Guide robot system based on high-computing-power ARM64-NPU module

Abstract

The invention relates to the technical field of intelligent service robots, and discloses a navigation robot system based on a high-computing-power ARM64-NPU module, which comprises a hardware layer, a sensing and data processing layer, a scene dynamic understanding and decision engine, a service strategy adaptive adjustment module and an application and interactive execution layer. The core method is as follows: a decision engine judges a current scene state by performing quantitative evaluation on multi-modal sensing data; and when the scene state is switched, the service strategy adaptive adjustment module matches a new service strategy from the strategy library, and gradually adjusts the moving speed and voice volume behavior parameters of the robot from old strategy values to new strategy values in a non-abrupt manner in a preset time window by adopting a smooth transition algorithm. The navigation robot can sense the environment in real time and intelligently and smoothly adjust self behaviors, and the service adaptability and the user interaction experience in complex and changeable scenes are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent service robot technology, specifically a tour guide robot system based on a high-performance ARM64-NPU module. Background Technology

[0002] With the continuous maturation of artificial intelligence and robotics technologies, tour guide robots, whose main functions are providing information consultation and guidance services, have become an important part of the intelligent upgrading of public places such as museums, corporate exhibition halls, commercial centers, and transportation hubs. These robots aim to improve service efficiency and visitor experience through autonomous movement and human-computer interaction.

[0003] However, existing tour guide robots generally suffer from a deep-seated problem of insufficient adaptability to complex and dynamic environments. On the one hand, their service strategies are pre-programmed and fixed, with movement speed, voice interaction volume, and interactive action parameters set to a fixed set of values ​​to deal with idealized or most common single scenarios. This static configuration leads to poor performance of the robots in real and changing environments. When a robot carefully configured in a quiet exhibition hall enters a noisy, crowded commercial event, its gentle voice will be completely drowned out by the ambient noise, and its slow movement will hinder passage, significantly reducing the service effect. This exposes the robot's lack of ability to perceive, quantify, and adjust its behavior accordingly in real time.

[0004] Meanwhile, the lack of adaptability also stems from the inherent limitations of the underlying hardware and system architecture. To achieve true environmental adaptation, robots not only need intelligent algorithmic strategies but also powerful hardware support. Existing robot platforms often lack core computing power, making it difficult to smoothly run complex real-time environmental perception and decision-making algorithms. Furthermore, their far-field speech recognition capabilities in noisy environments, the stability of network connections in large or outdoor spaces, and the continuity and accuracy of positioning when switching between different scenarios all face significant challenges. Communication methods relying entirely on a single Wi-Fi connection and positioning schemes based solely on vision are prone to failure in environments with signal obstruction or drastic changes in lighting, leading to service interruptions.

[0005] Furthermore, even when some systems attempt to introduce multiple working modes, the switching process is often abrupt and sudden. Sudden changes in behavioral parameters, such as a sudden increase in speed or volume, can feel jarring and uncomfortable for users, undermining the naturalness and user-friendliness of human-computer interaction. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a guide robot system based on a high-performance ARM64-NPU module. This system solves the hardware limitations of existing technologies, such as robots' inability to intelligently adapt to environmental changes due to fixed behavioral parameters, and insufficient core computing power, far-field speech recognition, network connectivity, and positioning stability, which prevent the support of advanced adaptive functions. Furthermore, it eliminates the unnatural feeling caused by abrupt and jarring transitions in robot behavior patterns.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a tour guide robot system based on a high-performance ARM64-NPU module, comprising;

[0008] The hardware layer includes a core computing unit for providing core computing power, a multimodal sensing unit for collecting environmental and interactive data, and an execution unit for performing physical actions.

[0009] The perception and data processing layer, connected to the hardware layer, is used to process the raw data collected by the multimodal perception unit, thereby extracting structured feature data.

