Method for controlling a robotic pet, robotic pet, and storage medium

Abstract

The embodiment of the application provides a kind of machine pet control method, machine pet and storage medium, method first obtains the distance information between user and machine pet, the orientation information of user relative to machine pet, the micro-motion information of user, wherein micro-motion information can represent the subtle body movement of user non-displacement or the signal change caused by physiological activity;Then according to the information obtained, construct human dynamic trend feature information;Subsequently, the interactive state of user is identified according to feature information, and the target behavior information of machine pet is determined according to interactive state, and the interactive behavior of machine pet is controlled according to target behavior information.The embodiment of the application identifies the interactive state of user by constructing human dynamic trend feature information, determines the behavior information of machine pet according to interactive state, and controls machine pet according to behavior information, so as to realize the natural behavior interaction of machine pet with user in various scenes, and improve user experience.

Description

Technical Field

[0001] The embodiments of this application relate to, but are not limited to, the field of information processing technology, and in particular to a control method for a robotic pet, a robotic pet, and a storage medium. Background Technology

[0002] Current consumer-grade robotic pets mostly rely on cameras, voice recognition modules, and distance sensors to achieve human-computer interaction. However, they have technical limitations in complex environments. For example, visual sensors are easily affected by lighting, and voice recognition is easily affected by environmental interference, making it difficult to achieve natural behavioral interaction with users. Furthermore, they may disturb users when they are in a low-interaction state, and they cannot respond in a timely manner when users actively interact. Summary of the Invention

[0003] This application provides a control method for a robotic pet, a robotic pet, and a storage medium, which enables the robotic pet to interact naturally with the user in various scenarios, thereby improving the user experience.

[0004] This application provides a method for controlling a robotic pet, comprising: acquiring distance information between a user and the robotic pet, orientation information of the user relative to the robotic pet, and micro-motion information of the user, wherein the micro-motion information represents the user's non-displacement subtle limb movements or signal changes caused by physiological activities; constructing human dynamic trend feature information based on the distance information, the orientation information, and the micro-motion information; identifying the user's interaction state based on the human dynamic trend feature information, and determining the target behavior information of the robotic pet based on the interaction state; and controlling the interactive behavior of the robotic pet based on the target behavior information.

[0005] On the other hand, embodiments of this application provide a robotic pet, including: at least one processor; at least one memory for storing at least one program; and when at least one of the programs is executed by at least one of the processors, implementing the robotic pet control method as described above.

[0006] On the other hand, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions for performing the control method for the robotic pet as described above.

[0007] On the other hand, embodiments of this application provide a computer program product, including a computer program or computer instructions, the computer program or computer instructions being stored in a computer-readable storage medium, the processor of a robotic pet reading the computer program or computer instructions from the computer-readable storage medium, and the processor executing the computer program or computer instructions to cause the robotic pet to perform the robotic pet control method described above.

[0008] This application provides a method for controlling a robotic pet, a robotic pet, a computer-readable storage medium, and a computer program product. The method first acquires distance information between the user and the robotic pet, the user's orientation relative to the robotic pet, and the user's micro-motion information. The micro-motion information can characterize the user's non-displacement, subtle limb movements or signal changes caused by physiological activities. Then, based on the distance, orientation, and micro-motion information, human dynamic trend feature information is constructed. Subsequently, the user's interaction state is identified based on the human dynamic trend feature information, and the target behavior information of the robotic pet is determined based on the interaction state. Finally, the interactive behavior of the robotic pet is controlled based on the target behavior information. This application, by constructing human dynamic trend feature information to identify the user's interaction state and determining the target behavior information of the robotic pet based on the interaction state, and controlling the robotic pet based on the determined behavior information, enables the robotic pet to interact naturally with the user in various scenarios, improving the user experience. Attached Figure Description

[0009] Figure 1 This is a flowchart of a control method for a robotic pet provided in one embodiment of this application; Figure 2 This is provided in one embodiment of the present application. Figure 1 The detailed flowchart of step S120; Figure 3 This is a flowchart of a control method for a robotic pet provided in another embodiment of this application; Figure 4 This is a block diagram of an intelligent interaction system for a robotic pet provided in one embodiment of this application; Figure 5 This is a block diagram of a millimeter-wave sensing module provided in one embodiment of this application; Figure 6 This is a block diagram of a human dynamic trend analysis module provided in one embodiment of this application; Figure 7 This is a block diagram of a pet social distancing decision module provided in one embodiment of this application; Figure 8 This is a flowchart of the pet social distancing decision module provided in one embodiment of this application; Figure 9 This is a block diagram of a social strategy learning and reconstruction module provided in one embodiment of this application; Figure 10 This is a flowchart of the social strategy learning and reconstruction module provided in one embodiment of this application; Figure 11 This is a block diagram of a motion control execution module provided in one embodiment of this application; Figure 12This is a block diagram of an exception handling adjustment module provided in one embodiment of this application; Figure 13 This is a flowchart of the intelligent interaction system for a robotic pet provided in one embodiment of this application; Figure 14 This is a schematic diagram of a robotic pet structure provided in one embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0011] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0012] In this application, the terms "furthermore," "in this embodiment," or "optionally" are used as examples, illustrations, or descriptions and should not be construed as being more preferred or advantageous than other embodiments or designs. The use of terms such as "furthermore," "in this embodiment," or "optionally" is intended to present the relevant concepts in a specific manner.

[0013] Current consumer-grade robotic pets primarily rely on multimodal detection, including cameras, voice recognition modules, and distance sensors, to achieve human-computer interaction. They use multiple camera modules to detect the user's position and follow them, trigger preset actions via voice commands, and use infrared ranging for obstacle avoidance and distance adjustment. However, when users are in a low-interaction state, such as focused work or rest, traditional robotic pets may still continuously approach, make sounds, or follow, disturbing the user. Conversely, when users actively express a desire to interact, traditional robotic pets cannot proactively approach in a timely manner, resulting in a mechanical interaction.

[0014] These solutions also suffer from significant technical limitations in complex environments: visual sensors are susceptible to light intensity, leading to inaccurate positioning and deviations in target action and facial expression recognition; the accuracy of speech recognition is also easily affected by the clarity of the user's speech and noisy environments; Bluetooth and WiFi technologies can detect the presence of a human body but cannot capture the user's micro-movements; ultra-wideband (UWB) technology is better at judging distances to targets but not at capturing subtle posture trends, and it requires positioning tags. In consumer products, the price of UWB base stations with tags is not advantageous compared to 60GHz millimeter-wave radar. These problems directly prevent robotic pets from achieving natural behavioral interactions with users in real-world, complex scenarios.

[0015] To effectively address the problem of robotic pets lacking the ability to proactively adjust social distance in complex environments, this application provides a robotic pet control method, a robotic pet, a computer-readable storage medium, and a computer program product. The method first acquires distance information between the user and the robotic pet, the user's orientation relative to the robotic pet, and the user's micro-motion information, whereby the micro-motion information can characterize the user's non-displacement, subtle limb movements or signal changes caused by physiological activities. Then, based on the distance, orientation, and micro-motion information, human dynamic trend feature information is constructed. Subsequently, the user's interaction state is identified based on the human dynamic trend feature information, and the target behavior information of the robotic pet is determined based on the interaction state. Finally, the interactive behavior of the robotic pet is controlled based on the target behavior information. This application, by constructing human dynamic trend feature information to identify the user's interaction state, determining the target behavior information of the robotic pet based on the interaction state, and controlling the robotic pet based on the determined behavior information, enables the robotic pet to interact naturally with the user in various scenarios, improving the user experience.

