minor

    def _render(self,info,init_obs=False,valid_actions=None):

        print("info",info)

        new_info=handel_info(info.copy())

        object_positions=new_info['obj_positions']

        other_information=new_info['other_info']

    This code demonstrates how to create an instance of the environment,

    reset it, and interact with it using manual input actions.

    """

    config = PrimitiveSkillEnvConfig(record_video=True, video_record_dir="./test_manipulation_video")

    # AlignTwoCube,PlaceTwoCube,PutAppleInDrawer,StackThreeCube

    config = PrimitiveSkillEnvConfig(record_video=True, video_record_dir="./test_manipulation_video",env_id="AlignTwoCube", render_mode="vision")

    env = PrimitiveSkillEnv(config)

    print(env.system_prompt())

        for name in self.object_list:

            info[f"is_{name}_grasped"] = self.agent.is_grasping(self.object_list[name])[0]

            info[f"{name}_pos"] = self.object_list[name].pose.p[0]

        info["cube_size"]=torch.ones_like(info["red_cube_pos"])*40

        info["cube_size"]=torch.ones_like(info["red_cube_pos"])*0.04

        info["gripper_pos"] = self.agent.tcp.pose.p[0]

        return info

            # convert to cm round to 2 decimal places

            obj_positions[k] = tuple(np.round(v*1000, 0).astype(int))

        elif k.endswith('_value'):

            obj_positions[k] = np.round(v*1000, 0).astype(int).item()

            other_info[k] = np.round(v*1000, 0).astype(int).item()

        elif k.endswith('_size'):

            other_info[k] = tuple(np.round(v*1000, 0).astype(int))

        else:

            if isinstance(v, np.ndarray):

            if isinstance(v, np.ndarray) and v.ndim == 0:

                other_info[k] = v.item()

            else:

                other_info[k] = v

    env_name: primitive_skill 

    env_config:

        render_mode: text

        env_id: "PlaceTwoCube" # AlignTwoCube,PlaceTwoCube,PutAppleInDrawer,StackThreeCube

        env_id: "StackThreeCube" # AlignTwoCube,PlaceTwoCube,PutAppleInDrawer,StackThreeCube

    train_size: 10000  

    test_size: 32

 No newline at end of file

    data.train_batch_size=8 \

    data.max_prompt_length=1024 \

    data.max_response_length=128 \

    data.max_trajectory_length=1500 \

    data.max_trajectory_length=1800 \

    data.image_key=images \

    data.truncation=error \

    actor_rollout_ref.model.path=Qwen/Qwen2.5-0.5B-Instruct \