[0010] The scene dynamic understanding and decision engine is connected to the perception and data processing layer. It is used to receive the structured feature data, construct a scene state vector based on the structured feature data, classify the current scene into a predefined scene state, and output scene data.

[0011] The service strategy adaptive adjustment module is connected to the scenario dynamic understanding and decision engine. It is used to match and activate the corresponding service strategy from the preset service strategy library according to the scenario data. The service strategy includes a set of behavioral parameters to guide the robot's behavior.

[0012] The application and interaction execution layer is connected to the service strategy adaptive adjustment module and the hardware layer, and is used to receive the service strategy and control the execution unit to perform corresponding actions according to the behavior parameters used to guide the robot's behavior.

[0013] Through the above technical solutions, a complete closed-loop adaptive system from perception and decision-making to execution has been constructed. Through a layered and decoupled architecture, the robot can quantitatively understand environmental changes in real time and automatically match and adjust service strategies. This solves the technical problem of existing robots being unable to adapt to dynamic scenarios due to rigid strategies, significantly improving the robot's intelligence level and environmental adaptability.

[0014] Preferably, the structured feature data extracted by the perception and data processing layer includes:

[0015] The ambient noise level obtained by the audio system in the multimodal sensing unit;

[0016] Crowd density and lighting conditions obtained by the visual system.

[0017] The above technical solutions provide specific and quantifiable key judgment criteria for scene dynamic understanding and decision-making engines. By transforming complex raw perception data into intuitive structured features such as noise levels and crowd density, the system can more accurately and reliably assess the actual state of the current environment, laying a solid data foundation for subsequent scene classification and adaptive strategy adjustment.

[0018] Preferably, the scene dynamic understanding and decision engine normalizes and weights multiple structured feature data output by the perception and data processing layer to calculate a comprehensive environment evaluation score, thereby constructing the scene state vector.

[0019] The above technical solution integrates multi-dimensional and heterogeneous environmental features into a single, standardized evaluation score. This quantification method simplifies the logic for judging scene complexity, enabling the system to objectively and continuously evaluate the environment based on a clear numerical value, effectively improving the accuracy of scene state judgment.

[0020] Preferably, the scene dynamic understanding and decision engine further includes a scene state machine, which is used to classify the current scene into the predefined scene state according to the scene state vector.

[0021] The above technical solutions provide a clear and stable logical framework for switching scene states. The use of a scene state machine avoids frequent and unstable state fluctuations near critical values, ensuring the stability and consistency of the robot's service mode. At the same time, the structure also makes the definition and expansion of scene states clearer and more convenient.

[0022] Preferably, the service policy library stores multiple service policies, and each service policy is a set of behavioral parameters corresponding to specific scenario data;

[0023] The behavioral parameters include maximum movement speed, safe obstacle avoidance distance, and broadcast volume.

[0024] Through the above technical solution, the scene state is directly linked to the specific robot behavior. The service strategy library provides an optimal combination of behavioral parameters for each predefined scene state, making the robot's adaptive adjustment based on evidence. This ensures that the robot's behavior in different scenarios not only meets functional requirements but also takes into account safety and user experience.

[0025] Preferably, the process by which the service policy adaptive adjustment module activates the corresponding service policy includes:

[0026] When the scene data output by the scene dynamic understanding and decision engine is switched, the behavioral parameters contained in the two service strategies before and after the switch are smoothly transitioned within a preset time window.

[0027] The above technical solution solves the problem of abrupt and jarring robot behavior mode switching in existing technologies. By smoothing the behavior parameters within a time window, the robot's state adjustment becomes a gradual and continuous process.

[0028] Preferably, the multimodal sensing unit includes:

[0029] Microphone arrays used for far-field speech pickup and sound source localization;

[0030] A camera used to acquire visual information;

[0031] A multi-source fusion positioning system consisting of Bluetooth beacons, an ultra-wideband module, a GPS module, and visual SLAM technology.