[0016] Based on the above analysis, the embodiments of this application will be further described below with reference to the accompanying drawings.

[0017] Reference Figure 1 , Figure 1 This is a flowchart of a control method for a robotic pet according to an embodiment of this application. The specific flow of this method may include, but is not limited to, steps S110 to S140.

[0018] Step S110: Obtain distance information between the user and the robot pet, orientation information of the user relative to the robot pet, and micro-motion information of the user, wherein the micro-motion information can characterize the user's non-displacement subtle limb movements or signal changes caused by physiological activities. Step S120: Construct human dynamic trend characteristic information based on distance information, orientation information, and micro-motion information; Step S130: Identify the user's interaction state based on the human body dynamic trend feature information, and determine the target behavior information of the robot pet based on the interaction state; Step S140: Control the interactive behavior of the robot pet based on the target behavior information.

[0019] In this embodiment, the user's micro-motion information refers to the signal feature information corresponding to small movements of the body parts when the user's overall position is basically still. It can characterize the user's non-displacement subtle limb movements or signal changes caused by physiological activities. For example, the rise and fall of breathing when the user is sitting still, the slight shaking of the fingers, and the slight vibration of the chest with the heartbeat are all micro-motion information.

[0020] In this embodiment, when acquiring the distance information between the user and the robot pet, the user's orientation information relative to the robot pet, and the user's micro-motion information, millimeter-wave sensing signals can be continuously sent to the surrounding environment and the echo signals of the millimeter-wave sensing signals can be continuously received. Then, environmental point cloud data can be generated based on the echo signals, and the distance information between the user and the robot pet, the user's orientation information relative to the robot pet, and the user's micro-motion information can be extracted from the environmental point cloud data.

[0021] In this embodiment, the millimeter-wave sensing signal refers to the millimeter-wave electromagnetic waves radiated outwards by the device's transmitter. After continuously sending millimeter-wave sensing signals to the surrounding environment, the device can continuously receive the echo signals. These echo signals are electromagnetic wave signals reflected back to the device after the millimeter-wave sensing signals come into contact with objects in the environment, and can be used to generate environmental point cloud data. Environmental point cloud data refers to a set of three-dimensional coordinate points obtained from the echo signals, used to characterize the spatial positions of various obstacles in the environment, such as the three-dimensional point information corresponding to indoor furniture, robotic pets, and the user's body. By analyzing this environmental point cloud data, the distance information, orientation information, and micro-motion information between the user and the robotic pet can be obtained.

[0022] In this embodiment, when extracting distance, orientation, and micro-motion information from point cloud data, static background information can be identified first in the environmental point cloud data and then filtered out to obtain dynamic object point cloud data. Initial user point cloud data is then identified from the dynamic object point cloud data, and the distance between the user and the robotic pet, the user's orientation relative to the robotic pet, and the user's micro-motion information are determined based on this initial user point cloud data. It should be understood that continuous multi-frame environmental point cloud data can separate static and dynamic environmental information. Static information corresponds to point clouds of objects with fixed positions, such as walls and furniture, while dynamic information corresponds to point clouds of targets that continuously move, such as users and robotic pets. For example, indoor wall points are static information, while the points corresponding to a moving user are dynamic information. By identifying and filtering out static background information from the environmental point cloud data, the obtained dynamic object point cloud data retains only the feature information corresponding to moving targets within the scene, thereby reducing data interference from irrelevant environmental objects and facilitating accurate tracking of user and robotic pet information.

[0023] In this embodiment, when identifying initial user point cloud data from dynamic object point cloud data, multiple target point cloud clusters can be identified first. Then, the second average micro-motion energy of each target point cloud cluster is obtained. Subsequently, among all target point cloud clusters, those whose second average micro-motion energy is higher than a preset energy threshold within a preset time period and whose corresponding point cloud cluster size meets a preset size threshold are determined as initial user point cloud data. Specifically, in this process, the dynamic object point cloud data is first clustered and segmented to obtain multiple target point cloud clusters, each target point cloud cluster corresponding to a single dynamic object in the environment. Then, the average micro-motion energy, i.e., the second average micro-motion energy, corresponding to each target point cloud cluster can be calculated using a micro-Doppler energy integration algorithm. Subsequently, continuous validation is performed based on a time sliding window to filter out items within a preset time period. Target point cloud clusters that consistently exceed a preset energy threshold, while maintaining the point cloud cluster size within a preset size threshold range (e.g., 0.1 to...). ) Target point cloud clusters that conform to the typical reflective area characteristics of the human body are used to obtain effective target point cloud clusters, which are the initial user point cloud data.

[0024] In this embodiment, when determining the distance information between the user and the robot pet, the orientation information of the user relative to the robot pet, and the micro-motion information of the user based on the initial user point cloud data, the orientation information of the user relative to the robot pet can be determined first based on the initial user point cloud data. Then, the transmission direction of the millimeter wave sensing signal can be directed towards the user based on the orientation information to focus the sensing range and continuously acquire the target user point cloud data. Subsequently, the distance information between the user and the robot pet and the micro-motion information of the user can be determined based on the target user point cloud data.

[0025] In this embodiment, human dynamic trend feature information refers to feature information that reflects the continuous change in the human body's position, including the direction of movement, speed of movement, and continuous motion state. For example, the changes in a user's behavior—continuously moving closer to the robotic pet or gradually moving away from it—belong to human dynamic trend feature information. Figure 2 As shown, the execution process of step S120, which constructs the dynamic trend feature information of the human body, may include, but is not limited to, steps S210 to S250.

[0026] Step S210: Determine the micro-motion energy level to characterize the user's activity intensity based on the micro-motion information; Step S220: Determine distance trend information based on the distance information to characterize the change in distance between the user and the robot pet; Step S230: Determine the orientation stability based on the orientation information to characterize whether the user is continuously facing the robot pet; Step S240: Determine the high-frequency disturbance intensity information to characterize whether the user has a tendency to actively interact based on the micro-motion information; Step S250: Combine features based on micro-motion energy level, distance trend information, orientation stability and high-frequency disturbance intensity information to obtain human dynamic trend feature information.

[0027] In this embodiment, the intensity of human activity is positively correlated with the energy of the micro-Doppler spectrum. Even when the human body remains stationary, various micro-movements such as breathing, fine-tuning of the center of gravity, and small limb movements can still generate micro-Doppler signals. By integrating the micro-Doppler energy within a continuous time window, the intensity of human activity can be quantitatively assessed. Different intensities of human activity correspond to different levels of micro-movement energy.

