Body-aware agent control method and apparatus, device, and medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses a body-equipped intelligent agent control method and device, equipment and a medium. The method comprises the following steps: acquiring detection data of a target user and motion parameters of a body-equipped intelligent agent; calculating a radial velocity component generated by the target user due to the motion of the body-equipped intelligent agent; obtaining the real radial velocity of the target user; when it is determined that the target user meets a preset approaching condition based on the real radial velocity and a target coordinate, driving the body-equipped intelligent agent to preliminarily turn around according to the target coordinate; in the process of performing the preliminary turning around, acquiring image frames collected by a visual sensor arranged in front of the body-equipped intelligent agent in real time, and when it is detected that a face feature of the target user exists in the image frames, driving the body-equipped intelligent agent to continue turning around according to the position deviation of the face feature in the image frames. According to the application, the body-equipped intelligent agent can reliably perceive the approaching intention of a target in a dynamic environment and realize accurate and smooth face-to-face posture alignment.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and medium for controlling an embodied intelligent agent. Background Technology

[0002] With the rapid development of artificial intelligence and robotics, embodied intelligent agents (such as mobile service robots and intelligent interactive devices) have gradually been integrated into daily life and public service scenarios. In practical applications, such embodied intelligent agents often rely on rear-mounted radar to detect and perceive users approaching from behind, and automatically adjust their orientation to establish a face-to-face interactive relationship.

[0003] However, existing solutions have significant shortcomings in actual operation. First, when the equipment itself is moving or turning, its own motion inevitably gets mixed in. Second, after triggering a turn, it often relies on a fixed angle or a fixed duration for rotation, making it difficult to adapt to changes in the target's actual location.

[0004] In summary, how to enable intelligent agents to reliably perceive the target state in dynamic environments and achieve accurate turning and alignment has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this application is to solve the above-mentioned problems by providing an embodied intelligent agent control method, device, equipment, and medium.

[0006] According to one aspect of this application, an embodied intelligent agent control method is provided, comprising the following steps: The system acquires detection data of the target user collected by a radar sensor configured behind the embodied intelligent agent, and motion parameters collected by a motion sensor configured on the embodied intelligent agent. The detection data includes target coordinates and initial radial velocity. Based on the motion parameters and target coordinates, calculate the radial velocity component of the target user caused by the motion of the embodied intelligent agent itself; The initial radial velocity is compensated based on the radial velocity component to obtain the true radial velocity of the target user; When it is determined that the target user meets the preset approach conditions based on the actual radial velocity and target coordinates, the embodied intelligent agent is driven to make an initial turn according to the target coordinates; During the initial turn, image frames are acquired in real time by a visual sensor positioned in front of the embodied intelligent agent. When the facial features of the target user are detected in the image frame, the embodied intelligent agent is driven to continue turning based on the positional deviation of the facial features in the image frame.

[0007] According to another aspect of this application, an apparatus for controlling an embodied intelligent agent is provided, comprising: The data acquisition module is used to acquire detection data of the target user collected by the radar sensor configured behind the embodied intelligent agent, and motion parameters collected by the motion sensor configured on the embodied intelligent agent. The detection data includes target coordinates and initial radial velocity. The self-velocity projection module is used to calculate the radial velocity component generated by the target user due to the movement of the embodied intelligent agent itself, based on the motion parameters and target coordinates. The speed compensation module is used to calculate the compensation of the initial radial velocity based on the radial velocity component to obtain the true radial velocity of the target user. The coarse-adjustment turning module is used to drive the embodied intelligent agent to make a preliminary turn according to the target coordinates when it is determined that the target user meets the preset approach conditions based on the real radial velocity and target coordinates. The visual fine-tuning module is used to acquire image frames collected by a visual sensor positioned in front of the embodied intelligent agent in real time during the initial turn. When the facial features of the target user are detected in the image frame, the embodied intelligent agent is driven to continue turning according to the positional deviation of the facial features in the image frame.

[0008] According to another aspect of this application, an electronic device is provided, including a central processing unit and a memory, wherein the central processing unit is configured to invoke and run a computer program stored in the memory to perform the steps of the embodied intelligent agent control method described in this application.

[0009] According to another aspect of this application, a non-volatile readable storage medium is provided, which stores a computer program implemented according to the embodied intelligent agent control method in the form of computer-readable instructions, wherein the computer program, when invoked by a computer, executes the steps included in the method.

[0010] Compared to existing technologies, this application calculates the radial velocity component based on motion parameters and target coordinates and compensates for the initial radial velocity to obtain the true radial velocity. Furthermore, during the initial turning process, it drives the turning to continue based on the positional deviation of facial features in the visual sensor image frame. This achieves the effect of eliminating the interference of the embodied intelligent agent's own motion to accurately determine the user's approach state and realizes the smooth connection and precise alignment of the turning process. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating one embodiment of the embodied intelligent agent control method of this application; Figure 2 This is a schematic block diagram of the embodied intelligent agent control device of this application; Figure 3 This is a schematic diagram of the structure of an electronic device used in this application. Detailed Implementation

[0012] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0013] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0014] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0015] Those skilled in the art will understand that the terms "client," "terminal," and "terminal device" as used herein include both devices that receive wireless signals, devices that only possess wireless signal receiver capabilities without transmission capabilities, and devices with receiving and transmitting hardware, devices that have receiving and transmitting hardware capable of bidirectional communication over a bidirectional communication link. Such devices may include: cellular or other communication devices such as personal computers or tablets, having single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service) that can combine voice, data processing, fax, and / or data communication capabilities; PDAs (Personal Digital Assistants) that may include radio frequency receivers, pagers, internet / intranet access, web browsers, notepads, calendars, and / or GPS (Global Positioning System) receivers; and conventional laptops and / or handheld computers or other devices that have and / or include radio frequency receivers. As used herein, "client," "terminal," and "terminal device" can be portable, transportable, installed in a means of transportation (air, sea, and / or land), or suitable and / or configured to operate locally and / or in a distributed manner, operating in any other location on Earth and / or in space. "Client," "terminal," and "terminal device" as used herein can also be a communication terminal, an internet access terminal, or a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or a smart TV, set-top box, etc.

[0016] The hardware referred to by the names "server," "client," and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann architecture, such as a central processing unit (including an arithmetic logic unit and a control unit), memory, input devices, and output devices. The computer program is stored in its memory, and the central processing unit loads the program stored in the secondary storage into the main memory to run it, execute the instructions in the program, and interact with the input and output devices to complete specific functions.

[0017] It should be noted that the concept of "server" used in this application can also be extended to apply to server clusters. Based on network deployment principles as understood by those skilled in the art, the servers should be logically divided; physically, these servers can be independent yet accessible through interfaces, or they can be integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method described in this application.

[0018] One or more of the technical features of this application, unless explicitly specified herein, can be deployed on a server and accessed by a client remotely calling the online service interface provided by the server, or can be directly deployed and run on a client for access.

[0019] Unless otherwise specified, all data involved in this application may be stored remotely on a server or on a local terminal device, as long as it is suitable for use by the technical solution of this application.

[0020] Those skilled in the art will understand that although the various methods in this application are described based on the same concept and thus present commonality among them, they can be performed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept; therefore, concepts expressed in the same way, as well as concepts that are appropriately changed for convenience but are expressed differently, should be understood equivalently.