[0032] The above technical solutions provide the system with high-quality and highly reliable raw perception data input, ensuring the accuracy of environmental perception from the source. The microphone array ensures effective voice interaction in noisy environments, while the multi-source fusion positioning system overcomes the limitations of single technologies in different scenarios, guaranteeing that the robot can achieve stable and accurate positioning and navigation in any environment.

[0033] Preferably, the core computing unit of the hardware layer includes an ARM64 architecture processor and a neural network processing unit, wherein the neural network processing unit is used to perform visual information processing for the perception and data processing layer.

[0034] The above technical solutions provide the system with powerful real-time computing capabilities, solving the bottleneck of insufficient computing power in existing platforms. The ARM64 processor ensures the smooth operation of business logic, while the dedicated neural network processing unit efficiently accelerates the complex visual AI algorithms for crowd density recognition, ensuring that the entire adaptive system can perform real-time perception and decision-making without lag or delay.

[0035] Preferably, the application and interaction execution layer is specifically used to: control the execution unit to move according to the navigation-related behavior parameters in the service strategy, and control the execution unit to perform voice broadcast according to the voice-related behavior parameters.

[0036] The above technical solution clarifies the transformation path from service strategy to physical execution. The application and interaction execution layer precisely parses abstract behavioral parameters into specific control commands for the robot chassis and voice broadcast hardware, ensuring that upper-level decisions are executed accurately and without compromise, thus completing the closed loop from decision-making to behavior.

[0037] Preferably, the hardware layer further includes a communication unit;

[0038] The communication unit receives update instructions from the cloud server in conjunction with the service policy adaptive adjustment module, thereby dynamically updating the content of the service policy library.

[0039] The above technical solution empowers the system with remote update and iteration capabilities. Operators can conveniently update or add content to the service strategy library via a cloud server based on new application scenarios or continuous optimization experience, without requiring on-site intervention on the robot itself. This effectively improves the system's maintainability, scalability, and long-term operational efficiency.

[0040] This invention provides a tour guide robot system based on a high-performance ARM64-NPU module. It offers the following advantages:

[0041] 1. This invention uses a multimodal perception unit to collect real-time data on environmental noise levels, crowd density, and lighting conditions. The perception and data processing layer extracts structured feature data. A scene dynamic understanding and decision engine constructs a scene state vector based on these features and classifies the scene states. This drives a service strategy adaptive adjustment module to match and activate corresponding service strategies from a service strategy library, and ensures a smooth transition of behavioral parameters during scene switching. This allows the robot to automatically adjust its maximum movement speed, safe obstacle avoidance distance, and voice broadcast volume according to real-time changes in its environment, adapting to different operating scenarios.

[0042] 2. This invention ensures stable data transmission between the robot and the cloud server through a built-in 5G communication module, unaffected by the coverage limitations of traditional Wi-Fi. Simultaneously, the integrated multi-source fusion positioning system, composed of a Bluetooth beacon, an ultra-wideband module, a GPS module, and visual SLAM technology, provides continuous pose information both indoors and outdoors, and in stationary or moving states. The neural network processing unit in the core computing unit provides computational support for visual information processing, assisting in positioning accuracy and ensuring the robot's continuous operation and accurate positional perception over a wide area.

[0043] 3. This invention provides computing power for the system to handle complex AI algorithms and business logic by integrating an ARM64 architecture processor and a neural network processing unit in the core computing unit. The communication unit and service policy adaptive adjustment module in the hardware layer work together to dynamically receive update instructions from the cloud server, thereby dynamically updating the content of the service policy library. Simultaneously, the closed-loop control system composed of the hardware layer, perception and data processing layer, scene dynamic understanding and decision engine, service policy adaptive adjustment module, and application and interaction execution layer ensures efficient data flow and collaborative work between modules, enabling the system to cope with long-term service scenarios. Attached Figure Description

[0044] Figure 1 This is a block diagram of the overall architecture of the tour guide robot system of the present invention;

[0045] Figure 2 This is a detailed structural block diagram of the hardware layer of the present invention;

[0046] Figure 3 This is a flowchart illustrating the scene dynamic understanding and decision-making engine of the present invention.