[0028] In this embodiment, since micro-motion information can reflect the micro-motion state of the human body, the micro-motion energy level used to characterize the intensity of user activity can be determined based on the micro-motion information. Specifically, the first average micro-motion energy within a preset time period can be determined based on the micro-motion information, and then the micro-motion energy level used to characterize the intensity of user activity can be determined based on the first average micro-motion energy. For example, the corresponding micro-motion information is extracted from each frame of echo signal, and the instantaneous micro-motion energy e(t) is calculated based on this information. A time window T containing N frames of data is selected as the preset time period, thereby allowing the calculation of the average micro-motion energy within the time window T, statistically obtained from multiple sets of micro-motion information. ; further By comparing the energy level with a pre-set threshold, the corresponding micro-movement energy level can be obtained, thereby determining the user's current activity intensity. For example, a maximum energy threshold can be pre-set. With minimum energy threshold ,when < When the user is in a low-intensity state, such as sitting, the corresponding micro-movement energy level is A; when ≤ ≤ At this time, it can be determined that the user is making slight limb movements, is in a state of moderate activity intensity, and the corresponding micro-movement energy level is B; when > When the user is in a state of high activity intensity, such as walking, the corresponding micro-movement energy level is C.

[0029] In this embodiment, distance trend information can be used to determine the dynamic change in the relative distance between the user and the robotic pet. When determining the distance trend information to characterize the change in distance between the user and the robotic pet, multiple user radial velocities within a preset time period can be determined based on the distance information. Then, the distance trend information characterizing the change in distance between the user and the robotic pet can be determined based on these multiple user radial velocities. The distance information represents the real-time spatial distance between the user and the robotic pet. Based on this distance information, the corresponding user radial velocity Vr within a continuous time period T can be calculated. If most Vr values ​​are positive, it can be determined that the user is moving closer to the robotic pet; if most Vr values ​​are negative, it can be determined that the user is moving away from the robotic pet; if most Vr values ​​approach zero, it can be determined that the user is in a relatively stationary state.

[0030] In this embodiment, azimuth stability is used to determine whether a user continuously faces the robotic pet, and is a key indicator for assessing the user's willingness to socialize with the robotic pet. For example, the user's azimuth angle θ is set as the criterion. When the value of azimuth angle θ approaches zero, it means the user is facing the robotic pet directly; if the azimuth angle θ is in any other range, it means the user is deviating from the robotic pet's front. When determining the azimuth stability based on the user's azimuth information relative to the robotic pet, the variance of the azimuth information within a preset time period, i.e., the variance of the azimuth angle, can be determined first. Then, the azimuth stability is determined based on this variance. By calculating the variance of the azimuth angle θ within the preset time period, the fluctuation and deviation of the user's orientation during that period can be quantified. Generally, the smaller the variance, the more stable the user's orientation; the larger the variance, the more frequently the user glances left and right, indicating a constantly changing orientation. The specific process of determining azimuth stability based on variance is as follows: When the variance is less than a preset threshold, and the average azimuth angle remains within the angle range corresponding to the front of the robot pet during that time period, the robot pet can determine that the user is paying attention to it and has the intention to interact, and can prepare to initiate interaction. If an extremely high azimuth velocity is detected in a short period of time, and the final calculated average azimuth angle has deviated from the center area in front, it can be determined that the user has quickly turned away from the robot pet or quickly shifted their gaze. The robot pet can then identify whether the user is refusing to interact, is about to end the interaction, or is preparing to leave.

[0031] In this embodiment, high-frequency disturbance intensity information can be used to identify micro-burst actions by the user with an intention to actively interact, such as raising a hand, reaching out a hand, or suddenly leaning forward. These actions generally have the characteristics of short duration and triggering micro-Doppler energy abrupt changes. When determining the high-frequency disturbance intensity information used to characterize whether the user has an intention to actively interact based on micro-motion information, the instantaneous micro-motion energy can be determined first based on the micro-motion information, and then the high-frequency disturbance intensity information used to characterize whether the user has an intention to actively interact can be determined based on the instantaneous micro-motion energy. In specific implementation, the instantaneous micro-motion energy e(t) is first determined based on the micro-motion information; then, the instantaneous micro-motion energy e(t) is subjected to high-pass filtering to filter out the energy components corresponding to low-frequency movements such as breathing and slow body movement, and the high-frequency components in the signal are obtained. Next, the high-frequency components within a preset time period (e.g., within 0.5 seconds) are analyzed. By performing integration, high-frequency energy is obtained. High-frequency energy Compare with the reference high-frequency energy corresponding to the user's quiet state, if If the threshold is exceeded, it can be determined that a single active interaction pulse has been detected; if two or more active interaction pulses are detected within a continuous duration T, it can be determined that the user has an active interaction trend.

[0032] In this embodiment, after processing steps S210 to S240, the corresponding micro-motion energy level, distance trend information, orientation stability, and high-frequency disturbance intensity information can be obtained. Further, when executing step S250, the micro-motion energy level, distance trend information, orientation stability, and high-frequency disturbance intensity information can be time-smoothed to eliminate state jitter caused by noise. Then, the current timestamp and the time-smoothed micro-motion energy level, distance trend information, orientation stability, and high-frequency disturbance intensity information can be spliced, weighted fused, or subjected to model prediction—any one of these processes—to finally obtain the human body dynamic trend characteristic information.

[0033] In this embodiment, when concatenating the current timestamp with time-smoothed micro-motion energy level, distance trend information, orientation stability, and high-frequency disturbance intensity information, the four types of features and the timestamp can be directly combined sequentially. For example, if the currently detected micro-motion energy level is B, the distance trend is approaching, the orientation stability is high, and the high-frequency disturbance intensity is low, these can be directly concatenated with the current timestamp to form a five-dimensional human dynamic trend feature vector.

[0034] In this embodiment, when weighting and fusing the current timestamp along with time-smoothed micro-motion energy levels, distance trend information, orientation stability, and high-frequency disturbance intensity information, different weights can be assigned to each feature based on the actual interaction scenario, and the fusion calculation can be performed accordingly. For example, when it is determined to be an active interaction, the weights of high-frequency disturbance intensity and distance trend can be increased, while the weight of orientation stability in a stable state can be decreased. The fusion calculation yields a comprehensive feature, which is then combined with the timestamp to form a five-dimensional human dynamic trend feature vector.

[0035] In this embodiment, when predicting the model using the current timestamp and time-smoothed micro-motion energy levels, distance trend information, orientation stability, and high-frequency disturbance intensity information, a feature sequence composed of the four types of features and the timestamp can be input into the time-series model, automatically outputting optimized human dynamic trend features. Common models that can be used include time-series networks such as LSTM, BiLSTM, and Transformer. For example, by inputting user micro-motion, distance, orientation, disturbance features, and timestamps from multiple consecutive frames into an LSTM model, the model automatically learns the patterns of user behavior changes and outputs a five-dimensional human dynamic trend feature vector.

[0036] In this embodiment, the user's interaction state reflects their current willingness to interact with the robotic pet. Human dynamic trend characteristic information includes micro-motion energy level, distance trend information, orientation stability, and high-frequency disturbance intensity information. When identifying the user's interaction state based on the human dynamic trend characteristic information, the identification and mapping can be specifically performed based on the micro-motion energy level, distance trend information, orientation stability, and high-frequency disturbance intensity information to obtain the user's interaction state. For example, when the high-frequency disturbance intensity information detects an active interaction pulse, the distance trend information shows that the user is continuously approaching the robotic pet, and the orientation stability is high while the micro-motion energy level is at a moderate level, it can be determined that the user has a strong willingness to interact. If the distance trend information indicates that the user is gradually moving away, and the orientation angle continues to deviate from the area in front, it is determined that the user's willingness to interact decreases, and they tend to end the interaction.