[0021] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0022] To facilitate understanding of the various embodiments of this application, exemplary mechanical structures and application scenarios will be introduced first.

[0023] In an exemplary mechanical structure, the embodied intelligent agent control method of this application is implemented based on the physical layout of the embodied intelligent agent body, mobile chassis, steering mechanism, radar sensor, motion sensor and vision sensor.

[0024] The mobile chassis is positioned at the bottom of the embodied intelligent entity, contacting the ground via drive wheels or tracks. It is responsible for supporting the entire device and enabling autonomous movement and path switching within the space. A steering mechanism connects the mobile chassis to the embodied intelligent entity, mechanically coupled to the chassis via a motor drive component. This mechanism receives control commands and drives the embodied intelligent entity to adjust its horizontal orientation. Radar sensors are fixedly positioned at the rear of the embodied intelligent entity, motion sensors are mounted on its body, and vision sensors are fixedly positioned at the front.

[0025] The radar sensor, motion sensor, vision sensor, mobile chassis and steering mechanism are all directly mounted on the embodied intelligent agent and electrically connected to the control unit inside the embodied intelligent agent through internal signal harnesses.

[0026] In one exemplary application scenario, the embodied intelligent agent control method of this application can be applied to open service environments such as shopping mall navigation, hospital guidance, and smart exhibition halls. In such scenarios, the embodied intelligent agent is usually in an autonomous cruising or low-speed movement state, with frequent movement of people around it. Target users often approach from behind or to the side of the embodied intelligent agent to initiate voice interaction or business consultation requests.

[0027] This real-world environment presents clear business requirements for the dynamic perception and posture adjustment of embodied intelligent agents: On the one hand, during its own movement, the embodied intelligent agent needs to accurately distinguish between the target user's true intention to approach and the relative speed change caused by its own displacement, so as to avoid false or missed responses due to data confusion; on the other hand, when the embodied intelligent agent turns to face the target user, it needs to balance turning efficiency and operational stability, so as to quickly bring the target user into the foreground view, while avoiding alignment deviations or repeated swaying caused by abrupt fixed-angle rotations, so as to ensure the natural and coherent service interaction.

[0028] As the embodied agent moves within the passage, a radar sensor positioned behind it continuously collects detection data of the target user, acquiring the target's coordinates and initial radial velocity. Simultaneously, a motion sensor positioned on the embodied agent collects motion parameters. The embodied agent combines these motion parameters with the target coordinates to calculate the radial velocity component generated by its own movement and compensates for the initial radial velocity, thus eliminating interference from its own movement and obtaining the target user's true radial velocity. When the target user meets the preset approach conditions based on the true radial velocity and target coordinates, the embodied agent performs an initial turn according to the target coordinates, roughly facing the target user's location. During this initial turn, a visual sensor positioned in front of the embodied agent collects image frames in real time. When the target user's facial features are detected in the image frame, the embodied agent immediately drives itself to continue turning based on the positional deviation of these facial features in the image frame, gradually and precisely aligning the target user with the center of the interactive field of vision.

[0029] In summary, this application achieves seamless operation of dynamic interference removal, coarse orientation guidance, and real-time visual correction. This enables the embodied agent to reliably identify the target user's actual approach trend even while in motion, and achieves a smooth transition from coarse positioning to fine alignment during turning. This effectively avoids misjudgments caused by the motion interference of traditional embodied agents, as well as the abrupt movements and field-of-view shifts caused by turning with fixed parameters. It provides a feasible application path for efficient human-computer interaction in dynamic environments.

[0030] Through the detailed description of the mechanical structure and application scenarios described above, a better understanding of the specific application of the embodied intelligent agent control method of this application in the field of artificial intelligence technology can be achieved. The following will elaborate on the detailed description of various specific embodiments of this application based on these exemplary contents.

[0031] like Figure 1 As shown, in one embodiment, the embodied intelligent agent control method of this application includes: Step S5100: Obtain the detection data of the target user collected by the radar sensor configured behind the embodied intelligent agent, and the motion parameters collected by the motion sensor configured on the embodied intelligent agent. The detection data includes the target coordinates and the initial radial velocity.

[0032] The control unit receives detection data from the radar sensor located behind the embodied intelligent agent and simultaneously receives motion parameters from the motion sensor located on the embodied intelligent agent. The control unit extracts the target coordinates and initial radial velocity from the detection data and completes data reading and buffering operations.

[0033] The control unit, integrated within the embodied intelligent agent's body, is responsible for receiving raw data from various sensors, executing speed compensation and attitude control algorithms, and outputting drive commands to the steering mechanism. It serves as the central hub for the embodied intelligent agent's environmental perception and autonomous decision-making. Target coordinates characterize the spatial position of the target user relative to the embodied intelligent agent, including distance and azimuth parameters. Initial radial velocity represents the measured rate at which the target user approaches or moves away along the radar beam direction. Motion parameters quantify the real-time operating state of the embodied intelligent agent, including linear velocity, angular velocity, and acceleration information.

[0034] The control unit can establish a data link with the radar sensor and motion sensor through a preset communication protocol.

[0035] Radar sensors can transmit detection signals to the rear at fixed intervals and receive reflected echoes. After internal signal processing, they output detection data packets containing the target coordinates and initial radial velocity.

[0036] Motion sensors can collect the operating status of the chassis or body of the embodied intelligent agent in real time and convert it into a motion parameter data stream.

[0037] The control unit can synchronously read the detection data and motion parameters at a preset sampling frequency, and use a timestamp alignment mechanism to match the detection data and motion parameters, eliminating data timing deviations caused by transmission delays. After matching is complete, the control unit can write the data to a designated buffer in the running memory.

[0038] The control unit can perform data integrity verification. If data packet loss or value out of bounds occurs, it will mark the abnormal state and call the valid data of the previous frame for linear interpolation compensation to ensure that the input data stream is continuous and stable.

[0039] Radar sensors can be millimeter-wave radar or lidar.

[0040] Motion sensors can be inertial measurement units, wheel encoders, or optical odometers.

[0041] Step S5200: Based on motion parameters and target coordinates, calculate the radial velocity component generated by the target user's own motion due to the embodied intelligent agent.

[0042] The control unit receives motion parameters and target coordinates, performs velocity decomposition calculations, separates the projected velocity formed by the embodied intelligent agent's own movement in the direction pointing towards the target user, and outputs the radial velocity component.

[0043] The radial velocity component is used to quantify the degree of interference caused by the embodied agent's own motion to radar detection data. When the embodied agent is moving or turning, its displacement continuously changes the relative distance between it and the target user, causing additional values ​​from the device's own motion to be mixed into the initial radial velocity collected by the radar sensor.

[0044] The control unit can analyze motion parameters and extract the real-time running speed of the embodied intelligent agent.

[0045] The control unit can analyze the target coordinates and determine the spatial orientation of the target user relative to the embodied intelligent agent.

[0046] The control unit can decompose the real-time operating speed along the spatial pointing direction, calculate the projected value in that direction, and output this projected value as the radial velocity component. The calculation process follows direction alignment and numerical mapping rules to ensure that the decomposition direction is consistent with the orientation indicated by the target coordinates.

[0047] The calculation of the radial velocity component can be achieved through geometric decomposition, coordinate transformation, or trigonometric function mapping. The key is to accurately map the running velocity to the target orientation and output the equivalent projected value.

