves every this many iterations
+        experiment_name = 'h1_stepping'
+        run_name = ''
+        # load and resume
+        resume = False
+        load_run = -1  # -1 = last run
+        checkpoint = -1  # -1 = last saved model
+        resume_path = None  # updated from load_run and ckpt

skill_policies/h1_stepping/0000_best/model_15001.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3fc8906af0901a5c0f6722f5fdbba12cdea69b3b9095af14f33502949378035
+size 7611377

skill_policies/h1_walking/0000_best/h1_walking_config.py

@@ -0,0 +1,292 @@
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# 1. Redistributions of source code must retain the above copyright notice, this
+# list of conditions and the following disclaimer.
+#
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+#
+# 3. Neither the name of the copyright holder nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Copyright (c) 2024 Beijing RobotEra TECHNOLOGY CO.,LTD. All rights reserved.
+
+
+from legged_gym.envs.base.legged_robot_config import LeggedRobotCfg, LeggedRobotCfgPPO
+
+
+class H1WalkingCfg(LeggedRobotCfg):
+    """
+    Configuration class for the H1 humanoid robot.
+    """
+    class env(LeggedRobotCfg.env):
+        # change the observation dim
+        num_actions = 19
+        frame_stack = 1
+        c_frame_stack = 3
+        command_dim = 3
+        num_single_obs = 3 * num_actions + 6 + command_dim # see `obs_buf = torch.cat(...)` for details
+        num_observations = int(frame_stack * num_single_obs)
+        single_num_privileged_obs = 4 * num_actions + 25
+        num_privileged_obs = int(c_frame_stack * single_num_privileged_obs)
+        # num_actions = 12
+        
+        num_envs = 4096
+        episode_length_s = 24  # episode length in seconds
+        use_ref_actions = False
+
+    class safety:
+        # safety factors
+        pos_limit = 1.0
+        vel_limit = 1.0
+        torque_limit = 0.85
+
+    class asset(LeggedRobotCfg.asset):
+        file = '{LEGGED_GYM_ROOT_DIR}/resources/robots/h1/urdf/h1_wrist.urdf'
+
+        name = "h1"
+        foot_name = "ankle"
+        knee_name = "knee"
+        elbow_name = "elbow"
+        torso_name = "torso"
+
+        terminate_after_contacts_on = ['pelvis', 'torso', 'shoulder', 'elbow']
+        penalize_contacts_on = ["hip", 'knee']
+        self_collisions = 0  # 1 to disable, 0 to enable...bitwise filter
+        flip_visual_attachments = False
+        replace_cylinder_with_capsule = False # replace collision cylinders with capsules, leads to faster/more stable simulation
+        fix_base_link = False
+
+    class terrain(LeggedRobotCfg.terrain):
+        # mesh_type = 'plane'
+        # curriculum = False
+        mesh_type = 'trimesh'
+        curriculum = True
+        
+        # rough terrain only:
+        measure_heights = False
+        static_friction = 0.6
+        dynamic_friction = 0.6
+        terrain_length = 8.
+        terrain_width = 8.
+        num_rows = 20  # number of terrain rows (levels)
+        num_cols = 20  # number of terrain cols (types)
+        max_init_terrain_level = 10  # starting curriculum state
+        # plane; obstacles; uniform; slope_up; slope_down, stair_up, stair_down
+        terrain_proportions = [0.2, 0.2, 0.4, 0.1, 0.1, 0, 0]
+        restitution = 0.
+
+    class noise:
+        add_noise = True
+        noise_level = 0.6    # scales other values
+
+        class noise_scales:
+            dof_pos = 0.05
+            dof_vel = 0.5
+            ang_vel = 0.1
+            lin_vel = 0.05
+            quat = 0.03
+            height_measurements = 0.1
+
+    class init_state(LeggedRobotCfg.init_state):
+        pos = [0.0, 0.0, 1.0] # x,y,z [m]
+
+        default_joint_angles = { # = target angles [rad] when action = 0.0
+           'left_hip_yaw_joint' : 0. ,   
+           'left_hip_roll_joint' : 0,               
+           'left_hip_pitch_joint' : -0.4,         
+           'left_knee_joint' : 0.8,       
+           'left_ankle_joint' : -0.4,     
+           'right_hip_yaw_joint' : 0., 
+           'right_hip_roll_joint' : 0, 
+           'right_hip_pitch_joint' : -0.4,                                       
+           'right_knee_joint' : 0.8,                                             
+           'right_ankle_joint' : -0.4,                                     
+           'torso_joint' : 0., 
+           'left_shoulder_pitch_joint' : 0., 
+           'left_shoulder_roll_joint' : 0, 
+           'left_shoulder_yaw_joint' : 0.,
+           'left_elbow_joint'  : 0.,
+           'right_shoulder_pitch_joint' : 0.,
+           'right_shoulder_roll_joint' : 0.0,
+           'right_shoulder_yaw_joint' : 0.,
+           'right_elbow_joint' : 0.,
+        }
+
+    class control(LeggedRobotCfg.control):
+        control_type = 'P'
+        # PD Drive parameters:
+        stiffness = {'hip_yaw': 200,
+                     'hip_roll': 200,
+                     'hip_pitch': 200,
+                     'knee': 300,
+                     'ankle': 40,
+                     'torso': 300,
+                     'shoulder': 100,
+                     "elbow":100,
+                     }  # [N*m/rad]
+        damping = {  'hip_yaw': 5,
+                     'hip_roll': 5,
+                     'hip_pitch': 5,
+                     'knee': 6,
+                     'ankle': 2,