[0037] In this embodiment, when identifying the user's interaction state based on human body dynamic trend feature information, the human body dynamic trend feature information can be identified and categorized, and mapped to the following four expected interaction states: A. Low Interaction State: The user is at a low level of activity, has no intention to approach, and does not take any initiative to interact. In this case, the robot pet should remain silent and avoid disturbing the user.

[0038] B. Potential Interaction State: The user shows a tendency to face or slowly approach the robotic pet, or their activity level has increased but a clear intention to interact has not yet been formed. In this case, the robotic pet can tentatively and slowly approach.

[0039] C. Proactive Interaction State: The user exhibits clear, rapid approaching behavior, or an approaching trend accompanied by strong, high-frequency disturbances, indicating that the user is actively seeking interaction. In this case, the robot pet should proactively approach and perform interactive behavior.

[0040] D. Rejection of Interaction: If the user quickly moves away from the robotic pet or turns their head away rapidly, regardless of whether it is accompanied by high-frequency disturbances, it is determined that the user intends to reject or refuse interaction. At this time, the robotic pet should immediately stop approaching and actively retreat.

[0041] In this embodiment, when determining the target behavior information of the robotic pet based on the interaction state, the corresponding behavioral intent and movement direction can be determined first based on the interaction state, and then the target behavior information of the robotic pet can be determined based on the behavioral intent and movement direction. For example, when the interaction state is low interaction, the corresponding behavioral intent is silent companionship, and the movement direction is set to not move or slowly wander; when the interaction state is potential interaction, the corresponding behavioral intent is slowly approaching, and the movement direction is set to slowly approach the user; when the interaction state is active interaction, the corresponding behavioral intent is actively approaching and interacting, and the movement direction is set to approach the user; when the interaction state is repulsive interaction, the corresponding behavioral intent is actively retreating, and the movement direction is set to move away from the user. Subsequently, based on the behavioral intent, movement direction, and the current actual distance between the robotic pet and the user, the target behavior information of the robotic pet can be determined. This behavior information can be used to drive the robotic pet to smoothly move to the desired companionship distance.

[0042] In this embodiment, when controlling the interactive behavior of the robotic pet based on target behavior information, corresponding motion parameters and interaction distances can be generated first based on the target behavior information, and then the interactive behavior of the robotic pet can be controlled based on the motion parameters and interaction distances. The interaction distance can be set according to the interaction state corresponding to the target behavior information. For example, when the target behavior information corresponds to a low interaction state, the expected interaction distance is set to 1.5 meters; when it corresponds to a potential interaction state, the expected interaction distance is set to 1 meter; when it corresponds to an active interaction state, the expected interaction distance is controlled within 0.4 meters; and when it corresponds to a repulsive interaction state, the expected interaction distance is kept beyond 2.0 meters. In this embodiment, by continuously comparing the current actual distance and interaction distance between the robotic pet and the user to adjust the motion parameters and limiting the output movement speed, it can be ensured that the upper limit of the movement speed under various interaction intentions meets the safety constraints and anthropomorphic behavior requirements. At the same time, the orientation can also be continuously adjusted according to the real-time azimuth difference to drive the robotic pet to always face the user, maintain a social gaze effect, and thus ensure that the overall movement process of the robotic pet is smoother and more natural.

[0043] Reference Figure 3 , Figure 3 This is a flowchart of a control method for a robotic pet according to another embodiment of this application. The specific flow of this method may include, but is not limited to, steps S310 to S340.

[0044] Step S310: Obtain information on the interaction process between the robot pet and the user; Step S320: Determine the user's personalized interaction preference information based on the interaction process information; Step S330: Generate candidate adjustment actions based on personalized interaction preference information; Step S340: Update the control strategy parameters of the robot pet based on the candidate adjustment actions.

[0045] In this embodiment, the interaction process information between the robot pet and the user may include at least the following data from the most recent N interactions: the duration, number of occurrences, and transition matrix of each interaction state; the user's response delay time to the robot pet's approach / remote behavior; the conditional probability of the user entering a repulsion state under different distance ranges; the user's average approach speed and high-frequency disturbance intensity when an active interaction state occurs; and the frequency and correlation characteristics of abnormal events.

[0046] In this embodiment, personalized interaction preference information includes at least one of the following: comfortable distance range, interaction enthusiasm, response speed, and rejection sensitivity. The comfortable distance range can be used to determine a reference value for the user's desired companionship distance; interaction enthusiasm can be used to determine whether the user prefers closer, faster interaction or farther, slower companionship; response speed can be used to determine the user's desired interaction speed; and rejection sensitivity can be used to adjust the target distance for proactive interaction.

[0047] In this embodiment, when generating candidate adjustment actions based on personalized interaction preference information, various parameters in the personalized interaction preference information can be extracted first to adjust control parameters such as expected interaction distance / movement speed limit / state switching threshold, and then candidate adjustment actions that can be executed by the robot pet can be generated based on the updated control parameters.

[0048] In this embodiment, the control strategy parameters of the robotic pet refer to a series of preset operating parameters used to manage the interactive behavior of the robotic pet, including the expected interaction distance, the upper limit of movement speed, and the judgment thresholds for various interaction states corresponding to different interaction states. When updating the control strategy parameters of the robotic pet according to the candidate adjustment actions, the parameter correction amount corresponding to the candidate adjustment actions can be analyzed first. The correction amount is then used to adjust the original control strategy parameters to obtain new control strategy parameters that adapt to the current user's personalized interaction preferences. Finally, the interactive behavior between the robotic pet and the user is controlled based on the updated control strategy parameters.

[0049] In this embodiment, if the environmental point cloud data detected by the robot pet is abnormal or lost, the robot pet's movement behavior can be stopped, and the robot pet's current behavior can be switched to silent companionship behavior.

[0050] In this embodiment, if the robot pet's sensor malfunctions for a preset duration, the robot pet's current behavior state can be switched to a standby state.

[0051] In this embodiment, if the robot pet receives conflicting or non-interactive commands from the user, it can remain in a standby state.

[0052] In this embodiment, if the target behavior information is determined to be unreasonable, the robot pet's default behavior can be used to control the robot pet's interactive behavior.

[0053] In this embodiment, if it is determined that there is a security risk in the target behavior information, the current behavior state of the robot pet can be switched to a standby state.

[0054] In this embodiment, if it is determined that the robot pet is currently in a low battery or abnormal power consumption state, the robot pet can be controlled to move to the vicinity of the charging dock and issue a charging prompt message.

[0055] In this embodiment, if it is detected that the current space is not suitable for large-scale movement, the current behavior state of the robot pet can be switched to a stationary interaction state.

[0056] The control method of this robotic pet will be described in detail below with a specific embodiment.