[0048] Step S5300: Compensate the initial radial velocity based on the radial velocity component to obtain the target user's true radial velocity.

[0049] The control unit receives the radial velocity component and the initial radial velocity, performs numerical compensation calculations to eliminate measurement deviations introduced by the embodied intelligent agent's own motion, and outputs the target user's true radial velocity.

[0050] True radial velocity is used to characterize the actual motion state of the target user relative to a stationary environment or a fixed reference frame.

[0051] The control unit can acquire the radial velocity component and the initial radial velocity, verify the directional reference and numerical dimensions of the two data, and ensure that they are in the same operational reference system.

[0052] The control unit can numerically superimpose or cancel the radial velocity component with the initial radial velocity, deduct the velocity influence generated by the device's self-motion in the detection direction, and output the target user's true radial velocity.

[0053] For numerical compensation calculations, interference removal can be achieved by algebraic subtraction, reference frame transformation, or superposition of signed values. As long as the interference values ​​can be removed under a unified benchmark and the actual movement speed of the target user can be output, it is acceptable.

[0054] Step S5400: When it is determined that the target user meets the preset approach conditions based on the actual radial velocity and target coordinates, drive the embodied intelligent agent to perform an initial turn according to the target coordinates.

[0055] The control unit determines the state of the actual radial velocity and the target coordinates. When it is confirmed that the target user meets the preset approach conditions, a turning control command is generated to drive the embodied intelligent agent to perform an initial turn according to the target coordinates.

[0056] The preset approach conditions are comprehensive judgment rules built into the control unit. These rules combine the approach trend reflected by the actual radial velocity with the spatial distance reflected by the target coordinates to filter out environmental noise or brief passing behavior and confirm that the target user has a stable intention to approach. The initial turn refers to a basic orientation adjustment action performed by the embodied intelligent agent based on the orientation indicated by the target coordinates. The purpose is to roughly turn the target user into the detection field of the forward visual sensor.

[0057] The control unit can input the real-time acquired radial velocity and target coordinates into the judgment process and match them with preset proximity conditions for verification. When the judgment result confirms that the conditions are met, the control unit can extract the orientation information contained in the target coordinates, convert it into the rotation direction parameters of the embodied intelligent agent, and generate the corresponding turning control command.

[0058] The turning control command can be sent to the driver of the embodied intelligent agent, controlling the embodied intelligent agent to perform a preliminary turning action according to the rotation direction parameters, so that the body turns towards the direction of the target user.

[0059] During the turning action, the control unit can continuously track the target coordinates to maintain the relative consistency between the turning direction and the target orientation, ensuring that the initial turning process advances synchronously with the target spatial position.

[0060] State determination and turn instruction generation can be achieved by threshold comparison, logic state machine polling, or software conditional branch judgment.

[0061] Step S5500: During the initial turn, image frames collected by the visual sensor configured in front of the embodied intelligent agent are acquired in real time. When the facial features of the target user are detected in the image frame, the embodied intelligent agent is driven to continue turning according to the positional deviation of the facial features in the image frame.

[0062] During the initial turn of the embodied intelligent agent, the control unit continuously receives image frames output by the front vision sensor and performs facial feature detection. When the facial features of the target user are identified in the image frame, the control unit extracts the positional deviation of the facial features in the image frame and generates a steering adjustment command, driving the embodied intelligent agent to continue turning according to the positional deviation.

[0063] An image frame refers to a unit of image data continuously output by a visual sensor in a time sequence. Facial features refer to the visual recognition area used to identify the shape and orientation of the target user's head. Positional deviation characterizes the relative offset between the actual display position of facial features in the image and the expected alignment position. Continued turning refers to the secondary angle correction performed based on image offset information after the initial turning to bring the target into the field of view.

[0064] After the initial turning motion is initiated, the control unit can activate the image data receiving channel and read the continuous image sequence output by the vision sensor at fixed intervals.

[0065] The control unit can input the read image sequence into the feature recognition process to perform facial region scanning and feature matching.

[0066] When the feature recognition process confirms the presence of the target user's facial features in the current image, the control unit can obtain the coordinate information of the facial features in the image plane and compare it with the reference position at the center of the image to calculate the position deviation data.

[0067] The control unit can convert position deviation data into angle adjustment amount of the embodied intelligent agent, generate corresponding steering adjustment commands, and drive the steering mechanism to perform a continued turning action.

[0068] If the feature recognition process fails to confirm the presence of the target user's facial features in the image frame within the preset monitoring period, the control unit can determine that the current visual channel has not captured a valid target. At this time, the control unit can maintain the initial turning execution state or switch to the preset field of view search mode to continuously receive and process new image frames until facial features are detected or the initial turning action reaches the preset termination condition.

[0069] During the continued turning process, the control unit can continuously read the image data and calculate the position deviation in real time, forming a dynamic feedback link until the position deviation meets the preset alignment requirements or the turning action reaches the termination state.

[0070] Facial feature detection can be achieved using convolutional neural network models, traditional image processing algorithms, or dedicated vision processing chips.

[0071] Positional deviations can be extracted using coordinate difference calculation, relative scale mapping, or grid region determination methods.

[0072] The state maintenance or timeout determination when no facial features are detected can be achieved by accumulating a time counter, counting consecutive empty frames, or accumulating angle increments.

[0073] As can be seen from the above embodiments, this application achieves the following outstanding effects: First, this application calculates the radial velocity component of the target user caused by the movement of the embodied intelligent agent by acquiring detection data collected by a radar sensor located behind the embodied intelligent agent and motion parameters collected by a motion sensor located on the embodied intelligent agent. Then, it compensates for the initial radial velocity based on this radial velocity component to obtain the target user's true radial velocity. This data processing method eliminates the numerical superposition interference caused by the embodied intelligent agent's own movement or turning on radar velocity measurement, accurately restoring the objective approach trend of the target user. It effectively overcomes the shortcomings of existing technologies that directly use the original velocity measurement while the device is moving, leading to false triggering and response delays. This allows the embodied intelligent agent to reliably perceive the target user's true approach state even in dynamic operation.

[0074] Secondly, when the target user meets the preset approach conditions based on the actual radial velocity and target coordinates, this application drives the embodied intelligent agent to perform an initial turn according to the target coordinates. During the initial turn, image frames collected by the visual sensor positioned in front of the embodied intelligent agent are acquired in real time. When the facial features of the target user are detected in the image frame, the embodied intelligent agent is driven to continue turning according to the positional deviation of the facial features in the image frame. This control strategy deeply integrates radar orientation guidance and visual image correction, breaking the rigid mode of traditional solutions that rely on fixed angles or fixed durations for turning. It can dynamically adjust according to the actual spatial position of the target user and the real-time offset of the image, effectively avoiding misalignment, loss of field of view, or repeated body swaying during the turning process, and achieving a smooth transition from coarse orientation alignment to fine image tracking.

[0075] Furthermore, the dynamic speed compensation mechanism and tiered turning strategy of this application form a complete closed loop during the operation of the embodied agent, seamlessly connecting interference elimination, approach intention determination, and real-time visual feedback. This collaborative control path improves the responsiveness and operational stability of the embodied agent in complex environments or under variable speed conditions, solving the technical bottleneck of existing agents' inability to accurately perceive the target state in dynamic environments and their stiff and sluggish turning and alignment. With coherent and adaptive attitude adjustment logic, it enables the embodied agent to reliably perceive the target state and accurately turn and align in dynamic environments.