[0047] Figure 4 This is a flowchart illustrating the service strategy adaptive adjustment module of the present invention. Detailed Implementation

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] To better understand the present invention, the above content will be described in detail below with reference to specific embodiments.

[0050] Please see the appendix Figure 1 - Appendix Figure 4 This invention provides a tour guide robot system based on a high-performance ARM64-NPU module. The tour guide robot system adopts a layered and decoupled architecture design, which consists of the following layers from bottom to top: hardware layer, perception and data processing layer, scene dynamic understanding and decision engine, service strategy adaptive adjustment module, and application and interaction execution layer.

[0051] The hardware layer is the physical foundation of the entire system, responsible for perception, computation, execution, and communication. In this embodiment, the hardware layer is based on an advanced humanoid robot platform and has been deeply customized and developed, including a core computing unit, a multimodal perception unit, an execution unit, and a communication unit.

[0052] The core computing unit is the brain of the system, responsible for all data processing, algorithm execution, and business logic decisions. In this embodiment, the core computing unit is implemented in an embedded computing device that integrates a high-performance ARM64 architecture processor and a neural network processing unit (NPU). The embedded computing device is robustly integrated into the back of the robot platform using specially designed custom structural components, ensuring stability during movement.

[0053] Specifically, the ARM64 architecture processor possesses powerful general-purpose computing capabilities, responsible for executing the robot's operating system, scheduling core service processes, and handling upper-layer general business logic. The neural network processing unit, through a highly parallel computing architecture, is dedicated to efficiently accelerating artificial intelligence algorithms involving large-scale matrix operations and tensor processing, especially for convolutional neural network inference tasks required for visual information processing in subsequent perception and data processing layers. This heterogeneous computing architecture design achieves a rational allocation of computing resources, ensuring that the system can maintain both the smooth operation of core business processes and high energy efficiency when executing complex AI tasks.

[0054] Furthermore, to address more complex application scenarios in the future, the hardware layer also reserves standardized expansion interfaces, which can be used to further integrate additional computing units or specific functional modules through a dedicated smart backpack, enabling on-demand enhancement of computing power. These expansion interfaces provide the system with excellent modular scalability.

[0055] A multimodal sensing unit is a set of highly integrated heterogeneous sensor components whose function is to acquire real-time data on the robot's environment and interacting objects in a multi-dimensional and continuous manner. In this embodiment, the multimodal sensing unit mainly includes:

[0056] Microphone array: Used for far-field voice pickup and sound source localization. Preferably, a linear four-microphone array configuration can be used, with deep integration of echo cancellation and beamforming audio processing technologies, aiming to achieve effective pickup and localization of sound sources within a 180-degree radius in front of the robot, up to 10 meters away. To achieve optimal sound pickup, the microphone array is integrated into a customized front panel that replaces the robot's original chest armor. This design not only optimizes the microphone's pickup orientation but also improves structural stability and overall aesthetics.

[0057] High frame rate cameras: one or more, used to capture real-time video stream data of the environment. This visual information is the basis for subsequent crowd density analysis, lighting condition assessment, face recognition, and visual localization.

[0058] Multi-source fusion positioning system: This system combines multiple positioning technologies to adapt to high-precision positioning needs in different indoor and outdoor scenarios through a complementary approach. Specifically, it includes:

[0059] Bluetooth beacon receiver: By receiving signals broadcast by multiple Bluetooth beacons deployed in the environment, it assists in area-level positioning in indoor environments, enabling rapid initial positioning or coarse positioning over a large area.

[0060] Ultra-wideband module: By measuring the time-of-flight (TOF) signal between the module and an ultra-wideband module base station in the environment, it provides indoor ranging and positioning data with centimeter-level accuracy. It is suitable for scenarios requiring high-precision navigation and positioning, and can accurately dock and align exhibits.