+                     'torso': 6,
+                     'shoulder': 2,
+                     "elbow":2,
+                     }  # [N*m/rad]  # [N*m*s/rad]
+        # action scale: target angle = actionScale * action + defaultAngle
+        action_scale = 0.25
+        # decimation: Number of control action updates @ sim DT per policy DT
+        decimation = 10  # 100hz
+
+    class sim(LeggedRobotCfg.sim):
+        dt = 0.001  # 1000 Hz
+        substeps = 1  # 2
+        up_axis = 1  # 0 is y, 1 is z
+
+        class physx(LeggedRobotCfg.sim.physx):
+            num_threads = 10
+            solver_type = 1  # 0: pgs, 1: tgs
+            num_position_iterations = 4
+            num_velocity_iterations = 0
+            contact_offset = 0.01  # [m]
+            rest_offset = 0.0   # [m]
+            bounce_threshold_velocity = 0.1  # [m/s]
+            max_depenetration_velocity = 1.0
+            max_gpu_contact_pairs = 2**23  # 2**24 -> needed for 8000 envs and more
+            default_buffer_size_multiplier = 5
+            # 0: never, 1: last sub-step, 2: all sub-steps (default=2)
+            contact_collection = 2
+
+    class domain_rand:
+        randomize_friction = True
+        friction_range = [0.1, 2.0]
+        randomize_base_mass = True
+        added_mass_range = [-5., 5.]
+        push_robots = True
+        push_interval_s = 4
+        max_push_vel_xy = 0.2
+        max_push_ang_vel = 0.4
+        dynamic_randomization = 0.02
+
+    class commands(LeggedRobotCfg.commands):
+        # Vers: lin_vel_x, lin_vel_y, ang_vel_yaw, heading (in heading mode ang_vel_yaw is recomputed from heading error)
+        num_commands = 4
+        resampling_time = 8.  # time before command are changed[s]
+        heading_command = True  # if true: compute ang vel command from heading error
+        curriculum = True # if true: curriculum update of commands
+
+        class ranges:
+            lin_vel_x = [-1.0, 2.0]  # min max [m/s]
+            lin_vel_y = [-1.0, 1.0]   # min max [m/s]
+            ang_vel_yaw = [-1.0, 1.0]    # min max [rad/s]
+            heading = [-3.14, 3.14]
+
+    class rewards:
+        base_height_target = 0.89
+        min_dist = 0.2
+        max_dist = 0.5
+        # put some settings here for LLM parameter tuning
+        target_joint_pos_scale = 0.17    # rad
+        target_feet_height = 0.06       # m
+        cycle_time = 0.64                # sec
+        # if true negative total rewards are clipped at zero (avoids early termination problems)
+        only_positive_rewards = True
+        # tracking reward = exp(error*sigma)
+        tracking_sigma = 5
+        max_contact_force = 700  # forces above this value are penalized
+
+        class scales:
+            # reference motion tracking
+            joint_pos = 1.6
+            feet_clearance = 1. * 2
+            feet_contact_number = 1.2 * 2
+            # gait
+            feet_air_time = 1.
+            foot_slip = -0.05
+            feet_distance = 0.2
+            knee_distance = 0.2
+            # elbow_distance = 0.4
+            # elbow_torso_distance = 0.4
+            # contact
+            feet_contact_forces = -0.01
+            # vel tracking
+            tracking_lin_vel = 1.2 * 2
+            tracking_ang_vel = 1.1 * 2
+            vel_mismatch_exp = 0.5  # lin_z; ang x,y
+            low_speed = 0.2
+            track_vel_hard = 0.5 * 2
+            # stand_still = -0.5
+            # base pos
+            default_joint_pos = 0.5
+            upper_body_pos = 0.5
+            orientation = 1.
+            base_height = 0.2
+            base_acc = 0.2
+            # energy
+            action_smoothness = -0.002
+            torques = -1e-5
+            dof_vel = -5e-4
+            dof_acc = -1e-7
+            collision = -1.
+
+    class normalization:
+        class obs_scales:
+            lin_vel = 2.
+            ang_vel = 1.
+            dof_pos = 1.
+            dof_vel = 0.05
+            quat = 1.
+            height_measurements = 5.0
+        clip_observations = 18.
+        clip_actions = 18.
+
+
+class H1WalkingCfgPPO(LeggedRobotCfgPPO):
+    seed = 5
+    runner_class_name = 'OnPolicyRunner'   # DWLOnPolicyRunner
+
+    class policy:
+        init_noise_std = 1.0
+        actor_hidden_dims = [512, 256, 128]
+        critic_hidden_dims = [768, 256, 128]
+
+    class algorithm(LeggedRobotCfgPPO.algorithm):
+        entropy_coef = 0.001
+        learning_rate = 1e-5
+        num_learning_epochs = 2
+        gamma = 0.994
+        lam = 0.9
+        num_mini_batches = 4
+
+    class runner:
+        policy_class_name = 'ActorCritic'
+        algorithm_class_name = 'PPO'
+        num_steps_per_env = 60  # per iteration
+        max_iterations = 15001 # 3001  # number of policy updates
+
+        # logging
+        save_interval = 100  # check for potential saves every this many iterations
+        experiment_name = 'h1_walking'
+        run_name = ''
+        # load and resume
+        resume = False
+        load_run = -1  # -1 = last run
+        checkpoint = -1  # -1 = last saved model
+        resume_path = None  # updated from load_run and ckpt

skill_policies/h1_walking/0000_best/model_15001.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c530e2b51d7e7da5af74dc1c96ba8afd9c9649a5fe4c5a2622fe4a9b5dac2349
+size 7992305