[0076] Based on any embodiment of the method in this application, the radial velocity component generated by the target user's own motion due to the embodied intelligent agent is calculated based on motion parameters and target coordinates, including: Step S5210: Determine the installation offset and installation angle of the radar sensor relative to the body coordinate system of the embodied intelligent agent.

[0077] The control unit determines the installation offset and installation angle of the radar sensor relative to the body coordinate system of the embodied intelligent agent.

[0078] The offset index radar sensor is installed to measure the straight-line distance and direction parameters between its physical installation center point and the origin of the embodied intelligent agent's coordinate system. The angle index radar sensor is installed to detect the angle between the main axis and the reference axis of the embodied intelligent agent's coordinate system.

[0079] The control unit can access the pre-stored parameter storage area to read the spatial position offset and attitude angle values ​​of the radar sensor. The control unit can perform format verification on the read values, convert them into a unified data structure, and load them into the running memory.

[0080] If online calibration is supported, the control unit can receive external calibration commands, calculate the installation offset and installation angle by collecting spatial position information of known reference points, and complete the parameter update.

[0081] Step S5220: Based on the installation offset and installation angle, transform the target coordinates from the radar coordinate system to the body coordinate system.

[0082] The control unit transforms the target coordinates from the radar coordinate system to the body coordinate system based on the installation offset and installation angle.

[0083] The radar coordinate system is a local measurement reference system centered on the radar sensor itself, while the body coordinate system is a global control reference system based on the geometric center or motion reference point of the embodied intelligent agent.

[0084] The control unit receives the target coordinates, invokes spatial transformation processing logic, and performs translation compensation calculations on the target coordinates based on the installation offset value to eliminate the influence of the physical distance between the sensor origin and the fuselage origin. Subsequently, the control unit performs rotation transformation calculations on the translated coordinates based on the installation angle value to correct the angular difference between the sensor detection orientation and the fuselage reference orientation. After the calculation is completed, the control unit outputs the transformed target coordinates in the body coordinate system and aligns them with the motion parameters.

[0085] Step S5230: Calculate the vehicle's speed at the target coordinates based on the motion parameters and the transformed target coordinates.

[0086] The control unit calculates the vehicle's speed at the target coordinates based on motion parameters and the converted target coordinates.

[0087] The vehicle's motion speed refers to the composite movement rate formed by the superposition of the overall displacement and rotation of the embodied intelligent agent at the target user's spatial location during movement or turning.

[0088] The control unit can analyze motion parameters and extract the translational linear velocity and rotational angular velocity data of the embodied intelligent agent.

[0089] The control unit can read the converted target coordinates and determine the spatial vector of the target position relative to the fuselage reference point.

[0090] The control unit can calculate the equivalent velocity of the body moving linearly at the target point based on the translational linear velocity, and at the same time calculate the tangential velocity generated by the body rotating at the target point based on the spatial geometric relationship between the rotational angular velocity and the target space vector.

[0091] The control unit can perform vector synthesis of the translational equivalent velocity and the rotational tangential velocity to calculate the vehicle's speed at the target coordinates.

[0092] Step S5240: Project the vehicle's motion velocity onto the radial direction corresponding to the target coordinates to obtain the radial velocity component.

[0093] The control unit projects the vehicle's speed onto the radial direction corresponding to the target coordinates to obtain the radial speed component.

[0094] The radial direction refers to the spatial direction of the line connecting the embodied intelligent agent's reference point to the target user.

[0095] The control unit can construct a radial direction unit vector pointing from the embodied intelligent agent's body reference point to the target user based on the converted target coordinates.

[0096] The control unit can align the calculated vehicle speed with the radial unit vector and perform vector projection calculations.

[0097] The control unit can extract the parallel component value of the vehicle's speed in the radial direction, filter out irrelevant speed components perpendicular to the radial direction, and output the calculated parallel component value as the radial speed component.

[0098] As can be seen from the above embodiments, this application transforms the target coordinates from the radar coordinate system to the body coordinate system by determining the installation offset and installation angle of the radar sensor relative to the body coordinate system of the embodied intelligent agent. Based on the motion parameters and the transformed target coordinates, the vehicle's velocity at the target coordinates is calculated, and the vehicle's velocity is projected onto the radial direction corresponding to the target coordinates to obtain the radial velocity component. This achieves the effect of accurately quantifying and eliminating the interference of the device's own motion, and restoring the true motion state of the target. This calculation path unifies the spatial calculation benchmark, eliminates the numerical superposition error caused by differences in sensor physical installation and device movement, and directly extracts the specific interference values ​​generated in the target detection direction when the embodied intelligent agent moves or turns. This ensures that the collected speed data objectively reflects the actual approach trend of the target user, and ensures that the embodied intelligent agent can stably and accurately perceive the target's approach state in a dynamic operating state.

[0099] Based on any embodiment of the method in this application, the initial radial velocity is compensated according to the radial velocity component to obtain the true radial velocity of the target user, including: Step S5310: Determine the reference direction for measuring the initial radial velocity, and calculate the projection value of the radial velocity component on the reference direction to obtain the compensated velocity value.

[0100] The control unit determines the initial radial velocity measurement reference direction, calculates the projection value of the radial velocity component on the measurement reference direction, and outputs the compensated velocity value.

[0101] The velocity measurement reference direction refers to the direction of the detection beam axis used by the radar sensor when acquiring the initial radial velocity, i.e., the direction of the radar sensor's detection line. The compensated velocity value refers to the equivalent interference value generated in the radar velocity measurement direction after the radial velocity component is aligned with the direction.

[0102] The control unit can read the target coordinates, construct a spatial line from the radar sensor installation location to the target user, and determine the direction of this line as the speed measurement reference direction.

[0103] The control unit can acquire the radial velocity component and align it with the velocity measurement reference direction.

[0104] After alignment, the control unit can perform projection calculations, extract the parallel component values ​​of the radial velocity component in the velocity measurement reference direction, filter the velocity components perpendicular to this direction, and output the extracted parallel component values ​​as the compensation velocity values.

[0105] The velocity measurement reference direction can be obtained directly by calculating the target coordinate difference or by using the radar beam axis parameters. As long as the self-moving interference value can be accurately mapped to the original radar measurement direction and the projection value with the same direction can be output, it is acceptable.

[0106] Projection calculations can be performed using vector dot product operations, trigonometric function decomposition, or coordinate axis mapping. The projection calculation process can use the same units as the initial radial velocity, and the velocity measurement reference direction vector can be normalized to eliminate calculation errors caused by modulus differences, ensuring that the compensated velocity value reflects the actual interference intensity of the self-motion in the radar measurement direction.

[0107] Step S5320: Add or subtract the initial radial velocity from the compensated velocity value algebraically to obtain the true radial velocity.

[0108] The control unit algebraically adds or subtracts the initial radial velocity from the compensated velocity value to obtain the true radial velocity.

[0109] After receiving the initial radial velocity and the compensated velocity values, the control unit can verify the sign definition and direction reference of both.

[0110] The control unit can perform algebraic addition or subtraction based on the relative pointing directions of the initial radial velocity and the compensated velocity values ​​in the velocity measurement reference direction. When the two values ​​point in opposite directions, the control unit can perform subtraction to cancel out interference; when the two values ​​point in the same direction, the control unit can perform addition to correct superposition deviation. After completing the numerical calculation, the result can be marked as the true radial velocity and output.