[0061] Global Positioning System (GPS) module: Used to receive satellite signals in open outdoor environments, providing absolute geographic coordinates, which is the foundation for robots to achieve outdoor navigation and movement.

[0062] Visual synchronous localization and mapping technology: Utilizing visual data collected by high frame rate cameras, autonomous pose estimation and environmental map construction are achieved by tracking environmental feature points in complex environments with no external signal coverage or poor signal.

[0063] The execution unit is responsible for accurately translating the decision instructions issued by the upper-level software modules into the robot's physical actions.

[0064] In this embodiment, the execution unit includes:

[0065] Chassis drive mechanism: Used to enable the robot to move on a two-dimensional plane. Depending on the robot platform design, servo motor-driven wheel sets can be used to achieve efficient and smooth movement, or a multi-jointed leg mechanism can be used to adapt to more complex terrain.

[0066] Multi-degree-of-freedom robotic arms and dexterous hands: used to perform anthropomorphic gestures, postures, or complete specific physical interaction tasks.

[0067] Audio output system: Used for voice broadcasting and audio interaction. The audio output system preferably consists of a dual-channel digital amplifier and high-fidelity speakers of Harman brand or equivalent quality, ensuring sufficient clarity, loudness and coverage for voice broadcasting in large or noisy spaces, thereby effectively overcoming interference from environmental noise.

[0068] The communication unit is responsible for establishing a data link between the robot and the external network.

[0069] 5G Communication Module: This module mainly consists of a high-performance 5G communication module and a matching omnidirectional antenna. Compared to traditional Wi-Fi, 5G provides a wireless data link with higher bandwidth, lower latency, and wider coverage. This enables robots to perform real-time and stable data transmission in large exhibition centers, outdoor plazas, and other wide areas or complex network environments. It allows robots to interact with cloud-based AI services, receive remote update commands, and report operational status data.

[0070] Internal Network Services: To ensure efficient and seamless data flow between the core computing unit, the robot platform's original underlying control system, and other expansion modules, the core computing unit is named PC3, and the original computing units responsible for underlying motion control and sensor integration are named PC1 and PC2, respectively. A Network Address Translation (NAT) service is also configured within the hardware layer. Running on PC3 (the main brain), the NAT service converts public network connections obtained from 5G modules or external Wi-Fi into a unified internal LAN, assigning private IP addresses to PC1, PC2, and all other network-enabled expansion modules and providing them with network access. This centralized network architecture, with PC3 as the gateway, simplifies communication configuration between internal modules, enabling upper-layer software to seamlessly access and manage all hardware resources through standard network protocols.

[0071] The perception and data processing layer is logically closely connected to the hardware layer. Its core function is to convert the various raw and unstructured sensor data collected by the multimodal perception unit into structured feature data that can be directly understood and used by the upper-level decision-making module through a series of signal processing and pattern recognition algorithms.

[0072] Audio data processing: For the raw audio stream acquired from the microphone array, audio data processing first performs echo cancellation to precisely filter out speech or sound effects played by the speakers of the execution unit, thereby isolating pure, realistic ambient background sound. Subsequently, by performing a Fast Fourier Transform (FFT) spectral analysis on the processed audio signal, the energy distribution or average power within a specific frequency band is calculated, ultimately yielding a quantified, time-varying ambient noise level N. level,t ;

[0073] Meanwhile, for scenarios requiring voice interaction, the clean user voice stream, after echo cancellation and noise reduction processing, will be sent to a speech recognition engine for processing. The speech recognition engine, a lightweight model deployed locally on the core computing unit, in this preferred embodiment, can transmit the audio stream in real-time to a cloud-based speech recognition service for processing via a communication unit, thereby obtaining highly accurate text transcription results and providing input for subsequent natural language understanding and dialogue management.

