Service robots with hardware-level visual privacy protection capabilities and their control methods

By integrating a privacy-preserving image data processing unit into the service robot, the raw data is converted into encrypted feature vectors, solving the problem of privacy information leakage in existing technologies. This achieves hardware-level visual privacy protection and adaptive visual mode selection, improving the robot's privacy security in private spaces.

CN123103597APending Publication Date: 2026-11-17E-SURFING DIGITAL LIFE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611547146.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-30
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

When existing service robots enter private spaces, they cannot fundamentally prevent image recording and privacy information leakage. Existing technologies mostly rely on software algorithm post-processing, which cannot defend against hardware-level leakage risks. Furthermore, their mode switching is not intelligent enough and cannot adapt to scenarios with different privacy requirements.

Method used

The service robot integrates a dedicated privacy-preserving image data processing unit, which is physically connected to a privacy camera. It directly converts the raw data into encrypted irreversible feature vectors and dynamically selects the visual mode through the context awareness and decision-making module. It then fuses the encrypted feature vectors and radar point cloud data to perform the task.

Benefits of technology

Ensuring that the service robot cannot output visible images at the physical level achieves hardware-level privacy protection in private environments, and adaptively selects visual modes, thereby improving privacy security and reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103597A_ABST
    Figure CN123103597A_ABST
Patent Text Reader

Abstract

This application discloses a service robot with hardware-level visual privacy protection capabilities and its implementation method, including a context perception and decision-making module for obtaining visual mode switching instructions based on semantic tags and current task instructions; a data acquisition module including a privacy camera, a general-purpose camera, a LiDAR, and an inertial measurement unit, for acquiring privacy-preserving raw data, non-privacy images, radar point cloud data, and inertial data, respectively; a privacy-preserving image data processing unit, electrically connected to the privacy camera, for encrypting the privacy-preserving raw data to obtain encrypted feature vectors; and a main controller module for executing target tasks based on encrypted feature vectors, radar point cloud data, and inertial data in privacy-preserving visual mode, and based on non-privacy images and radar point cloud data in traditional visual mode. This application can improve the security and reliability of visual privacy in service robots and can be widely applied in the field of intelligent robot technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent robot technology, and in particular to a service robot with hardware-level visual privacy protection capabilities and its control method. Background Technology

[0002] Service robots are increasingly being used in hotels, homes, and healthcare settings to provide services such as item delivery, environmental inspection, and companionship. However, when they enter private spaces like bedrooms, hotel rooms, and hospital wards, their onboard cameras continuously collect environmental videos or images, raising serious concerns about privacy breaches. Current solutions often use software algorithms to recognize and post-process images (such as blurring or animating) to cover up private areas after the robot has collected them, but this cannot physically prevent private information from being recorded during the initial collection and processing stages. Summary of the Invention

[0003] The main objective of this application is to propose a service robot with hardware-level visual privacy protection capabilities and its implementation method, which can provide necessary service functions while ensuring at the physical level that no image recording is made in the private environment.

[0004] To achieve the above objectives, one aspect of this application proposes a service robot with hardware-level visual privacy protection capabilities, the service robot comprising: The context awareness and decision-making module is used to construct a semantic map, obtain the semantic label corresponding to the current coordinate point based on the semantic map, and obtain a visual mode switching instruction based on the semantic label and the current task instruction. The data acquisition module includes a privacy camera, a general-purpose camera, a lidar, and an inertial measurement unit, which are used to acquire privacy raw data, non-privacy images, radar point cloud data, and inertial data, respectively. A privacy-protected image data processing unit, electrically connected to the privacy camera, is used to encrypt the original privacy data to obtain an irreversible encrypted feature vector. The main controller module, connected to the context awareness and decision-making module, the data acquisition module, and the privacy-preserving image data processing unit, is used to select either a privacy-preserving visual mode or a traditional visual mode according to the visual mode switching instruction. In the privacy-preserving visual mode, it executes the target task based on the encrypted feature vector, the radar point cloud data, and the inertial data. In the traditional visual mode, it executes the target task based on the non-privacy image and the radar point cloud data.

[0005] In some embodiments, the service robot further includes: The human-computer interaction module is connected to the main controller module and is used to receive user task instructions and perform voice interaction or display the current visual mode according to the task instructions. The power management module is connected to the main controller module, the privacy camera, the general-purpose camera, the lidar, and the inertial measurement unit, and is used to supply power to the main controller module, the privacy camera, the general-purpose camera, the lidar, and the inertial measurement unit.

[0006] In some embodiments, the service robot further includes: A physical switch, connected between the privacy camera and the power management module, is used to turn the power supply to the privacy camera on or off.

[0007] In some embodiments, the service robot further includes: An analog switch is provided, and both the privacy camera and the general-purpose camera are connected to one end of the analog switch. The privacy-protected image data processing unit and the main controller module are also connected to the other end of the analog switch. This is used to connect the data path between the privacy camera and the privacy-protected image data processing unit in the privacy vision mode, and to connect the data path between the general-purpose camera and the main controller module in the conventional vision mode.

[0008] To achieve the above objectives, another aspect of this application proposes a control method for a service robot with hardware-level visual privacy protection capabilities, used to control the service robot with hardware-level visual privacy protection capabilities as described above. The method includes the following steps: Construct a semantic map, obtain the semantic label corresponding to the current coordinate point based on the semantic map, and select a privacy visual mode or a traditional visual mode based on the semantic label and the current task instruction; In the privacy vision mode, privacy raw data, radar point cloud data, and inertial measurement unit are collected by privacy camera, LiDAR, and inertial measurement unit, respectively. The privacy raw data is encrypted by privacy protection image data processing unit to obtain irreversible encrypted feature vector. Then, the service robot is controlled to perform target tasks based on the encrypted feature vector, the radar point cloud data, and the inertial data. In the traditional vision mode, non-privacy images and radar point cloud data are acquired by a general-purpose camera and a LiDAR, respectively, and the service robot is controlled to perform the target task based on the non-privacy images and the radar point cloud data.