[0111] As can be seen from the above embodiments, this application obtains a compensated velocity value by determining the initial radial velocity measurement reference direction and projecting the radial velocity component onto that direction. Then, the initial radial velocity and the compensated velocity value are algebraically added or subtracted, achieving the effect of accurately removing the device's own motion interference and restoring the target's true motion state. This compensation process uniformly maps the interference values ​​generated by the embodied intelligent agent's movement or turning onto the actual radar measurement axis. Through direct numerical cancellation calculations, it eliminates the self-motion deviation superimposed in the relative velocity measurement, ensuring that the output velocity data objectively reflects the target user's actual approach or departure trend. This ensures that the device can stably and accurately perceive the target's true approach state during dynamic movement or turning.

[0112] Based on any embodiment of the method in this application, before driving the embodied intelligent agent to make an initial turn according to the target coordinates when it is determined that the target user meets the preset approach conditions based on the actual radial velocity and target coordinates, the method includes: Step S5410: Determine whether the actual radial velocity and target coordinates meet the preset velocity threshold and distance threshold, respectively.

[0113] The control unit compares the actual radial velocity with a preset velocity threshold, and compares the distance value corresponding to the target coordinates with a preset distance threshold.

[0114] A preset speed threshold is used to define the effective range for determining the target's approach rate, and a preset distance threshold is used to define the safe space range for triggering interaction. Only when both the actual radial velocity and the target coordinates fall within the set range is the target determined to have initial approach characteristics.

[0115] The control unit can read the real-time calculated radial velocity value and compare it with a pre-stored velocity threshold. The control unit can also parse the distance parameter contained in the target coordinates and compare it with a pre-stored distance threshold. Based on the comparison results, it generates velocity and distance compliance status flags respectively. When both status flags indicate compliance, the control unit determines that the target has passed the initial screening.

[0116] Threshold comparison can be achieved through direct numerical comparison, interval range matching, or weighted scoring.

[0117] Threshold parameters can be set through system presets, manual user configuration, or environmental adaptive learning.

[0118] The preset speed threshold and distance threshold can be set according to the safe braking distance of the embodied intelligent agent and the effective detection range of the radar.

[0119] Step S5420: If satisfied, perform continuous frame detection counting on the target user, and when the continuous frame detection count reaches a preset frame number threshold, determine that the target user meets the preset proximity condition.

[0120] After the speed and distance comparisons are both passed, the control unit performs continuous frame detection counting. When the continuous frame detection count reaches the preset frame number threshold, it determines that the target user meets the preset proximity conditions.

[0121] Continuous frame detection counting refers to the process by which the control unit accumulates and records the status of targets that have passed the initial screening in multiple consecutive data acquisition cycles. The preset frame threshold refers to the minimum number of consecutive valid cycles required to confirm that the target has a stable intention to approach.

[0122] After receiving the trigger signal that the initial screening has passed, the control unit can start the continuous verification process and maintain a counter variable. In each data acquisition cycle, if the target user continuously meets the speed threshold and distance threshold requirements at the same time, the counter variable is incremented.

[0123] If the target fails to meet both threshold requirements simultaneously within a certain period, the control unit can clear the count variable or reset it to the initial value.

[0124] The control unit can compare the current count variable value with the preset frame number threshold in real time. When the count variable value accumulates to the preset frame number threshold, the control unit can generate a status confirmation command to determine that the target user meets the preset approach conditions.

[0125] The preset frame rate threshold can be set to correspond to three to ten data acquisition cycles. This ensures that the system has sufficient anti-interference and fault tolerance capabilities while avoiding excessively long counting cycles that could lead to delayed interactive responses.

[0126] As can be seen from the above embodiments, this application compares the actual radial velocity and target coordinates with preset velocity thresholds and distance thresholds respectively, and performs continuous frame detection and counting until the preset frame number threshold is reached when both conditions are met. This achieves a precise triggering effect that combines dual-dimensional initial screening with temporal continuity verification. This judgment mechanism simultaneously filters irrelevant targets or instantaneous interference from both spatial distance and approach rate perspectives, and eliminates misjudgments caused by personnel passing by briefly or sensor noise through continuous periodic accumulation counting logic. It ensures that subsequent actions are only formally activated when the target is continuously and stably approaching, effectively avoiding frequent false triggers and response delays in dynamic environments. This enables the embodied intelligent agent to stably and accurately identify the target's approach intention and reliably enter the turning preparation stage.

[0127] Based on any embodiment of the method in this application, driving the embodied intelligent agent to perform an initial turn according to the target coordinates includes: Step S5510: Determine the lateral orientation of the target user based on the target coordinates.

[0128] The control unit analyzes the target coordinates, extracts the angle information of the target user relative to the reference direction of the embodied intelligent agent, and determines the lateral orientation of the target user.

[0129] Lateral orientation refers to the area where the target user is located on the left or right side of the embodied intelligent agent's body, or at a specific horizontal deflection angle.

[0130] The control unit can read the orientation values ​​contained in the target coordinates, establish a horizontal reference system with the geometric center of the embodied intelligent agent's body as the origin, calculate the angle between the line connecting the target coordinates and the reference axis directly in front of the body, and determine whether the target user is located on the left or right side of the body based on the sign of the angle or the range of the angle.

[0131] Step S5520: Match the preset open-loop control parameters according to the lateral orientation. The open-loop control parameters include the turning angular velocity and the turning duration.

[0132] The control unit matches the corresponding open-loop control parameters from the preset parameter library based on the lateral orientation. The open-loop control parameters include the turning angular velocity and the turning duration.

[0133] Open-loop control parameters refer to drive values ​​that are preset according to fixed rules and do not depend on real-time closed-loop feedback.

[0134] After receiving the lateral orientation indicator, the control unit can access the built-in parameter mapping table or execute preset matching logic to extract the corresponding turning angular velocity value and turning duration value based on the left and right direction and degree of deflection indicated by the lateral orientation.

[0135] The control unit can encapsulate the turning angular velocity and turning duration in a format to generate open-loop control commands that include speed and time settings.

[0136] The open-loop control parameters can be calibrated based on the steering inertia and safe operating boundaries of the embodied intelligent agent. This configuration avoids the instability of the body due to excessive turning speed or the insufficient field of vision due to excessively short turning time, ensuring that the initial turning action is executed efficiently within a safe range.

[0137] Step S5530: Drive the embodied intelligent agent to perform an initial turn according to the open-loop control parameters in order to turn the target user into the detection field of the visual sensor.

[0138] The control unit sends open-loop control parameters, including the turning angular velocity and the turning duration, to the steering mechanism. The steering mechanism drives the embodied intelligent body to perform a horizontal deflection at a set speed and time, completing the initial turning action and bringing the target user into the detection field of the vision sensor.

[0139] The initial turn refers to the unidirectional angular deflection completed by the embodied intelligent entity within a set time, driven by the steering mechanism.

[0140] The control unit can send the packaged open-loop control parameters to the steering mechanism through the internal signal harness.