[0074] Visual data processing: For the raw video stream captured from the camera, visual data processing, supported by the powerful computing capabilities of the neural network processing unit in the core computing unit, processes the video frames in real time. By running a pre-trained people detection model, visual data processing can efficiently identify and count the number of individuals in the image, and estimate the number of people per unit area by combining the camera's intrinsic and extrinsic parameters, thereby extracting the structured crowd density D. density,t Feature data. Simultaneously, by calculating the brightness histogram of each frame or the average grayscale value of all pixels, visual data processing can quantify the overall illumination intensity of the current scene, thereby extracting the ambient light level L. illum,t Feature data.

[0075] Localization data fusion: For various heterogeneous localization data collected from a multi-source fusion localization system, localization data fusion is responsible for efficient data fusion, outputting a unified, high-precision, and high-frequency robot pose estimate. Localization data fusion can run an extended Kalman filter algorithm. The extended Kalman filter algorithm takes noisy measurements from different sensors as input, and through a recursive prediction and update loop, continuously corrects the estimate of the robot's state vector, ultimately outputting a global pose information that is more continuous, smoother, and more reliable than any single data source.

[0076] The scene dynamic understanding and decision engine, connected to the perception and data processing layer, is the core decision-making hub for achieving environmental adaptation. Its core function is to receive multiple structured feature data output from the upper layer, and based on this structured feature data, to comprehensively evaluate and classify the robot's current scene, ultimately outputting a discrete and clear scene state.

[0077] The feature normalization and fusion scenario dynamic understanding and decision engine first normalizes the received multiple structured feature data. This step aims to eliminate the dimensional differences between different physical features, uniformly mapping them to an interval for subsequent fair fusion calculations. Taking environmental noise level as an example, the normalized value N' norm,t Calculated using the following formula:

[0078]

[0079] in:

[0080] N level,t This is the real-time environmental noise level measurement output by the perception and data processing layer at the current moment;

[0081] N min and N maxThese are the minimum and maximum noise levels, respectively, determined through prior surveys and calibrations of the system deployment environment. Crowd density and light levels are also processed using a similar minimum-to-maximum normalization method.

[0082] After normalization, the fusion scene dynamic understanding and decision engine calculates a comprehensive environmental assessment score C by weighted fusion of dimensionless feature values. env,t Comprehensive environmental assessment score C env,t As a key quantitative indicator, it intuitively reflects the overall state of the current environment. The calculation formula is as follows:

[0083] C env,t =w n ·N' norm,t +w d ·D' norm,t +w l ·L′ norm,t ;

[0084] in:

[0085] w n w d w l These are the preset weighting coefficients for environmental noise, crowd density, and illumination, respectively, and they satisfy w n +w d +w l =1. The weighting coefficients can be flexibly configured according to the emphasis of different application scenarios. For example, in a museum, w n The weight may be higher, and in the marketplace, w d The weight is higher.

[0086] The scene state machine, along with other state information acquired from the system, constitutes the scene state vector based on the comprehensive environmental assessment score. The integrated scene dynamic understanding and decision-making engine also includes a scene state machine. This machine receives the scene state vector as input and, according to a series of predefined logical rules and thresholds, maps the continuously changing scene state vector to a specific state in a predefined set of discrete scene states. When the comprehensive environmental assessment score C... env,t If the predefined logical rules and thresholds are exceeded, the state switches from a quiet exhibition hall to a noisy lobby. The use of a state machine can effectively avoid frequent and invalid switching of scene states caused by small fluctuations in evaluation scores around the threshold, thus ensuring the stability of decision-making.

[0087] Finally, the scene dynamic understanding and decision engine outputs the classified scene state as scene data and passes it to the next layer service strategy adaptive adjustment module.

[0088] The service strategy adaptive adjustment module connects with the scene dynamic understanding and decision engine, serving as a bridge between scene understanding and robot behavior. Its function is to retrieve and activate a matching service strategy from a pre-set service strategy library based on the scene data output by the engine, and then output the specific behavioral parameters contained in the service strategy to the application and interaction execution layer.