[0009] In some embodiments, constructing a semantic map specifically includes: Environmental information is collected using LiDAR and visual cameras; A two-dimensional grid map is constructed based on the environmental information to obtain the environmental map; The environmental information is subjected to image recognition, and semantic labels are marked in different areas of the environmental map according to the image recognition results to obtain the semantic map.

[0010] In some embodiments, obtaining the semantic label corresponding to the current coordinate point based on the semantic map, and selecting a privacy-oriented visual mode or a traditional visual mode based on the semantic label and the current task instruction, specifically includes: Obtain the current coordinates of the service robot; The current coordinate point is matched with the semantic map to obtain the region polygon where the current coordinate point is located, and then the semantic label corresponding to the region polygon is obtained. Based on the semantic tags, the current task instructions, and the preset decision rule base, the privacy visual mode or the traditional visual mode is selected.

[0011] In some embodiments, the encryption of the original privacy data by the privacy-preserving image data processing unit to obtain an irreversible encrypted feature vector specifically includes: The original privacy data is subjected to black level correction to obtain the first privacy data; The first privacy data is subjected to median filtering to obtain the second privacy data; The second private data is input into a pre-trained feature extraction model, which outputs a feature vector. The feature extraction model is obtained through game-like adversarial training, making the feature vector irreversible. The feature vector is symmetrically encrypted and authenticated to obtain the irreversible encrypted feature vector.

[0012] In some embodiments, controlling the service robot to perform the target task based on the encrypted feature vector, the radar point cloud data, and the inertial data specifically includes: Decrypting the encrypted feature vector yields the plaintext feature vector; The plaintext feature vector and the inertial data are matched and fused to obtain the relative pose transformation of the service robot. The radar point cloud data is matched with the environmental map to obtain the pose correction amount of the service robot; A state vector is constructed, and an extended Kalman filter is performed to fuse the state vector, the relative pose transformation amount, and the pose correction amount to obtain the current pose estimate of the service robot. A hybrid sensing map is constructed based on the radar point cloud data, and path planning is performed based on the current pose estimation and the hybrid sensing map to obtain control commands. The service robot is controlled to perform the target task according to the control instructions.

[0013] In some embodiments, the method further includes: In the privacy vision mode, the power supply to the privacy camera is turned on or off in response to the user's physical triggering operation of the physical switch; If the power supply to the privacy camera is cut off, the service robot is controlled to perform the target task based on the radar point cloud data and the inertial data.

[0014] The embodiments of this application include at least the following beneficial effects: The service robot and its implementation method with hardware-level visual privacy protection capabilities of this application include a context perception and decision-making module, a data acquisition module, a privacy-protected image data processing unit, and a main controller module. This application integrates a dedicated privacy-protected image data processing unit within the robot body and physically connects it to a privacy camera, constructing a unique data path from the sensor to this unit. This directly converts the original privacy data into encrypted, irreversible feature vectors, ensuring that the service robot cannot output any visible images at the physical level in visual privacy mode. Furthermore, it automatically selects the visual mode based on the semantic label corresponding to the current coordinate point. In privacy visual mode, it fuses encrypted feature vectors, radar point cloud data, and inertial data to execute the target task; in traditional visual mode, it executes the target task based on non-privacy images and radar point cloud data. This application can provide necessary service functions while ensuring, at the physical level, that no image recording of the private environment is performed, thereby improving the security and reliability of the service robot's visual privacy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments of this application are described below. It should be understood that the drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions in this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a structural block diagram of a service robot with hardware-level visual privacy protection capabilities provided in one embodiment of this application; Figure 2 A structural block diagram of a service robot with hardware-level visual privacy protection capabilities provided in another embodiment of this application; Figure 3This is an overall structural block diagram of a service robot with hardware-level visual privacy protection capabilities provided in one embodiment of the present application; Figure 4 A flowchart illustrating the steps of a control method for a service robot with hardware-level visual privacy protection capabilities, provided in one embodiment of this application. Figure 5 This is a flowchart illustrating a control method for a service robot with hardware-level visual privacy protection capabilities according to an embodiment of this application. Figure 6 A flowchart illustrating the steps of constructing a semantic map according to one embodiment of this application; Figure 7 A flowchart illustrating the steps of visual mode selection according to one embodiment of this application; Figure 8 A flowchart illustrating the processing of a privacy-preserving image data processing unit provided in one embodiment of this application; Figure 9 This is a schematic diagram illustrating the execution of a target task in a privacy-oriented visual mode according to an embodiment of this application. Figure 10 A flowchart illustrating the steps of a control method for a service robot with hardware-level visual privacy protection capabilities, provided in another embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0019] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0020] Privacy-Preserving Image Data Processing Unit (PPIU) is used to replace the imaging link of traditional image signal processors (ISPs). The PPIU receives the raw RAW data output by the image sensor, processes it through a solidified pipeline, and outputs only encrypted feature vectors. It does not generate any visible images internally, thus achieving fundamental privacy protection at the hardware level.

[0021] Semantic map: A map that adds high-level semantic information (such as bedroom, living room, corridor) to the environmental geometric map built by the robot using SLAM technology.

[0022] Privacy Vision Mode: A working state of the robot's vision system. In this state, the privacy camera and privacy-preserving image data processing unit are activated, outputting only encrypted feature vectors for limited navigation, without generating or storing any images.

[0023] Privacy Vision Function Library: A software algorithm library specifically developed for processing encrypted feature vectors, providing limited functions such as visual odometry and obstacle detection, and is the foundation for robots to move in hardware privacy mode.

[0024] Privacy raw data: refers to the raw RAW data output by the privacy camera that has not been processed by the image signal processor (ISP); this data always exists in a non-visual form within the privacy-preserving image data processing unit, and no visible image is generated within the privacy-preserving image data processing unit.

