SKRL PPO Agent, How to learn get up after a fall?

[AndreaRossetto]

Hi,
I'm wondering how I can make my biped model stand up from a fall, at the moment if The model fall down the eviroment resets. I'm able to train it to walk / stand but not to getting up after a fall.
This will be my next step, but I don't have so many ideas on reward and the termination of the simulation.

At the moment my reward are

# -- task
    track_lin_vel_xy_exp = RewTerm(
        func=mdp.track_lin_vel_xy_exp, weight=1.0, params={"command_name": "base_velocity", "std": math.sqrt(0.25)}
    )
    track_ang_vel_z_exp = RewTerm(
        func=mdp.track_ang_vel_z_exp, weight=0.5, params={"command_name": "base_velocity", "std": math.sqrt(0.25)}
    )

    # -- penalties
    lin_vel_z_l2 = RewTerm(func=mdp.lin_vel_z_l2, weight=-2.0)
    ang_vel_xy_l2 = RewTerm(func=mdp.ang_vel_xy_l2, weight=-0.05)
    joint_torques_l2 = RewTerm(func=mdp.joint_torques_l2, weight=-1.0e-5)
    action_rate_l2 = RewTerm(func=mdp.action_rate_l2, weight=-0.01)
    '''base_height_l2 = RewTerm(
        func=mdp.c_base_height_l2,
        params={
            "asset_cfg": SceneEntityCfg("robot", body_names="body_front"),
            "command_name": "base_orientation",
            "sensor_cfg": SceneEntityCfg("robot", body_names="height_scanner"),
        },
    )'''
    feet_air_time = RewTerm(
        func=mdp.feet_air_time,
        weight=2.0,
        params={
            "sensor_cfg": SceneEntityCfg("feet_contact_sensor", body_names=".*"),
            "command_name": "base_velocity",
            "threshold_min": 0.2,
            "threshold_max": 0.5,
        },
    )
    feet_slide = RewTerm(
        func=mdp.feet_slide,
        weight=-0.25,
        params={
            "sensor_cfg": SceneEntityCfg("feet_contact_sensor", body_names=".*"),
            "asset_cfg": SceneEntityCfg("robot", body_names="foot.*"),
        },
    )
    undesired_contacts = RewTerm(
        func=mdp.undesired_contacts,
        weight=-1.0,
        params={"sensor_cfg": SceneEntityCfg("contact_forces", body_names=["motor2motor_HR_.*", "motor2motor_HA_.*", \
                                                                           "spacer_leg_.*", "calf_.*"]),
                "threshold": 1.0},
    )
    joint_deviation_hip = RewTerm(
        func=mdp.joint_deviation_l1,
        weight=-0.1,
        params={"asset_cfg": SceneEntityCfg("robot", joint_names=["HR_HA_right_joint", "thigh_HA_right_joint", "HF_right_joint", "HR_HA_left_joint", "thigh_HA_left_joint", "HF_left_joint"])},
    )
    joint_deviation_knee = RewTerm(
        func=mdp.joint_deviation_l1,
        weight=-0.01,
        params={"asset_cfg": SceneEntityCfg("robot", joint_names=["ankle_right_joint", "ankle_left_joint"])},
    )
    # -- optional penalties
    flat_orientation_l2 = RewTerm(func=mdp.flat_orientation_l2, weight=-0.5)
    dof_pos_limits = RewTerm(func=mdp.joint_pos_limits, weight=-1.0)

The enviroment terminate with death when the model body touches the ground or when the timeout elapsed.

I'm asking some help on that part, My idea to terminate before the timeout can be something like "if body is touching the ground then I wait too see if after half of the timeout is still on the ground or not, if still on the ground end with death otherwise keep alive.
In a managedRL how can I do this kind of termination?

Second question, Is necessary to have some new rewards or the ones I have are enough ? what could be a reward to make it stand from a fall?

Thank you for your help