[0141] After receiving parameters, the steering mechanism can activate the motor drive components, controlling the embodied intelligent agent to begin horizontal rotation at a set turning angular velocity. The control unit or the timing module built into the steering mechanism can simultaneously start timing, accumulating the rotation duration in real time. When the accumulated time reaches the set turning duration, the control unit can send a stop command, the steering mechanism cuts off power output, and the embodied intelligent agent stops turning. The entire execution process operates according to preset parameters, ensuring the device completes a predetermined angle of orientation change, moving the target user, originally located behind or to the side, into the effective shooting range of the front visual sensor.

[0142] As can be seen from the above embodiments, this application determines the lateral orientation by analyzing the target coordinates, matches open-loop control parameters including turning angular velocity and turning duration, and drives the steering mechanism to perform an initial turn according to these parameters, achieving the effect of quickly and smoothly turning the target user into the forward visual detection range. This control strategy directly converts the target spatial orientation into standardized rotational speed and time commands, effectively overcoming the defects of misalignment and loss of field of vision caused by fixed angles or blind turns. This allows the embodied intelligent agent to complete the orientation switch according to the actual position of the target, achieving a stable and reliable attitude transition without the need for real-time video intervention.

[0143] Based on any embodiment of the method in this application, driving the embodied intelligent agent to continue turning around according to the positional deviation of facial features in the image frame includes: Step S5610: Extract the center point of the face detection box in the image frame and calculate the pixel deviation between the center point of the face detection box and the center point of the image frame.

[0144] The control unit extracts the center coordinates of the detection box of the target user's facial features in the image frame, calculates the difference between the center coordinates and the center coordinates of the image frame, and outputs the pixel deviation.

[0145] A face detection bounding box refers to the rectangular region of the target user's face identified and marked by image processing algorithms. The center point of the face detection bounding box and the center point of the image frame represent the visual centroid of the target face and the geometric center of the camera's view, respectively. Pixel deviation is the difference in distance between the two in the horizontal and vertical directions, used to characterize the alignment error between the current orientation of the camera and the frontal orientation of the target.

[0146] After receiving the image frame output by the vision sensor, the control unit can run a face region detection algorithm to locate a rectangular region containing the facial features of the target user within the image frame, read the coordinates of the four corner points of the rectangular region, calculate its geometric center point, and obtain the coordinates of the center point of the face detection box.

[0147] The control unit can obtain the resolution parameters of the image frame, calculate the coordinates of the geometric center point of the image frame, and calculate the difference between the coordinates of the center point of the face detection box and the coordinates of the center point of the image frame to obtain the pixel difference in the horizontal direction and the pixel difference in the vertical direction, and then output them as the pixel deviation.

[0148] Center point extraction can be achieved through centroid calculation, corner mean calculation, or feature point aggregation. Pixel deviation calculation can be achieved through two-dimensional vector subtraction or projected coordinate difference. The goal is to accurately extract the center position of the facial recognition region and calculate its relative offset from the image center.

[0149] Step S5620: Update the pixel deviation to the preset historical state queue, and calculate the deviation change rate of the pixel deviation based on the historical state queue.

[0150] The control unit stores the pixel deviation amount into a preset historical state queue, extracts multiple historical pixel deviation values ​​stored in the queue, calculates the rate of change between adjacent values, and outputs the deviation change rate.

[0151] The historical state queue is a circular buffer that stores pixel deviation data from recent multiple frames in chronological order. The deviation change rate refers to the rate at which pixel deviation increases or decreases over time, used to characterize how fast and in what direction a target moves or deviates from the frame.

[0152] After receiving the pixel deviation calculated for the current frame, the control unit can append it to the end of the historical state queue and maintain a constant queue length.

[0153] The control unit can read the historical values ​​of multiple pixel deviations arranged in a time sequence in the queue, perform differentiation or difference operations on the historical value sequence, extract the difference in value changes between adjacent data points, and calculate the change amplitude per unit time in combination with the data acquisition cycle to obtain the deviation change rate.

[0154] Step S5630: Using pixel deviation and deviation change rate as input parameters, call the preset nonlinear mapping model to output dynamic proportional gain.

[0155] The control unit inputs the pixel deviation and deviation change rate into a preset nonlinear mapping model, and outputs a dynamic proportional gain after the model is calculated.

[0156] Dynamic proportional gain refers to the adjustment sensitivity coefficient calculated in real time based on the input state.

[0157] The control unit can encapsulate the pixel deviation value and the deviation change rate value into an input vector and transmit it to the nonlinear mapping model.

[0158] Nonlinear mapping models can perform nonlinear combination operations on two input dimensions based on built-in mapping functions or training parameters. When the input deviation is large or the rate of change is high, the output proportional gain value is higher; when the input deviation is small or the rate of change is low, the output proportional gain value is lower.

[0159] The control unit can receive the numerical values ​​output by the nonlinear mapping model and label them as dynamic proportional gain.

[0160] Nonlinear mapping models can be implemented using piecewise linear functions, fuzzy logic rule tables, BP neural networks, or multinomial regression models. The key is to accept the inputs of the deviation and rate of change, and output the proportional coefficient that changes with the state.

[0161] The internal parameters of the nonlinear mapping model can be calibrated based on the steering response characteristics of the embodied agent and the field of view of the visual sensor.

[0162] Step S5640: Multiply the pixel deviation by the dynamic scaling gain to obtain the target adjustment angular velocity, and drive the embodied agent to continue turning according to the target adjustment angular velocity.

[0163] The control unit multiplies the pixel deviation by the dynamic proportional gain to calculate the target adjustment angular velocity, and generates control commands to drive the embodied intelligent agent to continue turning according to the target adjustment angular velocity.

[0164] The target adjustment angular velocity refers to the real-time rotation rate required to control the fuselage for corrective rotation.

[0165] The control unit can read the current pixel deviation value and dynamic proportional gain value, and multiply the two to obtain the target adjustment angular velocity value.

[0166] The control unit can convert this angular velocity value into a drive signal that the steering mechanism can execute and send it to the steering mechanism.

[0167] After receiving the drive signal, the steering mechanism can control the embodied intelligent agent to rotate at a target angular velocity. During the rotation, the control unit can continuously monitor the changes on the screen and repeat steps S5610 to S5640 to form a closed-loop control link.

[0168] As can be seen from the above embodiments, this application extracts the pixel deviation of facial features in the image frame, calculates the deviation change rate by combining it with the historical state queue, calls the nonlinear mapping model to output the dynamic proportional gain, and multiplies the pixel deviation by the dynamic proportional gain to obtain the target adjustment angular velocity to drive continued turning, achieving the effect of adaptive smooth fine-tuning based on the deviation trend of the visual image. This adjustment mechanism breaks through the limitations of fixed control parameters, and changes the body rotation sensitivity in real time according to the actual degree of target offset in the image and the speed of offset change. This allows the embodied intelligent agent to flexibly respond to target position changes and its own rotational inertia during tracking, effectively eliminating repeated swaying, excessive deflection, or response lag when continuing to turn. It achieves a smooth transition from coarse orientation alignment to fine visual alignment, ensuring that the embodied intelligent agent can accurately face the target user with a coherent and flexible posture in a dynamic interactive environment.

[0169] Based on any embodiment of the method in this application, the target adjustment angular velocity is obtained by multiplying the pixel deviation by the dynamic scaling gain, including: Step S5710: Multiply the pixel deviation by the dynamic scaling gain to obtain the initial adjustment angular velocity.