[0025] Service robots are increasingly being used in hotels, homes, and healthcare settings to provide services such as item delivery, environmental inspection, and companionship. However, when they enter private spaces like bedrooms, hotel rooms, and hospital wards, their onboard cameras continuously collect environmental video or images, raising serious concerns among users about privacy breaches. Existing visual privacy protection solutions for service robots primarily rely on software algorithms to identify and mask sensitive areas (e.g., image content replacement or blurring) after capturing a complete visual image. Current technologies focus on post-processing remediation but fail to adequately consider a fundamental premise: the original image containing all privacy information already exists in the robot's system memory before software processing begins. Existing solutions largely depend on the accuracy and processing speed of software algorithms, making it impossible to defend against the risk of data being stolen by malicious programs or hardware vulnerabilities at the system's lower levels between image capture and processing. Essentially, this is a strategy of leaking first and then covering up, failing to address users' fundamental concerns about robot privacy and security.

[0026] Existing technology proposes a dual-mode privacy recording device and method based on the fusion of infrared thermal imaging and visible light. This scheme integrates dual-mode imaging switching and can be triggered based on geofence signals. However, this scheme has the following problems: 1. Incomplete privacy protection, with the risk of leakage of non-identity-related visual information still present: The output in privacy mode is infrared thermal distribution video. Although direct identity recognition information such as face and skin color is removed, it still contains visual information such as environmental spatial outline and human / object thermal outline, which can restore scene layout and personnel activity status. Moreover, this solution is essentially a switching of imaging source, and does not block the generation path of visual images at the hardware level. The original visible light image still has a leakage window before switching, which does not solve the root problem. Second, the mode switching is based on crude criteria and is prone to mis-triggering: The solution uses geofences (coordinate ranges) as the switching trigger condition, which cannot distinguish between rooms with different functions within the same coordinate range, such as the living room and bedroom, or the ward and the general office area on the same floor. This can easily lead to misjudgments of different privacy attributes at the same coordinates, requiring manual calibration by the user, and is not intelligent enough.

[0027] Third, the solution does not cover the mobility needs of service robots in privacy mode: This solution is only a recording device and does not involve the navigation, obstacle avoidance, and task execution logic of the mobile platform. Furthermore, it does not consider the core issue of how to maintain mobility after disabling conventional imaging. Therefore, it cannot directly meet the essential needs of service robots to deliver goods and conduct inspections in private spaces.

[0028] Fourth, no ultimate hardware control by the user: The mode switching of this solution relies on software signals / user commands, which can be bypassed by malicious programs at the underlying level. Users cannot completely cut off the power supply to the imaging unit at the hardware level, and there is no clear privacy status indication, resulting in low user trust.

[0029] In view of this, this application proposes a service robot with hardware-level visual privacy protection capabilities, including a context perception and decision-making module, a data acquisition module, a privacy-preserving image data processing unit, and a main controller module. This application integrates a dedicated privacy-preserving image data processing unit within the robot body and physically connects it to a privacy camera, establishing a unique data path from the sensor to this unit. This directly converts raw privacy data into encrypted, irreversible feature vectors, ensuring that the service robot cannot output any visible images at the physical level in a privacy-preserving visual mode. Furthermore, it automatically selects the visual mode based on the semantic label corresponding to the current coordinate point. In privacy-preserving visual mode, it fuses encrypted feature vectors, radar point cloud data, and inertial data to execute the target task; in traditional visual mode, it executes the target task based on non-privacy images and radar point cloud data. This application can provide necessary service functions while physically ensuring that no image recording is made in the private environment.

[0030] Reference Figure 1 , Figure 1 This is a structural block diagram of a service robot with hardware-level visual privacy protection capabilities according to an embodiment of this application. The embodiment of this application proposes a service robot with hardware-level visual privacy protection capabilities, which includes: The context awareness and decision-making module is used to build a semantic map, obtain the semantic label corresponding to the current coordinate point based on the semantic map, and obtain the visual mode switching instruction based on the semantic label and the current task instruction. The data acquisition module includes a privacy camera, a general-purpose camera, a lidar, and an inertial measurement unit, which are used to acquire privacy raw data, non-privacy images, radar point cloud data, and inertial data, respectively. A privacy-preserving image data processing unit, electrically connected to a privacy camera, is used to encrypt raw privacy data to obtain an irreversible encrypted feature vector. The main controller module, connected to the context awareness and decision-making module, the data acquisition module, and the privacy-preserving image data processing unit, is used to select either a privacy-preserving visual mode or a traditional visual mode according to the visual mode switching command. In the privacy-preserving visual mode, it executes the target task based on encrypted feature vectors, radar point cloud data, and inertial data. In the traditional visual mode, it executes the target task based on non-privacy images and radar point cloud data.

[0031] It should be noted that, to address the concern that existing service robots, when performing tasks in private spaces such as bedrooms and hotel rooms, inevitably collect, generate, and store complete images of the visual environment through their vision systems, thus failing to fundamentally eliminate user privacy concerns, this application proposes a service robot with hardware-level visual privacy protection capabilities. This robot, while providing necessary service functions, ensures through its hardware architecture that it does not record any images of the private environment it inhabits.

[0032] The service robot integrates a dedicated privacy-preserving image data processing unit (PPIU) and forms a unique physical connection with the privacy camera via PCB traces, creating a single data path from the sensor to this unit. It directly converts the collected raw RAW data (i.e., privacy-preserving raw data) into encrypted, irreversible feature vectors before outputting them, ensuring that the service robot cannot output any visible images at the physical level, fundamentally eliminating the possibility of raw privacy information being collected and recorded. The service robot is also equipped with a context awareness and decision-making module and a main controller module. Based on the current coordinates (such as a bedroom or guest room based on SLAM recognition) and the current task (such as delivering items or patrolling), it dynamically decides whether to use a privacy-preserving visual mode or a traditional visual mode. In privacy-preserving visual mode, it integrates encrypted feature vectors, radar point cloud data, and inertial data for path navigation and task execution; in traditional visual mode, it integrates non-privacy images and radar point cloud data for path navigation and task execution. This achieves an adaptive capability to perform complex tasks in public areas and automatically activate hardware-level privacy protection in private areas.

