[Question] Unable to reset the position of an object fixed in the world

[seanxd0727]

Question

Hello, I want to randomize the state of an object during training and add a platform fixed in the world coordinate system below the object. But write_root_link_pose_to_sim is not work. Platform always initialized in the default position.

Here is my randomization code:

def reset_object_state_with_platform_uniform(
    env: ManagerLocoManipRLEnv,
    env_ids: torch.Tensor,
    pose_range: dict[str, tuple[float, float]],
    velocity_range: dict[str, tuple[float, float]],
    object_cfg: SceneEntityCfg = SceneEntityCfg("object"),
    platform_cfg: SceneEntityCfg = SceneEntityCfg("platform"),
):
    # extract the used quantities (to enable type-hinting)
    object: RigidObject | Articulation = env.scene[object_cfg.name]
    platform: RigidObject = env.scene[platform_cfg.name]
    # get default root state
    root_states = object.data.default_root_state[env_ids].clone()
    # poses
    range_list = [pose_range.get(key, (0.0, 0.0)) for key in ["x", "y", "z", "roll", "pitch", "yaw"]]
    ranges = torch.tensor(range_list, device=object.device)
    rand_samples = math_utils.sample_uniform(ranges[:, 0], ranges[:, 1], (len(env_ids), 6), device=object.device)

    if env.object_aabb_local_cache is None:
        prim_paths = sim_utils.find_matching_prim_paths(object.cfg.prim_path)
        cache = bounds_utils.create_bbox_cache()
        aabb_list = [bounds_utils.compute_aabb(cache, prim_path=path) for path in prim_paths]
        aabb_np = np.stack(aabb_list, axis=0)  # (N,6)
        size_np = aabb_np[:, 3:] - aabb_np[:, :3]  # (N,3)
        env.object_aabb_local_cache = torch.from_numpy(size_np).to(env.device).to(torch.float32)  # (N,3)

    positions = root_states[:, 0:3] + env.scene.env_origins[env_ids] + rand_samples[:, 0:3]
    platform_pos_z = positions[:, 2] - 0.5 * env.object_aabb_local_cache[env_ids, 2] - 0.01
    platform_pos = torch.cat([positions[:, :2], platform_pos_z.unsqueeze(-1)], dim=-1)
    orientations_delta = math_utils.quat_from_euler_xyz(rand_samples[:, 3], rand_samples[:, 4], rand_samples[:, 5])
    orientations = math_utils.quat_mul(root_states[:, 3:7], orientations_delta)
    # velocities
    range_list = [velocity_range.get(key, (0.0, 0.0)) for key in ["x", "y", "z", "roll", "pitch", "yaw"]]
    ranges = torch.tensor(range_list, device=object.device)
    rand_samples = math_utils.sample_uniform(ranges[:, 0], ranges[:, 1], (len(env_ids), 6), device=object.device)

    velocities = root_states[:, 7:13] + rand_samples

    # set into the physics simulation
    object.write_root_pose_to_sim(torch.cat([positions, orientations], dim=-1), env_ids=env_ids)
    object.write_root_velocity_to_sim(velocities, env_ids=env_ids)
    platform.write_root_link_pose_to_sim(torch.cat([platform_pos, orientations], dim=-1), env_ids=env_ids)

PLATFORM_CFG = RigidObjectCfg(
    spawn=sim_utils.UrdfFileCfg(
        asset_path=f"{XD_LAB_DESCRIPTION_DIR}/objects/platform/platform0.urdf",
        usd_dir=f"{XD_LAB_DESCRIPTION_DIR}/objects/platform/usd",
        usd_file_name="platform0.usd",
        fix_base=True,
        activate_contact_sensors=False,
        articulation_props=sim_utils.ArticulationRootPropertiesCfg(articulation_enabled=False),
        rigid_props=sim_utils.RigidBodyPropertiesCfg(),
        joint_drive=sim_utils.UrdfConverterCfg.JointDriveCfg(
            gains=sim_utils.UrdfConverterCfg.JointDriveCfg.PDGainsCfg(
                stiffness=0,
                damping=0,
            ),
        ),
    ),
    init_state=RigidObjectCfg.InitialStateCfg(),
)

Build Info

• Isaac Lab Version: [e.g. 2.3.0]
• Isaac Sim Version: [e.g. 5.1]