[0170] The control unit performs a multiplication operation between the pixel deviation and the dynamic proportional gain to calculate the initial adjustment angular velocity.

[0171] The initial adjustment angular velocity refers to the theoretical rotational rate calculated solely based on the current degree of deviation and control weights.

[0172] The control unit can read the pixel deviation value calculated from the current image frame and obtain the dynamic scaling gain value. The control unit can input the two values ​​into the multiplication unit to perform a multiplication operation and obtain the initial adjustment angular velocity.

[0173] Step S5720: Limit the initial adjustment angular velocity within the preset upper and lower limits of the angular velocity to obtain the amplitude-limited adjustment angular velocity.

[0174] The control unit constrains the initial adjustment angular velocity between the preset upper and lower limits of the angular velocity and outputs the limited adjustment angular velocity.

[0175] The preset upper and lower limits of angular velocity refer to the allowable rotational speed range set according to the mechanical structure strength, chassis drive capability, and safety specifications of the embodied intelligent agent.

[0176] After receiving the initial adjustment angular velocity value, the control unit compares it with the system's pre-stored upper and lower limits for angular velocity. If the initial adjustment angular velocity value is greater than the upper limit, the control unit replaces it with the upper limit; if the value is less than the lower limit, the control unit replaces it with the lower limit; if the value is between the upper and lower limits, it remains unchanged. After processing, the control unit outputs the limited adjustment angular velocity after boundary correction.

[0177] Limiting can be achieved using hard threshold truncation, soft saturation functions, or dynamic safety boundary mapping. Upper and lower limit parameters can be statically calibrated at the factory or dynamically adjusted based on the device's real-time load status. The goal is simply to constrain the calculation speed within a preset safety range.

[0178] The upper and lower limits of angular velocity can be set based on the rated peak speed of the steering motor of the embodied intelligent agent and the test data of the stability of the body's center of gravity.

[0179] Step S5730: Perform a first-order low-pass filter on the limiting adjustment angular velocity to obtain the target adjustment angular velocity.

[0180] The control unit performs a first-order low-pass filter operation on the limited adjustment angular velocity to suppress high-frequency numerical fluctuations and outputs the target adjustment angular velocity.

[0181] First-order low-pass filtering refers to using a mathematical smoothing algorithm with inertial attenuation characteristics to perform a gradual transition on the input velocity sequence. The target adjustment angular velocity refers to the actual rotational speed used for final transmission to the steering mechanism after smoothing.

[0182] The control unit can input the limiting adjustment angular velocity into the filtering processing unit.

[0183] The filtering unit can combine the historical angular velocity output from the previous calculation cycle with the current input limiting adjustment angular velocity, perform a weighted average calculation according to the preset filtering time constant, and use the weighted calculation result as the target adjustment angular velocity output for the current cycle.

[0184] First-order low-pass filtering can be achieved using exponential smoothing algorithms, moving average calculations, or equivalent analog RC filtering.

[0185] The filtering time constant can be configured with fixed parameters or adaptively adjusted according to the current operating posture of the embodied agent.

[0186] The filtering time constant can be set to an interval that matches the mechanical response delay of the embodied agent's steering.

[0187] As can be seen from the above embodiments, this application obtains the initial adjustment angular velocity by multiplying the pixel deviation by the dynamic proportional gain, limits the initial adjustment angular velocity to a preset upper and lower limit range to obtain a limited adjustment angular velocity, and performs a first-order low-pass filtering on the limited adjustment angular velocity to obtain the target adjustment angular velocity, thus achieving the effect of outputting a continuous and smooth turning speed command within a safe boundary. This signal optimization path, when converting visual position errors into body rotation commands, prevents speed overruns and loss of control through numerical boundary constraints, and absorbs image recognition jitter and calculation jumps through filtering algorithms, effectively eliminating mechanical stuttering, excessive deflection, or repeated left-right swaying caused by sudden command changes during continued turning. This makes the posture adjustment movements of the embodied intelligent agent smooth and continuous, achieving a balance between high response sensitivity and operational stability during dynamic alignment.

[0188] like Figure 2 As shown, an embodied intelligent agent control device according to one aspect of this application includes a data acquisition module 5100, a self-velocity projection module 5200, a speed compensation module 5300, a coarse adjustment turning module 5400, and a visual fine adjustment module 5500. The data acquisition module 5100 is used to acquire detection data of a target user collected by a radar sensor configured behind the embodied intelligent agent, and motion parameters collected by a motion sensor configured on the embodied intelligent agent. The detection data includes target coordinates and an initial radial velocity. The self-velocity projection module 5200 is used to calculate the radial velocity component of the target user caused by the movement of the embodied intelligent agent itself, based on the motion parameters and the target coordinates. The speed... The compensation module 5300 is used to perform compensation calculations on the initial radial velocity based on the radial velocity component to obtain the true radial velocity of the target user; the coarse adjustment turning module 5400 is used to drive the embodied intelligent agent to perform an initial turning based on the target coordinates when it is determined that the target user meets the preset approach conditions based on the true radial velocity and target coordinates; the visual fine adjustment module 5500 is used to acquire image frames collected by a visual sensor configured in front of the embodied intelligent agent in real time during the initial turning process, and when the facial features of the target user are detected in the image frame, drive the embodied intelligent agent to continue turning based on the positional deviation of the facial features in the image frame.

[0189] Based on any embodiment of the device in this application, the self-velocity projection module 5200 includes: The installation parameter determination module is used to determine the installation offset and installation angle of the radar sensor relative to the body coordinate system of the embodied intelligent agent; The coordinate transformation module is used to transform the target coordinates from the radar coordinate system to the body coordinate system according to the installation offset and installation angle. The vehicle speed calculation module is used to calculate the vehicle speed at the target coordinates based on the motion parameters and the transformed target coordinates. The radial projection module is used to project the vehicle's motion speed onto the radial direction corresponding to the target coordinates to obtain the radial velocity component.

[0190] Based on any embodiment of the device in this application, the speed compensation module 5300 includes: The compensation value calculation module is used to determine the velocity measurement reference direction of the initial radial velocity and calculate the projection value of the radial velocity component on the velocity measurement reference direction to obtain the compensation velocity value. The speed correction module is used to algebraically add or subtract the initial radial speed from the compensated speed value to obtain the true radial speed.

[0191] Based on any embodiment of the device in this application, prior to the coarse adjustment turning module 5400, the device includes: The threshold determination module is used to determine whether the actual radial velocity and the target coordinates meet preset velocity thresholds and distance thresholds, respectively. The continuous frame calculation module is used to perform continuous frame detection and counting on the target user if the condition is met, and to determine that the target user meets the preset proximity condition when the continuous frame detection count reaches a preset frame number threshold.

[0192] Based on any embodiment of the device in this application, the coarse adjustment turning module 5400 includes: The lateral orientation determination module is used to determine the lateral orientation of the target user based on the target coordinates. An open-loop parameter matching module is used to match preset open-loop control parameters according to the lateral orientation, wherein the open-loop control parameters include turning angular velocity and turning duration; The coarse adjustment execution module is used to drive the embodied intelligent agent to make an initial turn according to the open-loop control parameters, so as to turn the target user into the detection field of the visual sensor.