[0033] Specifically, the service robot includes a robot mobile chassis, a context perception and decision-making module, a data acquisition module, a privacy-preserving image data processing unit, and a main controller module; wherein: The robot's mobile chassis carries all upper-level modules and achieves precise positional movement and attitude adjustment based on navigation commands via wheeled or tracked mechanisms. The chassis employs a two-wheel differential drive combined with omnidirectional wheels. The drive motor is a DC brushed motor equipped with an optical encoder for odometer feedback, and an integrated gearbox. A collision detection sensor is mounted on the lower front end.

[0034] The data acquisition module includes at least a privacy camera dedicated to privacy protection mode, a general-purpose camera, a LiDAR, and an inertial measurement unit (IMU), used to acquire raw privacy data, non-privacy images, radar point cloud data, and inertial data, respectively. In privacy mode, the privacy camera provides the privacy-protected image data processing unit with raw, unprocessed RAW data, ensuring data invisibility from the source. This privacy camera is a low-sensitivity, low-resolution privacy-dedicated camera, such as a global shutter CMOS image sensor that supports outputting Bayer array RAW10 / RAW12 data with an MIPICSI-22-lane interface. The general-purpose camera can use a high-resolution rolling shutter sensor that supports autofocus and outputs partially processed YUV422 or RGB data for traditional vision tasks. The LiDAR can be a two-dimensional LiDAR, and the IMU can be a six-axis IMU.

[0035] The Privacy-Preserving Image Data Processing Unit (PPIU) receives raw RAW data (i.e., privacy-preserving raw data) from the image sensor, performs preprocessing, feature extraction, and encryption, and outputs only irreversible encrypted feature vectors without generating any visible images internally. It employs a dedicated neural network processing chip with a peak computing power of 1 TOPS (INT8) and integrates a MIPICSI-2 interface and a high-speed serial output interface (such as PCIe).

[0036] The context awareness and decision-making module comprehensively analyzes the robot's current environmental context and task objectives to determine the appropriate visual mode, enabling the robot to proactively and intelligently adapt to scenarios with varying privacy requirements. This module runs as software within the main controller module. Inputs include semantic location information from laser SLAM (such as whether coordinates (x,y) are within the bedroom polygon), current task instructions, and system status. It internally maintains a decision rule base and outputs visual mode switching instructions.

[0037] The main controller module processes sensor information, runs navigation and task programs, and receives, decrypts, and utilizes the encrypted feature vectors output by the privacy-preserving image data processing unit (PPIU). This module employs an embedded high-performance SoC (such as Rockchip RK3588 or NVIDIA Jetson OrinNano), integrating a multi-core CPU, GPU, and NPU, running a Linux-based robot operating system (ROS2), and featuring a built-in secure boot mechanism and a dedicated secure execution environment for key processing and decryption operations.

[0038] Furthermore, in this embodiment, the encrypted feature vector output by the Privacy-Preserving Image Data Processing Unit (PPIU) can only be decrypted using AES-256-GCM within the secure execution environment of the main controller module (such as TrustZone). The decrypted plaintext feature vector is only transmitted to the predefined privacy visual function library in the main controller module and is not allowed to be directly accessed by any user-space process or third-party application. Since the feature vector itself is irreversible, and the decryption process and secure memory area are protected by hardware isolation, even if the main controller module is compromised, the attacker cannot obtain reversible visual information.

[0039] Reference Figure 2 , Figure 2 The structural block diagram of a service robot with hardware-level visual privacy protection capabilities provided in another embodiment of this application, further includes, as an optional implementation, the service robot: The human-computer interaction module, connected to the main controller module, is used to receive user task commands and perform voice interaction or display the current visual mode according to the task commands. The power management module is connected to the main controller module, privacy camera, general-purpose camera, LiDAR, and inertial measurement unit, and is used to supply power to the main controller module, privacy camera, general-purpose camera, LiDAR, and inertial measurement unit.

[0040] Specifically, the human-computer interaction module receives user task commands, displays system status, provides voice feedback, and clearly indicates the current visual mode to enhance user experience and trust. This module includes a touchscreen display for showing a task list and status icons such as "Privacy Visual Mode," a microphone array, a speaker, and an LED indicator ring; it communicates with the backend server or the user's mobile app via Wi-Fi / 4G. When the robot is in privacy visual mode, the LED indicator ring on the front of the robot changes to a specific color (e.g., blue), a prominent "eye with a slash" icon is displayed on the robot's chest screen, and a voice announcement stating "Privacy visual mode has been entered" is provided to enhance user transparency and trust.

[0041] The power management module provides a stable and efficient power supply to all electronic modules of the robot and manages the charging and discharging of the batteries. This module uses a rechargeable lithium battery pack and integrates an intelligent battery management system, supporting power display, overcharge and over-discharge protection, and automatic recharging. It provides the required voltage to different modules through multiple DC-DC buck converters.

[0042] Reference Figure 3 , Figure 3 The diagram shows the overall structure of a service robot with hardware-level visual privacy protection capabilities according to one embodiment of this application. Further, as an optional implementation, the service robot may include: A physical switch, connected between the privacy camera and the power management module, is used to turn the privacy camera on or off.

[0043] Specifically, the physical switch allows users to cut off the power supply to the privacy camera with a direct physical action, establishing user trust. When the physical switch is open, the privacy camera is completely powered off, no RAW data flows into the PPIU's MIPI input, and its output is always a zero eigenvector. This physical switch consists of a mechanical self-locking switch, a Schmitt trigger debounce circuit, and a power MOSFET. The mechanical switch only outputs high and low level control signals and does not directly carry the power current supplied to the camera. After debounce by the debounce circuit, the level signal directly drives the gate of the power MOSFET, which controls the on / off state of the privacy camera's power supply circuit. The switch can also be equipped with an LED indicator to visually display the power supply status. The entire control chain is implemented purely in hardware, completely independent of the main controller software system, and cannot be bypassed or overridden by any program. The physical switch is located on the back or top of the robot's outer surface, with a protective ring and clear markings to prevent accidental activation. The entire control chain is physically isolated from the main controller software system and has the highest priority veto power.