[0057] This embodiment provides an intelligent interaction system for a robotic pet. This system can be equipped with and run the perception, judgment, and control logic in the aforementioned control method. Through the collaborative work of multiple modules, it completes the entire process of millimeter-wave signal analysis, user state recognition, interaction intent judgment, behavior strategy generation, and device control, ultimately achieving intelligent interaction effects. For example... Figure 4 As shown, the interactive system includes a millimeter-wave sensing module 410, a human dynamic trend analysis module 420, a pet social distance decision-making module 430, a social strategy learning and reconstruction module 440, a motion control execution module 450, and an anomaly handling and adjustment module 460. Among them: The millimeter-wave sensing module 410 is connected to the human dynamic trend analysis module 420. It periodically transmits millimeter-wave frequency-modulated continuous wave (FMCW) signals into the environment and receives reflected signals from human targets, providing raw sensing input for human dynamic trend analysis. Its module block diagram is shown below. Figure 5As shown, the millimeter-wave sensing module 410 mainly consists of three parts: an antenna array, a radio frequency transceiver system, and a signal processing unit. The millimeter-wave antenna array, as the signal transceiver front end of the sensing module, can adopt a 4×4 planar array architecture. At the 60GHz high-frequency band, its physical array size is only about 10×10mm², which can be perfectly embedded in the robot pet's head without compromising its overall biomimetic appearance. This array, combined with a digital phase control network, enables flexible beamforming and dynamic scanning, covering approximately ±60° of two-dimensional scanning, with a 3dB beamwidth converged to approximately 15°. This narrow beam characteristic gives the radar extremely high spatial resolution, allowing it not only to accurately distinguish the target's location but also to keenly capture micro-Doppler disturbances such as chest rise and fall and slight adjustments in the center of gravity. As the robot pet moves or its head rotates, the antenna array can adjust its orientation through mechanical and electronic coordination, achieving seamless spatial coverage, effectively eliminating detection blind spots, and ensuring comprehensive information collection from the human-machine interaction environment. The radio frequency transceiver system consists of millimeter-wave radio frequency front-end devices and a transceiver. It generates FMCW signals and receives reflected signals. It can switch operating frequencies and adjust transmit power in real time according to environmental interference intensity, significantly improving the module's robustness and signal-to-noise ratio in complex home environments, providing stable underlying physical signals for subsequent micro-motion feature extraction. The signal processing unit consists of an AD / DA module, a digital down-conversion module, and a filtering module. It is responsible for preprocessing and solving the raw intermediate frequency signal output from the radio frequency front-end, extracting the intermediate frequency signal containing information on the distance, speed, and orientation of the human target from the echo, and generating a point cloud data matrix to provide raw data support for subsequent interactive decision-making by the robotic pet.

[0058] The human body dynamic trend analysis module 420 is connected to the millimeter-wave sensing module 410 and the pet social distance decision module 430, respectively. It can be used to process the raw point cloud data input by the millimeter-wave sensing module 410 and output the processed user human body dynamic trend feature data to the pet social distance decision module 430. Its structure is as follows: Figure 6 As shown, it consists of three parts: target detection and continuous tracking unit, dynamic trend feature extraction unit, and feature fusion and smoothing unit.

[0059] The target detection and continuous tracking unit is used to detect and filter human targets from the point cloud data of the millimeter-wave sensing module 410, continuously track user activity trajectories, and provide stable time-series data for trend analysis. This unit first filters out static backgrounds, retaining dynamic point clouds caused by human movement or micro-motions. Since the position and intensity of point clouds reflected by static objects such as walls and furniture remain stable over a long period, if the distance, angle, and intensity changes of a detection point within N consecutive frames are all less than a threshold, it is marked as static background and filtered out. To eliminate interference from other dynamic objects (such as robotic vacuum cleaners, fans, etc.), the unit can calculate the average micro-motion energy of the point cloud within each target cluster using micro-Doppler integration. .when The value remains above the threshold for a period of time, and the size of the target point cloud cluster matches the typical reflective area of ​​the human body (e.g., 0.1 to 1.5 oz). When the target is detected, it can be determined that the object is a human target. The target detection and continuous tracking unit can feed back the azimuth angle θ(t) of the current human target to the radar perception module in real time, so that it can adjust the beam direction to continuously track the target.

[0060] The dynamic trend feature extraction unit is used to perform statistical analysis on the tracked time series and extract human dynamic trend features. This unit can determine the motion trend parameters from four dimensions: micro-motion energy characteristics, distance change trend, orientation stability, and high-frequency disturbance characteristics, as detailed below: a. The intensity of human activity is positively correlated with the energy of the micro-Doppler spectrum. Even when the human body remains stationary, breathing, changes in center of gravity, and limb movements will still generate micro-Doppler signals. The dynamic trend feature extraction unit can quantify the intensity of human activity by integrating the micro-Doppler energy within a continuous time window. Let the instantaneous micro-motion energy of each frame be e(t), then the average energy within the time window T (N frames) is... .Will With the preset maximum energy threshold and minimum energy threshold The comparison outputs the energy level of the exercise to determine the intensity of the user's activity. The comparison results are shown in Table 1.

[0061] Table 1. Micro-movement Energy Level Comparison Table

[0062] b. Distance change trend is used to determine the relative distance between the user and the robotic pet, and whether the user is approaching or moving away from the robotic pet. This feature is determined by the radial velocity. The results are obtained by performing symbol statistics over a continuous time period T. If the majority of sampling points... A positive value indicates that the user is approaching the robotic pet; if the majority of sampling points are positive... A negative value indicates that the user is moving away from the robotic pet; if most sampling points... If the value is close to zero, it indicates that the user is in a static state.

[0063] c. Orientation stability is used to determine whether the user continuously faces the robotic pet, which is key to assessing the user's willingness to socialize with the robotic pet. If the user's orientation angle... A value close to 0° indicates the user is directly facing the robotic pet; other values ​​indicate the user is off-center from the pet. This is determined through calculation. Variance within time window T This can measure the degree of deviation of the user's orientation during this period. Typically, A smaller value indicates a more stable user gaze direction; A higher value indicates that the user is looking around. When... If the angular velocity is below a threshold and the average angle remains within a certain range directly in front of the robotic pet, the robotic pet can determine that the user is paying attention to it and is ready to initiate interaction. If an extremely high angular velocity is detected within a short period of time, and the final average azimuth angle deviates from the center area (e.g., greater than 45°), it indicates that the user suddenly turns their back to the robotic pet or quickly looks away. This can be considered a signal of "refusal to interact" or "imminent departure." Other situations indicate that the user has no specific social orientation.

[0064] d. High-frequency perturbation features are used to detect "micro-burst" actions by users that contain active interaction intentions, such as raising a hand, reaching out, or suddenly leaning forward. These actions are usually accompanied by short-duration, high-frequency micro-Doppler energy abrupt changes. The dynamic trend feature extraction unit performs high-pass filtering on the instantaneous micro-motion energy e(t), which can filter out low-frequency components such as breathing and slow movement, obtaining the high-frequency components. For within 0.5 seconds Integrating to obtain ,Will The signal is compared with the high-frequency energy threshold of the user in a quiet state. If the signal exceeds the threshold, it is determined as an active interaction pulse. If more than two active interaction pulses occur within a continuous time T, the signal indicates that there is an active interaction trend.

[0065] The feature fusion and smoothing unit applies first-order low-pass filtering to smooth the four features (micro-motion energy level, distance trend, orientation stability, and high-frequency disturbance intensity) output by the dynamic trend feature extraction unit to eliminate state jitter caused by noise. The smoothed features are then combined with timestamps to form a five-dimensional output vector. This vector can serve as complete human dynamic trend feature data and is transmitted to the pet social distance decision module 430 for subsequent interaction strategy generation.