[0089] Service Policy Library: The service policy adaptive adjustment module contains a service policy library. This library can be implemented using a hash table or similar data structure, storing multiple service policies. Each service policy corresponds to a predefined scenario state and includes a set of numerical behavioral parameters to guide the robot's specific actions. In this embodiment, the behavioral parameters include at least:

[0090] Maximum movement speed v max The highest speed the robot is allowed to reach in the current scenario;

[0091] Safe obstacle avoidance distance d safe The minimum safe distance that a robot must maintain from obstacles;

[0092] Voice broadcast volume V spk The volume of the robot's voice broadcast.

[0093] The strategy corresponding to the quiet exhibition hall state is {v max =0.3m / s,d safe =1.0m,V spk =40}, while the strategy corresponding to the noisy lobby state might be {v max =0.8m / s,d safe =0.5m,V spk =85}.

[0094] For a smooth transition of the strategy, when the service strategy adaptive adjustment module receives scene data from the scene dynamic understanding and decision engine indicating a change in the current scene state, it does not immediately abruptly change all behavioral parameters to the new strategy's values. Instead, it initiates a strategy smooth transition processing mechanism. Within a preset time window ΔT, the service strategy adaptive adjustment module performs linear interpolation calculations on each behavioral parameter included in the two service strategies before and after the switch, thereby achieving a smooth and natural transition of robot behavior. At any time t (0 ≤ t ≤ ΔT) during the transition process, the currently effective behavioral parameter vector P of the system... current (t) is calculated using the following formula:

[0095]

[0096] in:

[0097] Pcurrent (t) is the current behavior parameter vector at time t after the start of the transition;

[0098] P old This is the old behavior parameter vector corresponding to the service policy before the switch;

[0099] P new The new behavior parameter vector corresponding to the newly switched service strategy;

[0100] P is a variable containing v max d safe V spk A vector of multiple behavioral parameters.

[0101] Cloud-based dynamic updates: The service policy adaptive adjustment module, in conjunction with the communication unit in the hardware layer, possesses remote update capabilities. It can receive update commands from the cloud server via the communication unit and dynamically update the contents of the locally stored service policy library accordingly. This can be achieved by modifying behavioral parameters corresponding to existing scenario states or adding new scenario states and corresponding service policies. This improves the system's maintainability and adaptability to new application scenarios.

[0102] The application and interaction execution layer connects with the service policy adaptive adjustment module and the hardware layer, forming the final link between upper-layer system decision-making and lower-layer physical execution. Its function is to receive service policies containing specific behavioral parameters output by the service policy adaptive adjustment module and parse these parameters into low-level control instructions for the execution units in the hardware layer.

[0103] Navigation behavior execution: For navigation-related behavior parameters, the application and interaction execution layer will receive the maximum movement speed v. max and safe obstacle avoidance distance d safe As dynamic constraints, these are input into the path planning and motion control algorithm module of the robot chassis. The path planning and motion control algorithm module can preferably use the dynamic window method, which, within each control cycle, combines the robot's kinematic model and dynamic constraints to sample and select an optimal path in the velocity space, thereby controlling the chassis drive mechanism in the execution unit to strictly adhere to the current velocity. max Upper limit and d safe Under constraints, the movement can be carried out safely and smoothly.

[0104] Voice behavior execution: For voice-related behavior parameters, the application and interaction execution layer will receive the voice broadcast volume V. spk This signal is converted into a gain control signal for the audio output system in the hardware layer, which directly adjusts the actual output power of the speaker, thereby controlling the execution unit to broadcast voice messages at an appropriate volume.

[0105] Other interactive behavior execution: In addition, if the service strategy includes gesture interaction parameters, the application and interaction execution layer will also parse them into target angles or motion trajectory instructions for each joint of the multi-degree-of-freedom robotic arm in the execution unit, and drive the execution unit to complete the specified anthropomorphic actions, such as waving or pointing, through the underlying motion controller.