[0044] It should be noted that, in this embodiment, an independent physical switch is provided on the outer surface of the service robot, allowing users to directly and physically cut off the power supply to the privacy camera. Simultaneously, the robot clearly indicates, through body lights, screen icons, or voice commands, that the current vision system is in privacy vision mode, granting users the highest level of control and informed consent, and constructing an intuitive and trustworthy interaction.

[0045] Reference Figure 3 As an optional implementation, the service robot also includes: The analog switch, privacy camera, and general camera are all connected to one end of the analog switch, and the privacy-protected image data processing unit and the main controller module are all connected to the other end of the analog switch. This is used to connect the data path between the privacy camera and the privacy-protected image data processing unit in privacy vision mode, and to connect the data path between the general camera and the main controller module in conventional vision mode.

[0046] Specifically, the analog switch dynamically switches the camera used by the robot and its corresponding data path according to the vision mode switching command, realizing hardware-level switching between privacy vision mode and traditional vision mode. This switch is implemented by a multiplexer chip, using a dual-channel MIPI CSI-2 analog switch (such as the MAX96705) to select whether to output signals from the privacy camera or the general-purpose camera to the subsequent processing device (PPIU or SoC built-in ISP). The switching is controlled by a GPIO pin level of the main controller module, rather than through software-level data routing. The physical characteristics of this analog switch ensure that only one data path is active at any given time—either connecting the privacy camera to the PPIU or the general-purpose camera to the traditional ISP; there is no intermediate state where both paths are active simultaneously. In privacy vision mode, the data line of the general-purpose camera is hardware disconnected, and any data output from it cannot reach the main controller, physically preventing parallel acquisition of two data paths.

[0047] The structure and working principle of the service robot with hardware-level visual privacy protection capabilities according to the embodiments of this application have been described above. It can be recognized that the embodiments of this application have the following advantages: First, the robot integrates a dedicated privacy-preserving image data processing unit, which is physically connected to the privacy camera to create a unique data path from the sensor to the unit. This directly converts the collected raw privacy data into encrypted, irreversible feature vectors before outputting them, ensuring that the service robot cannot output any visible images at the physical level, thus fundamentally eliminating the possibility of raw privacy information being collected and recorded.

[0048] Second, it is equipped with a context awareness and decision-making module and a main controller module, which can dynamically decide whether to use a privacy vision mode or a traditional vision mode based on the current coordinates and the current task. In the privacy vision mode, it integrates encrypted feature vectors, radar point cloud data and inertial data for path navigation and task execution. In the traditional vision mode, it integrates non-privacy images and radar point cloud data for path navigation and task execution, realizing the adaptive capability of performing complex tasks in public areas and automatically enabling hardware-level privacy protection in private areas.

[0049] Third, the encrypted feature vector output by the privacy-protected image data processing unit can only be decrypted within the secure execution environment of the main controller module. The decrypted plaintext feature vector is only transmitted to the privacy visual function library. Even if the main controller module is compromised, the attacker cannot obtain reversible visual information, which further improves the security of the original privacy data.

[0050] Fourth, an independent physical switch is installed on the outer surface of the service robot, allowing users to directly and physically cut off the power supply to the privacy camera. At the same time, the user can clearly indicate that the current vision system is in privacy vision mode through body lights, screen icons, or voice, which can give the user the highest level of control and right to know, and build an intuitive and trustworthy interaction.

[0051] Reference Figure 4 and Figure 5 , Figure 4 This is a flowchart illustrating the steps of a control method for a service robot with hardware-level visual privacy protection capabilities according to an embodiment of this application. Figure 5 This is a flowchart illustrating a control method for a service robot with hardware-level visual privacy protection capabilities according to an embodiment of this application. The embodiment of this application provides a control method for a service robot with hardware-level visual privacy protection capabilities, including the following steps S101 to S103: Step S101: Construct a semantic map, obtain the semantic label corresponding to the current coordinate point based on the semantic map, and select either privacy visual mode or traditional visual mode based on the semantic label and the current task instruction. Reference Figure 6 , Figure 6 The flowchart for constructing a semantic map provided in one embodiment of this application is further, as an optional implementation, the step of constructing the semantic map can be further divided into the following steps S1011 to S1013: Step S1011: Collect environmental information using LiDAR and a visual camera; Step S1012: Construct a two-dimensional grid map based on the environmental information to obtain an environmental map; Step S1013: Perform image recognition on the environmental information, and label the corresponding semantic tags in different areas of the environmental map according to the image recognition results to obtain a semantic map.

[0052] Specifically, users can activate the mapping function through an app or the robot's own interface. The robot autonomously moves within its environment using its onboard LiDAR and traditional vision cameras, running SLAM algorithms for simultaneous localization and map building, simultaneously constructing a two-dimensional grid map of the environment. Simultaneously, semantic labels are assigned to connected regions in the map using image recognition or pre-defined rules. For example, an area containing a bed is labeled as a bedroom, an area containing a toilet and bathtub is labeled as a bathroom, and an open area is labeled as a living room. Ultimately, a semantic map is generated, containing (x,y) coordinates, occupancy status, and semantic labels.

[0053] Reference Figure 7 , Figure 7The flowchart of the steps for visual mode selection provided in one embodiment of this application is further, as an optional implementation, the step of obtaining the semantic label corresponding to the current coordinate point based on the semantic map, and selecting the privacy visual mode or the traditional visual mode based on the semantic label and the current task instruction, can be further divided into the following steps S1014 to S1016: Step S1014: Obtain the current coordinates of the service robot; Step S1015: Match the current coordinate point with the semantic map to obtain the region polygon where the current coordinate point is located, and then obtain the semantic label corresponding to the region polygon; Step S1016: Select either privacy visual mode or traditional visual mode based on semantic tags and a preset decision rule library.