[0066] The pet social distance decision module 430 is connected to the human dynamic trend analysis module 420, the social strategy learning and reconstruction module 440, the motion control execution module 450, and the anomaly handling and adjustment module 460. The pet social distance decision module 430 receives the human dynamic trend feature vector output by the human dynamic trend analysis module 420, combines it with the strategy parameters provided by the social strategy learning and reconstruction module 440 and the anomaly correction signal fed back by the anomaly handling and adjustment module 460, generates specific social distance control instructions, and sends them to the motion control execution module 450 for execution. The structure of the pet social distance decision module 430 is as follows: Figure 7 As shown, it consists of four parts: an interaction state recognition unit, a behavior intention recognition unit, a motion parameter generation unit, and a behavior coordination output unit. Its workflow is as follows: Figure 8As shown. When the pet social distance decision module 430 receives human dynamic trend feature data input from the human dynamic trend analysis module 420, it begins to work. The interaction state recognition unit can determine the current user's interaction state based on the combination of trend features and transmit the result to the behavior intention recognition unit, which generates the robot pet's behavior intention matching the user's interaction state. After receiving the behavior intention instruction, the motion parameter generation unit and the behavior coordination output unit can generate specific motion control instructions and instructions such as sound, light, and limb movements based on the expected distance, respectively, and output specific control parameters to the motion control execution module 450 for execution. The interaction state recognition unit is responsible for recognizing and classifying the feature vectors output by the human dynamic trend analysis module 420, mapping them to four expected interaction states, and eliminating state jumps caused by noise through a state preservation mechanism. This unit makes judgments based on multiple conditions of feature vectors, specifically mapping user behavior to the following four states: a. Low-interaction state: The user is at a low level of activity, has no intention to approach, and does not take any initiative to interact. In this case, the robot pet should remain silent and avoid disturbing the user.

[0067] b. Potential interaction state: The user shows a tendency to face or slowly approach the robotic pet, or the activity level has increased but no clear intention to interact has yet been formed. At this time, the robotic pet can tentatively and slowly approach.

[0068] c. Proactive Interaction State: The user exhibits clear, rapid approaching behavior, or an approaching trend accompanied by strong, high-frequency disturbances, indicating that the user is actively seeking interaction. In this case, the robot pet should proactively approach and perform interactive behavior.

[0069] d. Rejection of Interaction: If the user quickly moves away from the robotic pet or turns their head away rapidly, regardless of whether it is accompanied by high-frequency disturbances, it is determined that the user intends to reject or refuse interaction. At this time, the robotic pet should immediately stop approaching and actively retreat.

[0070] The specific determination criteria are shown in Table 2. The interactive state must be maintained continuously for a certain duration before output can be confirmed. For example, low interaction and potential interaction states must be maintained continuously for 0.5 seconds; active interaction and repulsive interaction states must be maintained continuously for at least 1.5 seconds to prevent brief actions from triggering invalid behavior. It should be noted that the duration here is only an example; the specific duration can be flexibly set according to the actual application scenario, and this embodiment does not impose any restrictions on it.

[0071] Table 2. User Expected Interaction State Determination Table

[0072] The behavior intent generation unit is responsible for converting the interaction state into specific behaviors of the robot pet, and fine-tuning them by integrating the personalized strategies of the social strategy learning and reconstruction module 440. The default corresponding behaviors are shown in Table 3: low interaction maps to silent companionship, potential interaction maps to tentative approach, active interaction maps to proactive approach, and rejection maps to active avoidance.

[0073] Table 3. Correspondence between Interaction State and Behavior Distance

[0074] The motion parameter generation unit is responsible for generating specific motion control commands based on the behavioral intent recognition results and the current actual distance between the robot pet and the user, enabling the robot pet to smoothly move to the desired companionship distance. For example, in the initial configuration, the desired distance for a low-interaction state can be set to 1.5 meters, for a potential interaction state to 1 meter, for an active interaction state to be within 0.4 meters, and for a repulsion state to be beyond 2.0 meters. This desired distance can be dynamically adjusted according to the strategy parameters iterated by the social strategy learning and reconstruction module 440. The motion parameter generation unit limits the speed output by comparing the current distance with the target distance, ensuring that the upper speed limit under different intents meets safety and anthropomorphic requirements. Simultaneously, it drives the robot pet to always face the user directly based on the azimuth difference, maintaining social gaze, thereby ensuring the smoothness and naturalness of the robot pet's movement.

[0075] The behavior coordination output unit is responsible for coordinating the robot pet's vocalizations, lights, and body movements to ensure consistency in emotional expression with its motion behavior. The specific mapping relationship is shown in Table 4. When the robot pet's behavior intensity is high, the unit can increase the light flashing frequency or vocalization frequency to match the intensity of the multimodal feedback with the robot pet's movement amplitude, avoiding a sense of incoordination where the movements are intense but the expressions are stiff. Conversely, it reduces the intensity of the audio-visual feedback to create a more layered overall performance. This unit converts the generated sound, light, and movement patterns into standard commands and sends them to the motion control execution module 450 for execution.

[0076] Table 4. Behavior Coordination Map

[0077] The social strategy learning and reconstruction module 440 is responsible for continuously learning user interaction preferences and dynamically adjusting the strategy parameters (such as target distance, speed factor, state threshold, etc.) used by the pet social distance decision module 430 to personalize and adapt to the habits of different users, thereby improving the naturalness of long-term companionship interactions. Figure 9As shown, the social strategy learning and reconstruction module 440 consists of four parts: an interaction data acquisition and recognition unit, a user preference inference unit, a strategy parameter generation and reconstruction unit, and a strategy smoothing and safety constraint unit. It periodically collects behavior execution results and user response data, and after learning and strategy optimization, generates new strategy parameters to guide the pet social distance decision-making module 430 in making decisions. Among them: The interaction data acquisition and recognition unit is responsible for long-term recording and summarizing the interaction data between the robot pet and the user, providing a foundation for preference inference. This unit can statistically analyze the following data from the most recent N interactions: the duration, frequency, and transition matrix of each interaction state; the user's response delay to the robot pet's approach / remote behavior; the conditional probability of the user entering a repulsion state at different distance intervals; the user's average approach speed and high-frequency disturbance intensity when an active interaction state occurs; and the frequency and correlation characteristics of abnormal events. After each statistical cycle, this unit outputs a set of statistical feature vectors to the user preference inference unit.

[0078] The user preference inference unit is responsible for inferring the user's personalized interaction preferences, including comfort distance range, interaction enthusiasm, and aversion sensitivity. The specific mechanism is as follows: (1) Distance comfort zone inference: Statistically determine the probability of a user entering a rejection state within each distance zone. Mark the distance range where the rejection probability is lower than the first preset threshold (e.g., 10%) as the comfort zone, and mark the distance range where the rejection probability is higher than the second preset threshold (e.g., 30%) as the rejection zone. Use the median of the comfort zone as the reference value for the companionship distance expected by the user.

[0079] (2) Evaluation of interaction enthusiasm: Calculate the ratio of the total duration of active interaction state to the total interaction time per unit time. If the ratio is higher than the preset enthusiasm threshold (e.g., 0.3), the user is judged to be highly enthusiastic and tends to interact more closely and faster; otherwise, the user is judged to be low enthusiastic and tends to accompany more distantly and slower.