[0106] Benefiting from the powerful computing capabilities of the core computing unit, especially the acceleration capabilities of the neural network processing unit for visual algorithms, this system is also able to achieve more advanced humanoid-specific interactive capabilities.

[0107] For interactive imitation capabilities: The application and interaction execution layer can receive and process skeletal keypoint sequence data about a specific interactive object from the perception and data processing layer. Through an inverse kinematics solution module, the guide robot system can convert observed user actions into motion commands for the robot's multi-degree-of-freedom robotic arm and body joints in real time, thereby achieving the imitation of user actions.

[0108] Multimodal task decomposition capability: For a complex instruction given in natural language, the task planner within the application and interaction execution layer can decompose it into a series of ordered subtasks and schedule different functional modules to execute collaboratively. By combining multimodal perception with complex behavior planning, robots can complete tasks with practical significance that go far beyond simple gestures.

Claims

1. A high-performance ARM64-NPU module-based tour guide robot system, Comprising, characterized in that; A hardware layer, comprising a core computing unit for providing core computing power, a multi-modal perception unit for collecting environmental and interaction data, and an execution unit for performing physical actions; A perception and data processing layer connected with the hardware layer, for processing raw data collected by the multi-modal perception unit, so as to extract structured feature data; A scene dynamic understanding and decision engine connected with the perception and data processing layer, for receiving the structured feature data, constructing a scene state vector based on the structured feature data, classifying the current scene into a predefined scene state, and outputting scene data; A service policy adaptive adjustment module connected with the scene dynamic understanding and decision engine, for matching and activating a corresponding service policy from a preset service policy library according to the scene data, the service policy including a set of behavior parameters for guiding robot behavior; An application and interaction execution layer connected with the service policy adaptive adjustment module and the hardware layer, for receiving the service policy and controlling the execution unit to perform corresponding actions according to the behavior parameters for guiding robot behavior. 2.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The structured feature data extracted by the perception and data processing layer includes: Ambient noise level processed by the audio system in the multi-modal perception unit; Crowd density and lighting conditions processed by the vision system. 3.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The scene dynamic understanding and decision engine calculates a comprehensive environmental evaluation score by normalizing and weightedly fusing multiple structured feature data output by the perception and data processing layer, thereby constructing the scene state vector. 4.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The scene dynamic understanding and decision engine further includes a scene state machine for classifying the current scene into the predefined scene state according to the scene state vector. 5.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The service policy library stores multiple service policies, which are a set of behavior parameters corresponding to specific scene data; The behavior parameters include maximum moving speed, safe obstacle avoidance distance, and broadcast volume. 6.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The process of the service policy adaptive adjustment module activating a corresponding service policy includes: When the scene data output by the scene dynamic understanding and decision engine switches, smooth transition processing is performed on the behavior parameters contained in the two service policies before and after the switch within a preset time window. 7.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The multi-modal perception unit includes: A microphone array for far-field voice pickup and sound source positioning; A camera for acquiring visual information; A multi-source fusion positioning system composed of a Bluetooth beacon, an ultra-wideband module, a GPS module, and a visual SLAM technology. 8.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The core computing unit of the hardware layer includes an ARM64 architecture processor and a neural network processing unit, which is used for visual information processing for the perception and data processing layer. 9.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The application and interaction execution layer is specifically configured to control the execution unit to move according to navigation-related behavior parameters in the service policy, and to control the execution unit to perform voice broadcast according to voice-related behavior parameters. 10.The tour guide robot system based on high-performance ARM64-NPU module according to claim 1, wherein, The hardware layer further includes a communication unit; The communication unit receives an update instruction from a cloud server through cooperation of the service policy adaptive adjustment module, thereby dynamically updating the content of the service policy library.

Images