[0054] Specifically, to achieve adaptive control of the robot's visual behavior, the context perception and decision-making module running on the robot's main controller executes position-label-rule query and judgment logic when the robot's pose is updated or a new task is received. First, it queries the robot's current pose in real time. current =(x,y,θ), where (x,y) is the current coordinate point and θ is the orientation. Next, the current coordinate point (x,y) is matched with the semantic map to query the polygon area containing that coordinate point and obtain its semantic label L; for example, L=query label The algorithm returns "Bedroom" from a map (x, y). Then, it selects either a privacy-focused visual mode or a traditional visual mode based on an internally pre-defined decision rule base. For example, if L belongs to {"Bedroom", "Bathroom", "Guest Room"}, the target mode is privacy-focused; otherwise, the target mode is traditional. Finally, the calculated visual mode switching command is sent to the main controller module. Alternatively, when a task command pointing to a private area is received (such as delivering goods into a bedroom), a visual mode switching command to switch to privacy-focused visual mode is generated directly without waiting for the robot to actually enter the private area.

[0055] Step S102: In privacy vision mode, privacy raw data, radar point cloud data and inertial data are collected by privacy camera, LiDAR and inertial measurement unit respectively. Privacy raw data is encrypted by privacy protection image data processing unit to obtain irreversible encrypted feature vector. Then, the service robot is controlled to perform target tasks according to the encrypted feature vector, radar point cloud data and inertial data. Specifically, when the main controller module determines that the visual mode is privacy visual mode based on the visual mode switching command output by the context awareness and decision-making module, it sets the GPIO pin of the main controller module low to turn off the power of the general-purpose camera, configures the MIPI switch chip (i.e., analog switch) via I2C to switch the data path from the traditional ISP to the PPIU unit, and then sets the GPIO pin high to turn on the power of the privacy camera, sending a start command to the PPIU unit via I2C. During the switching process, the robot relies on wheel odometry and inertial measurement unit to maintain short-term pose estimation.

[0056] After the camera switch is complete, the system no longer receives images. Instead, it receives encrypted feature streams (including initialization vector IV, ciphertext C, and authentication tag), radar point cloud data, and inertial data from the PPIU unit. In a secure operating environment, the main controller module uses a preset key K to decrypt the ciphertext C using AES-256-GCM. After verifying the integrity of the authentication tag, the 128-dimensional plaintext feature vector F is recovered. Next, the decrypted feature stream is used for feature point matching to estimate inter-frame motion, and then fused with the inertial data to obtain a high-frequency pose estimate. The radar point cloud data is then matched with the constructed grid map to obtain a low-drift pose correction. Finally, through Kalman filter or graph optimization, the results of the visual-inertial odometry, LiDAR localization, and wheel odometry are fused to output a stable and accurate robot current pose estimate. fused It is used for subsequent path planning and obstacle avoidance.

[0057] Reference Figure 8 , Figure 8 The flowchart of the privacy-preserving image data processing unit provided in one embodiment of this application is further shown. As an optional implementation, the step of encrypting the original privacy data by the privacy-preserving image data processing unit to obtain the encrypted feature vector can be further divided into the following steps S1021 to S1024: Step S1021: Perform black level correction on the original privacy data to obtain the first privacy data; Step S1022: Perform median filtering on the first privacy data to obtain the second privacy data; Step S1023: Input the second privacy data into the pre-trained feature extraction model and output the feature vector. The feature extraction model is obtained through game-like adversarial training, which makes the feature vector irreversible. Step S1024: Perform symmetric encryption and authentication on the feature vector to obtain an irreversible encrypted feature vector.

[0058] Specifically, the privacy-preserving image data processing unit first performs black-level correction on the input RAWBayer data (i.e., the original privacy data) using the loaded BLC[col] table to obtain the first privacy data. Then, it performs median filtering on the first privacy data to repair bad pixels based on a 3x3 window, resulting in the second privacy data. Next, it embeds a feature extraction model with a lightweight network as its backbone. The model receives a 4-channel tensor input after Bayer reconstruction, goes through several layers of convolution and pooling, and finally outputs a 128-dimensional feature vector F through global average pooling and fully connected layers. This feature vector F is then encrypted by the integrated hardware AES-256-GCM engine. The encryption process is: C, Tag = AES-256-GCM_Encrypt(K, IV, F, AAD), where K is a 256-bit symmetric key, IV is a real-time generated initialization vector, and AAD is additional authentication data. After encryption, the PPIU unit encapsulates the initialization vector IV, ciphertext C, and authentication tag into a data packet conforming to the MIPI CSI-2 protocol. The data packet structure is: [Frame Header Synchronization Word][IV(12B)][C Length Field][C Data][Tag(16B)][Frame Tail Check Word], and sends it to the main processor through the MIPI CSI-2 Tx interface.

[0059] In this embodiment, the feature extraction model is generated through game-theoretic adversarial training. The training framework comprises a feature extractor, an image reconstruction network (simulating a reverse attacker), and a discriminator. The optimization objective is to maximize the irreversibility of features while ensuring task accuracy. A joint loss function L is used during training. total = 1.0 L task + 0.05 L percept +0.01 L adv Optimize the feature extractor, where L task Cross-entropy loss is used to ensure the effectiveness of features in the task; L percept The perceptual loss based on VGG19 features is used to constrain features to preserve high-level semantics; L adv The least-squares adversarial loss is optimized so that the reconstructed image is judged as false by the discriminator. That is, the more difficult it is for the features to be reversed, the lower the loss is, and finally the balance between the task effectiveness of the features and the visual irreversibility is achieved.