[0080] (3) Response speed analysis: Statistically analyze the response delay of users to the robot pet's approach behavior. If the average response delay is less than the first response threshold (e.g., 0.8 seconds), it indicates that users expect fast interaction and the approach speed factor can be increased; if the delay is greater than the second response threshold (e.g., 2.5 seconds), the approach speed factor should be decreased.

[0081] (4) Abnormal event correlation analysis: If the rejection state is always triggered when the actual distance of the robot pet is less than a certain value (e.g., 0.6 meters), then the value is recorded as the rejection distance threshold and used to adjust the target distance of the active interaction state.

[0082] The policy parameter generation and reconstruction unit employs a policy gradient reinforcement learning algorithm to automatically learn the optimal mapping from user preference states to policy parameter adjustment amounts. Its complete workflow is as follows: a. At the start of the strategy update cycle, the unit first reads the state vector provided by the user preference inference unit. These features together characterize the user's interaction tendencies in the current time period. b. This unit generates candidate adjustment actions based on the current state and current policy parameters. The mean action is obtained by a linear combination of state features, and the actual values ​​are normally distributed around this mean. The dispersion of the distribution is controlled by a preset standard deviation. The standard deviation is larger initially to encourage exploration, and gradually decreases later to make full use of learned knowledge. An action is a continuous vector, for example, simultaneously including the adjustment amount of the expected distance of active interaction and the adjustment amount of the approach speed factor. The randomness in the sampling process ensures that the algorithm can explore the unknown policy space and avoid getting trapped in local optima. The sampled actions are superimposed on the currently running policy parameter set to obtain the candidate parameter set. In this embodiment, the adjustment amount of the expected distance of active interaction and the adjustment amount of the approach speed factor can both be limited to preset adjustment intervals. For example, the adjustment amount of the expected distance of active interaction can be limited to the interval between -0.2 meters and +0.2 meters, and the adjustment amount of the approach speed factor can be limited to the interval between -0.15 and +0.15.

[0083] c. The candidate parameter set is first processed by the policy smoothing and safety constraint unit, including limiting the magnitude of single changes and pruning to a safe boundary, and then distributed to the pet social distance decision module 430 and the motion control execution module 450. Throughout the subsequent update cycle, the robot pet operates entirely with the new policy parameters.

[0084] d. During the implementation of the new strategy, the pet social distance decision module 430 and the motion control execution module 450 continuously record interaction data, including: the total duration of the active interaction state, the total duration of the repulsion state, the number of collisions, and the absolute value of the deviation between the robot pet's actual average companion distance and the user's comfortable distance. At the end of the update cycle, the unit can calculate a scalar reward value based on these statistical data to quantitatively evaluate the interaction quality under the new parameters. The reward function is designed with the principle that the longer the active interaction time, the higher the reward; excessive repulsion time, collision count, and distance deviation will result in penalties. These factors are combined with certain weights to serve as the basis for strategy optimization. When combining factors with certain weights, the weight of the active interaction duration can be set to positive, and the rest to negative. Simultaneously, the unit records the state at the start of the next cycle.

[0085] e. Each unit maintains a cyclic experience buffer, storing the current cycle's state, selected action, calculated reward, and next state as an experience entry. When the accumulated experience in the buffer reaches a preset batch size, a policy parameter update is triggered. The update process is as follows: For each experience in the buffer, calculate the discounted cumulative reward from the start of that step to the end of the buffer. Earlier rewards are multiplied by a gradually decreasing discount factor (e.g., 0.95) to make the robot pet focus more on recent gains. For each experience, calculate the update direction according to the policy gradient theorem: if an action brings a higher-than-average cumulative reward, increase the probability of that action being selected in the future; otherwise, decrease its probability. Under a linear Gaussian policy, this update direction has a concise form, involving only state features, action bias, and reward value. The update directions of all experiences are summed, multiplied by a small learning rate (e.g., 0.001 to 0.01), and added to the current policy parameters to complete one parameter iteration.

[0086] f. After completing the parameter update, clear the experience buffer and wait for the next update cycle to begin, repeating the above steps. As the number of iterations increases, the strategy parameters gradually converge, enabling the robot pet to select the optimal or near-optimal strategy parameter adjustment amount for the current user, thereby achieving personalized social distance adjustment.

[0087] After each action sampling, the policy smoothing and safety constraint unit first hard-limits the action vector to ensure that the adjustment amount does not exceed the preset range, and further restricts its rate of change to prevent dangerous behaviors caused by parameter mutations during the learning process. When a serious anomaly is detected (such as three consecutive collisions), the unit can forcibly terminate the current cycle, give a large negative reward, and roll back the policy parameters to the most recent stable version to ensure system safety.

[0088] The workflow diagram of the social strategy learning reconstruction module is as follows: Figure 10As shown, the interaction data acquisition and recognition unit first summarizes the data from the most recent N interactions. The user preference inference unit outputs the current state vector, including the interaction activity score, the average comfortable distance, the rejection sensitivity, and the user's average response delay to approach. Then, the strategy parameter generation and reconstruction unit generates candidate adjustment actions based on the current state and strategy parameters. The average action is obtained by a linear combination of state features, and the actual value is normally distributed around the average. Sampling randomness ensures exploration capability. Afterward, the strategy smoothing and safety constraint unit hard-limits the action vector and restricts the rate of change. If a serious anomaly is detected, the current cycle is forcibly terminated and rolled back to the most recent stable version. Otherwise, the strategy parameters are sent to the pet social distance decision module 430 and the motion control execution module 450. During the update cycle, the robot pet runs with new parameters. The interaction data acquisition and recognition unit continuously records feedback data such as active interaction duration, rejection duration, collision count, and distance deviation. Scalar rewards are calculated based on the feedback data. Active interaction duration contributes positive rewards, while rejection duration, collisions, and distance deviations incur negative penalties. The state, action, reward, and next state data are stored in the circular experience buffer. After accumulating M experiences, the unit can calculate the cumulative reward for each experience, calculate the update direction of each interaction state according to the policy gradient, accumulate the rewards, multiply them by the learning rate and add them to the current policy parameters to complete one parameter iteration. Finally, the buffer is cleared and the unit re-enters the next update cycle.

[0089] The motion control execution module 450 is connected to the social strategy learning and reconstruction module 440 and the exception handling and adjustment module 460, respectively, and is used to execute the actions issued by the decision-making instructions. The structure of the motion control execution module 450 is as follows: Figure 11 As shown. The limb movement unit and joint motion unit control the robot pet's body movement and posture, serving as the core device for expressing emotions. They can exhibit interactive actions such as slowly approaching, joyfully circling, and retreating in fear. The facial display and vocalization unit can change the pet's facial expressions and emit non-verbal sound effects, such as purring and whimpering, conveying emotions through changes in eye shape and brightness, as well as breathing light changes. The tactile feedback unit simulates a sense of life or provides tactile responses by changing the temperature and vibration of various body parts, mimicking the physical sensations of a real pet during intimate interactions. When the decision-making system issues commands, the motion control execution module 450 can coordinate the work of each unit to achieve a consistent goal.