[0193] Based on any embodiment of the device in this application, the visual fine-tuning module 5500 includes: The pixel deviation calculation module is used to extract the center point of the face detection box in the image frame of the facial features, and calculate the pixel deviation between the center point of the face detection box and the center point of the image frame. The deviation change rate module is used to update the pixel deviation amount to a preset historical state queue, and calculate the deviation change rate of the pixel deviation amount based on the historical state queue. The dynamic gain mapping module is used to take the pixel deviation and the deviation change rate as input parameters, call a preset nonlinear mapping model, and output a dynamic proportional gain. An angular velocity generation module is used to multiply the pixel deviation by the dynamic proportional gain to obtain the target adjustment angular velocity, and drive the embodied intelligent agent to continue turning according to the target adjustment angular velocity.

[0194] Based on any embodiment of the device in this application, the angular velocity generation module includes: An angular velocity adjustment module is used to multiply the pixel deviation by the dynamic proportional gain to obtain an initial adjusted angular velocity; An angular velocity upper and lower limit module is used to limit the initial adjusted angular velocity within a preset range of upper and lower limits to obtain a limited adjusted angular velocity. A low-pass filter module is used to perform first-order low-pass filtering on the limiting adjustment angular velocity to obtain the target adjustment angular velocity.

[0195] like Figure 3 As shown in the diagram, another embodiment of this application also provides an electronic device, wherein the internal structure of the electronic device is illustrated. The electronic device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When the computer-readable instructions are executed by the processor, the processor can implement the embodied intelligent agent control method described in this application. The processor of the electronic device provides computing and control capabilities to support the operation of the entire electronic device. The memory of the electronic device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the embodied intelligent agent control method described in this application. The network interface of the electronic device is used for communication with a terminal. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0196] In this execution method, the processor is used for execution. Figure 2 The system contains the specific functions of each module and its submodules, and the memory stores the program code and various data required to execute these modules or submodules. The network interface is used for data transmission between the user terminal and the server. In this execution method, the memory stores the program code and data required to execute all modules / submodules in the embodied intelligent agent control device of this application, and the server can call the server's program code and data to execute the functions of all submodules.

[0197] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the method described in any embodiment of this application.

[0198] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM).

[0199] The above description is only a partial implementation method of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for controlling an embodied intelligent agent, characterized in that, include: The system acquires detection data of the target user collected by a radar sensor configured behind the embodied intelligent agent, and motion parameters collected by a motion sensor configured on the embodied intelligent agent. The detection data includes target coordinates and initial radial velocity. Based on the motion parameters and target coordinates, calculate the radial velocity component of the target user caused by the motion of the embodied intelligent agent itself; The initial radial velocity is compensated based on the radial velocity component to obtain the true radial velocity of the target user; When it is determined that the target user meets the preset approach conditions based on the actual radial velocity and target coordinates, the embodied intelligent agent is driven to make an initial turn according to the target coordinates; During the initial turn, image frames are acquired in real time by a visual sensor positioned in front of the embodied intelligent agent. When the facial features of the target user are detected in the image frame, the embodied intelligent agent is driven to continue turning based on the positional deviation of the facial features in the image frame.

2. The embodied intelligent agent control method according to claim 1, characterized in that, The step of calculating the radial velocity component of the target user due to the motion of the embodied intelligent agent itself, based on the motion parameters and target coordinates, includes: Determine the installation offset and installation angle of the radar sensor relative to the body coordinate system of the embodied intelligent agent; Based on the installation offset and installation angle, the target coordinates are transformed from the radar coordinate system to the body coordinate system; Based on the motion parameters and the transformed target coordinates, calculate the vehicle's velocity at the target coordinates; The vehicle's velocity is projected onto the radial direction corresponding to the target coordinates to obtain the radial velocity component.

3. The embodied intelligent agent control method according to claim 1, characterized in that, The step of calculating the compensation for the initial radial velocity based on the radial velocity component to obtain the true radial velocity of the target user includes: Determine the velocity measurement reference direction of the initial radial velocity, and calculate the projection value of the radial velocity component on the velocity measurement reference direction to obtain the compensated velocity value; The true radial velocity is obtained by algebraically adding or subtracting the initial radial velocity from the compensated velocity value.

4. The embodied intelligent agent control method according to claim 1, characterized in that, Before driving the embodied intelligent agent to make an initial turn based on the target coordinates when it is determined that the target user meets the preset approach conditions based on the true radial velocity and target coordinates, the process includes: Determine whether the actual radial velocity and the target coordinates satisfy preset velocity thresholds and distance thresholds, respectively; If the condition is met, then continuous frame detection counting is performed on the target user, and when the continuous frame detection count reaches a preset frame number threshold, it is determined that the target user meets the preset proximity condition.

5. The embodied intelligent agent control method according to claim 1, characterized in that, The process of driving the embodied intelligent agent to make an initial turn based on the target coordinates includes: Based on the target coordinates, determine the lateral orientation of the target user; According to the lateral orientation, preset open-loop control parameters are matched, including the turning angular velocity and the turning duration; The embodied intelligent agent is driven to make an initial turn according to the open-loop control parameters, so as to turn the target user into the detection field of the visual sensor.

6. The embodied intelligent agent control method according to claim 1, characterized in that, The process of driving the embodied intelligent agent to continue turning based on the positional deviation of the facial features in the image frame includes: Extract the center point of the face detection box in the image frame from the facial features, and calculate the pixel deviation between the center point of the face detection box and the center point of the image frame; The pixel deviation is updated to a preset historical state queue, and the deviation change rate of the pixel deviation is calculated based on the historical state queue. Using the pixel deviation and the deviation change rate as input parameters, a preset nonlinear mapping model is invoked to output a dynamic proportional gain. The target adjustment angular velocity is obtained by multiplying the pixel deviation by the dynamic proportional gain, and the embodied intelligent agent is driven to continue turning according to the target adjustment angular velocity.

7. The embodied intelligent agent control method according to claim 6, characterized in that, The step of multiplying the pixel deviation by the dynamic scaling gain to obtain the target adjustment angular velocity includes: Multiply the pixel deviation by the dynamic scaling gain to obtain the initial adjustment angular velocity; The initial adjustment angular velocity is limited to a preset upper and lower limit range to obtain a limited adjustment angular velocity; The target adjustment angular velocity is obtained by performing a first-order low-pass filter on the limiting adjustment angular velocity.

8. A control device for an embodied intelligent agent, characterized in that, include: The data acquisition module is used to acquire detection data of the target user collected by the radar sensor configured behind the embodied intelligent agent, and motion parameters collected by the motion sensor configured on the embodied intelligent agent. The detection data includes target coordinates and initial radial velocity. The self-velocity projection module is used to calculate the radial velocity component generated by the target user due to the movement of the embodied intelligent agent itself, based on the motion parameters and target coordinates. The speed compensation module is used to calculate the compensation of the initial radial velocity based on the radial velocity component to obtain the true radial velocity of the target user; The coarse-adjustment turning module is used to drive the embodied intelligent agent to make a preliminary turn according to the target coordinates when it is determined that the target user meets the preset approach conditions based on the real radial velocity and target coordinates. The visual fine-tuning module is used to acquire image frames collected by a visual sensor positioned in front of the embodied intelligent agent in real time during the initial turn. When the facial features of the target user are detected in the image frame, the embodied intelligent agent is driven to continue turning according to the positional deviation of the facial features in the image frame.

9. An electronic device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A non-volatile storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images