[0060] As an optional implementation, the step of controlling the service robot to perform the target task based on the encrypted feature vector, radar point cloud data, and inertial data can be further divided into the following steps S1025 to S10210: Step S1025: Decrypt the encrypted feature vector to obtain the plaintext feature vector; Step S1026: Perform feature point matching and fusion on the plaintext feature vector and inertial data to obtain the relative pose transformation of the service robot; Step S1027: Match the radar point cloud data with the environmental map to obtain the pose correction amount of the service robot; Step S1028: Construct a state vector, and perform extended Kalman filter fusion based on the state vector, relative pose transformation amount and pose correction amount to obtain the current pose estimate of the service robot. Specifically, such as Figure 9 The diagram illustrates the execution flow of the target task in privacy vision mode. First, the privacy camera, LiDAR, and inertial measurement unit required for navigation in privacy vision mode are activated. Each sensor begins publishing data through ROS topics, preparing for multi-source data fusion. Next, multi-source data is collected synchronously. The `message_filters` tool is used to perform approximate time synchronization on topics from ` / camera / features` (encrypted features), ` / imu / data` (inertial data), and ` / scan` (radar point cloud data), generating time-aligned data packets for subsequent processing.

[0061] Then, the encrypted feature vector is decrypted in a secure operating environment by the main controller module, recovering the 128-dimensional plaintext feature vector F. Next, multi-sensor fusion positioning is performed, using the decrypted feature stream and inertial data as input, and running the visual-inertial odometry calculation method. Let F be the set of feature points at time k and time k+1. k and F k+1 By combining feature descriptor matching and epipolar geometry constraints with inertial data pre-integration, a relative pose transformation quantity T is calculated. k,k+1 And give a covariance matrix Σ vio This indicates its uncertainty. Simultaneously, the ICP or CSM algorithm is used to analyze the current frame's laser point cloud L... k+1 Matching with a local submap, a pose correction T is calculated. lidar and its covariance Σ lidar Finally, a system is established containing position p, attitude q, velocity v, and accelerometer bias b. a gyroscope bias b g The state vector x=[p,q,v,b] a ,b g ] T Then, the relative pose change T provided by the visual inertial odometry k,k+1 As a prediction step, the pose correction T provided by the lidar matching is used. lidar As an observation update step, an extended Kalman filter update is performed to obtain the optimal current pose estimate. fused and its covariance.

[0062] Step S1029: Construct a hybrid perception map based on radar point cloud data, perform path planning based on current pose estimation and hybrid perception map, and obtain control commands; Step S10210: Control the service robot to perform the target task according to the control instructions.

[0063] Specifically, such as Figure 9 As shown, firstly, a global cost map and a local cost map are generated. Radar point cloud data is inserted into the cost map and obstacle marking is performed. Simultaneously, plaintext feature vectors are input into a lightweight classifier to identify potential obstacles or special regions in the foreground area. The classification results are then filled into the corresponding regions of the cost map with different costs, resulting in a hybrid sensing map. Finally, a local planner such as DWA or TEB is used to estimate the pose based on the current pose. fused Starting from the target and ending at the objective, the robot performs real-time path search on the constructed hybrid perception map. The planner considers the robot's kinematic constraints to avoid high-cost areas, generates linear velocity and angular velocity commands (v, ω), and sends them to the chassis motor controller for execution.

[0064] In privacy vision mode, the target tasks also include: matching the decrypted plaintext feature vector with the local pre-encrypted target item feature library, and determining that the target item has been found when the matching degree exceeds the threshold, without involving image comparison throughout the process; or, analyzing the change pattern of feature vectors in consecutive frames, and inferring abnormal events by combining sound sensor information when the change pattern meets the abnormal conditions, and going to the corresponding rough location to conduct voice inquiry; after the robot moves to the target location, it completes the delivery through a robotic arm or voice prompts, and the entire process data log is an encrypted feature sequence that cannot be restored to a scene image.

[0065] Furthermore, when the current pose is detected, the pose is estimated. fused Once the target mode is switched to traditional vision mode after leaving the polygonal area with the semantic label "private" or when the task state machine reports task completion, the privacy camera is turned off, the general camera is turned on, and the data path is switched back to the traditional ISP.

[0066] Step S103: In the traditional vision mode, non-privacy images and radar point cloud data are collected by a general-purpose camera and a LiDAR, respectively. The service robot is controlled to perform the target task based on the non-privacy images and radar point cloud data.

[0067] Specifically, when the main controller module determines that the visual mode is the traditional visual mode based on the visual mode switching command output by the context awareness and decision-making module, it sends a command to the analog switch to connect the data path of the general-purpose camera to the traditional image signal processor inside the main controller module. Simultaneously, it sends power-on and start signals to the general-purpose camera and power-off or standby signals to the privacy-protected camera. The traditional ISP then begins operation, receiving raw RAW image data (i.e., non-privacy images) from the general-purpose camera and executing a complete image processing pipeline to obtain high-quality processed YUV or RGB image frames. Next, a feature-based visual SLAM algorithm (such as ORB-SLAM3) is run for localization, and combined with radar point cloud data for fusion navigation. After successful localization and approaching the target, high-definition vision is used to complete target tasks requiring fine-grained identification, such as accurately grasping specific objects, reading text information, or performing precise docking operations.

[0068] Reference Figure 10 , Figure 10 The flowchart illustrates the steps of a control method for a service robot with hardware-level visual privacy protection capabilities, provided in another embodiment of this application. Further, as an optional implementation, the control method may include the following steps S104 to S105: Step S104: In privacy vision mode, in response to the user's physical triggering operation on the physical switch, turn on or off the power supply to the privacy camera. Step S105: If the power supply to the privacy camera is cut off, control the service robot to perform the target task based on radar point cloud data and inertial data.

[0069] Specifically, in privacy vision mode, if a physical trigger operation is received to turn off the physical switch by the user, or if the privacy-protected image data processing unit stops outputting feature streams due to a malfunction, navigation and positioning can be performed solely based on radar point cloud data and inertial data. In this case, the positioning accuracy will decrease but will not collapse. This embodiment of the application, at the cost of sacrificing some positioning accuracy, resolutely blocks the privacy imaging path, effectively reducing the risk of robot visual privacy leakage due to system anomalies.

[0070] It is understood that the content of the above service robot embodiments is applicable to this method embodiment. The specific functions implemented in this method embodiment are the same as those in the above service robot embodiments, and the beneficial effects achieved are also the same as those achieved in the above service robot embodiments.

[0071] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0072] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0073] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0075] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0076] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0078] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0079] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0080] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A service robot with hardware-level visual privacy protection capabilities, characterized in that, The service robots include: The context awareness and decision-making module is used to construct a semantic map, obtain the semantic label corresponding to the current coordinate point based on the semantic map, and obtain a visual mode switching instruction based on the semantic label and the current task instruction. The data acquisition module includes a privacy camera, a general-purpose camera, a lidar, and an inertial measurement unit, which are used to acquire privacy raw data, non-privacy images, radar point cloud data, and inertial data, respectively. A privacy-protected image data processing unit, electrically connected to the privacy camera, is used to encrypt the original privacy data to obtain an irreversible encrypted feature vector. The main controller module, connected to the context awareness and decision-making module, the data acquisition module, and the privacy-preserving image data processing unit, is used to select either a privacy-preserving visual mode or a traditional visual mode according to the visual mode switching instruction. In the privacy-preserving visual mode, it executes the target task based on the encrypted feature vector, the radar point cloud data, and the inertial data. In the traditional visual mode, it executes the target task based on the non-privacy image and the radar point cloud data.

2. The service robot according to claim 1, characterized in that, The service robot also includes: The human-computer interaction module is connected to the main controller module and is used to receive user task instructions and perform voice interaction or display the current visual mode according to the task instructions. The power management module is connected to the main controller module, the privacy camera, the general-purpose camera, the lidar, and the inertial measurement unit, and is used to supply power to the main controller module, the privacy camera, the general-purpose camera, the lidar, and the inertial measurement unit.

3. The service robot according to claim 2, characterized in that, The service robot also includes: A physical switch, connected between the privacy camera and the power management module, is used to turn the power supply to the privacy camera on or off.

4. The service robot according to claim 1, characterized in that, The service robot also includes: An analog switch is provided, and both the privacy camera and the general-purpose camera are connected to one end of the analog switch. The privacy-protected image data processing unit and the main controller module are also connected to the other end of the analog switch. This is used to connect the data path between the privacy camera and the privacy-protected image data processing unit in the privacy vision mode, and to connect the data path between the general-purpose camera and the main controller module in the conventional vision mode.

5. A control method for a service robot with hardware-level visual privacy protection capabilities, characterized in that, The method for controlling a service robot with hardware-level visual privacy protection capabilities as described in any one of claims 1 to 4 includes the following steps: Construct a semantic map, obtain the semantic label corresponding to the current coordinate point based on the semantic map, and select a privacy visual mode or a traditional visual mode based on the semantic label and the current task instruction; In the privacy vision mode, privacy raw data, radar point cloud data, and inertial measurement unit are collected by privacy camera, LiDAR, and inertial measurement unit, respectively. The privacy raw data is encrypted by privacy protection image data processing unit to obtain irreversible encrypted feature vector. Then, the service robot is controlled to perform target tasks based on the encrypted feature vector, the radar point cloud data, and the inertial data. In the traditional vision mode, non-privacy images and radar point cloud data are acquired by a general-purpose camera and a LiDAR, respectively, and the service robot is controlled to perform the target task based on the non-privacy images and the radar point cloud data.

6. The method according to claim 5, characterized in that, The construction of the semantic map specifically includes: Environmental information is collected using LiDAR and visual cameras; A two-dimensional grid map is constructed based on the environmental information to obtain the environmental map; The environmental information is subjected to image recognition, and semantic labels are marked in different areas of the environmental map according to the image recognition results to obtain the semantic map.

7. The method according to claim 5, characterized in that, The step of obtaining the semantic label corresponding to the current coordinate point based on the semantic map, and selecting the privacy visual mode or the traditional visual mode based on the semantic label and the current task instruction, specifically includes: Obtain the current coordinates of the service robot; The current coordinate point is matched with the semantic map to obtain the region polygon where the current coordinate point is located, and then the semantic label corresponding to the region polygon is obtained. Based on the semantic tags, the current task instructions, and the preset decision rule base, the privacy visual mode or the traditional visual mode is selected.

8. The method according to claim 5, characterized in that, The encryption process performed by the privacy-preserving image data processing unit to obtain an irreversible encrypted feature vector specifically includes: The original privacy data is subjected to black level correction to obtain the first privacy data; The first privacy data is subjected to median filtering to obtain the second privacy data; The second private data is input into a pre-trained feature extraction model, which outputs a feature vector. The feature extraction model is obtained through game-like adversarial training, making the feature vector irreversible. The feature vector is symmetrically encrypted and authenticated to obtain the irreversible encrypted feature vector.

9. The method according to claim 6, characterized in that, The step of controlling the service robot to perform the target task based on the encrypted feature vector, the radar point cloud data, and the inertial data specifically includes: Decrypting the encrypted feature vector yields the plaintext feature vector; The plaintext feature vector and the inertial data are matched and fused to obtain the relative pose transformation of the service robot. The radar point cloud data is matched with the environmental map to obtain the pose correction amount of the service robot; A state vector is constructed, and an extended Kalman filter is performed to fuse the state vector, the relative pose transformation amount, and the pose correction amount to obtain the current pose estimate of the service robot. A hybrid sensing map is constructed based on the radar point cloud data, and path planning is performed based on the current pose estimation and the hybrid sensing map to obtain control commands. The service robot is controlled to perform the target task according to the control instructions.

10. The method according to any one of claims 5 to 9, characterized in that, The method further includes: In the privacy vision mode, the power supply to the privacy camera is turned on or off in response to the user's physical triggering operation of the physical switch; If the power supply to the privacy camera is cut off, the service robot is controlled to perform the target task based on the radar point cloud data and the inertial data.