[0090] The anomaly handling and adjustment module 460 is connected to the pet social distance decision module 430 and the motion control execution module 450, respectively, to enhance the system robustness in complex and dynamic real-world environments, ensuring that pet behavior is always safe, reliable, and in line with social expectations. Its structure is as follows: Figure 12As shown. The core processing flow of the exception handling adjustment module 460 can include at least the following: module self-check, assessing the impact of the exception on user experience, invoking preset security rules, executing the safest conservative degradation behavior, waiting for recovery, and recording exception information for learning. The specific mechanisms of each unit of the module are as follows: Sensor anomaly handling unit: Used to handle situations where millimeter-wave point cloud data is abnormal or lost. Once an anomaly occurs, the pet will immediately stop all movement behaviors that rely on precise positioning and switch to silent companionship behavior; if the sensor is abnormal for a long time, it will switch to standby mode in place and convey a "confused" prompt through sound and facial expressions.

[0091] User Interaction Anomaly Handling Unit: This unit is used to handle situations where the pet's behavior does not receive the user's expected response or when conflicting commands are received. If conflicting commands or atypical interaction signals are received from the user, the pet prioritizes the user's safety and comfort, executing conservative behaviors (such as maintaining distance and observing) until the signal becomes clear.

[0092] System Status Anomaly Handling Unit: Used to handle abnormal output from the decision module and power management issues. When the behavioral instructions output by the decision core are unreasonable, the pet calls the default behavior library; if a safety hazard is identified in the output behavior, the current behavior is forcibly interrupted and the "back up standby" command is executed; if the pet is in a low battery or abnormal power consumption state, the behavior energy consumption is gradually reduced, the pet moves autonomously to the vicinity of the charging dock, and gives a clear "needs charging" prompt through facial expressions and sounds.

[0093] Environmental anomaly handling unit: Used to prevent pets from being hindered or endangered in narrow or complex environments. When the system determines that the current space is not suitable for large-scale movement, it automatically switches to "stationary interaction" mode.

[0094] In this intelligent interactive system, the millimeter-wave sensing module 410 operates in various millimeter-wave bands, including 60GHz, 28GHz, and 77GHz, enabling human detection and micro-motion sensing. Different bands offer varying degrees of precision and detection range, allowing developers to select the appropriate band based on product requirements. The antenna structure is not limited to a 4×4 array; various specifications of two-dimensional antenna arrays, MIMO antennas, and conformal antennas can all achieve angle estimation and beam pointing. The transmitted waveform is not limited to frequency-modulated continuous waves; pulse modulation, stepped-frequency continuous waves, and OFDM radar waveforms can all be used to extract target distance, velocity, and angle information.

[0095] In this embodiment, the realization of human dynamic trend perception is not limited to millimeter-wave radar. The following wireless sensing technologies can replace some or all of the functions of the millimeter-wave sensing module 410 at different levels: (1) WiFi sensing scheme: By analyzing the changes in channel state information (CSI) caused by the reflection of WiFi signals by the human body, it is possible to sense whether the user exists and whether the user is active, and realize user activity detection (such as keeping quiet when sitting still and actively approaching when active), but it cannot obtain accurate distance, location and micro-limb movement information.

[0096] (2) UWB sensing scheme: By measuring the time of flight (ToF) of the signal, the precise distance, orientation and speed of movement between the robot pet and the user can be obtained, which can realize the approach / distance trend detection. However, its micro-Doppler detection capability is weaker than that of millimeter-wave radar, and it is difficult to capture subtle movements such as breathing and body tremors.

[0097] (3) ToF ranging scheme: The distance between the user and the robot pet and the approach / distance trend are obtained by measuring the time of flight of infrared light or laser. However, the depth map resolution is low, it cannot detect millimeter-level micro-movements and is easily affected by ambient light.

[0098] like Figure 13 As shown, the workflow of the intelligent interaction system of the robotic pet in this embodiment is as follows: The robotic pet starts the intelligent interaction system, the millimeter-wave sensing module 410 periodically emits FMCW signals, receives human body reflected echoes, and generates target point cloud data; the human body dynamic trend analysis module 420 performs static background filtering, target detection and continuous tracking on the point cloud data, and extracts features such as micro-motion energy, distance change trend, orientation stability, and high-frequency disturbance intensity; the pet social distance decision module 430 determines the user interaction state (low interaction / potential interaction / active interaction / repulsion) based on the dynamic trend features, and generates target distance, movement speed and multimodal behavior parameters; the motion control execution module 450 executes limb movement and joint movement. The robot pet uses facial displays and vocalizations, along with tactile feedback, to smoothly approach the desired distance and maintain its orientation towards the user. The social strategy learning and reconstruction module 440 periodically collects interaction feedback data (such as state duration and rejection probability), optimizes strategy parameters (such as desired distance and speed factor) through a strategy gradient algorithm, and transmits the optimized strategy parameters to the pet social distance decision module 430. The anomaly handling and adjustment module 460 continuously monitors sensor activity, user interaction, system status, and environmental anomalies, executes degradation behaviors according to preset safety rules, and records anomaly information for the social strategy learning and reconstruction module 440 to learn from. Through the above closed loop, the robot pet achieves active social distance adjustment based on millimeter-wave human dynamic trend perception.

[0099] This application addresses the problem of robotic pets lacking the ability to proactively adjust social distance in complex indoor environments, thus affecting user experience. By continuously sensing the relative distance and orientation between the user and the robot, as well as the user's micro-movement characteristics, and determining the interaction state based on the user's dynamic trends over a period of time, the system controls the robotic pet to adjust its companionship distance and behavior patterns in real time. For example, when a user is sitting quietly reading on the sofa, the system detects a stable human-computer distance and subtle changes in human movement. At this time, the robotic pet can maintain a suitable distance (e.g., 1.5 meters) and silently accompany the user without disturbing them. However, when the user closes the book or gets up, multiple high-frequency micro-movements occur within a short period, and the human-computer distance continuously shortens. The robotic pet can then recognize the user's desire to interact, proactively approaching the user and making affectionate interactions such as wagging its tail. This solution does not rely on a single instantaneous action recognition but rather on judging based on the user's continuous behavioral trends, thus making the human-computer interaction of the robotic pet more intelligent and natural.

[0100] The embodiments of this application have the following beneficial technical effects: (1) Behavioral trend perception based on millimeter-wave radar can avoid the risk of privacy leakage. This application uses a 60GHz millimeter-wave radar as the core sensing device, which captures the micro-Doppler perturbation features of the human body by emitting FMCW signals, instead of using a camera to collect image data. Compared with current visual recognition technology, this application completely avoids the impact of light changes on detection accuracy and achieves stable operation in all weather conditions. At the same time, since radar data does not contain visual privacy information such as faces and clothing, it fundamentally solves the concern of privacy leakage of robotic pets.

[0101] (2) Based on the prediction of micro-motion trends, the robot pet can proactively adjust its social distance and enhance the naturalness of its interaction. Currently, robot pets generally adopt a passive response mode in human-computer interaction, requiring the user to issue clear instructions (such as voice wake-up, gestures, or approach) before reacting, resulting in a mechanical interaction. This application uses dynamic trend feature extraction to match the expected interaction state, enabling the robot pet to clearly understand the user's interaction intention.

[0102] (3) Based on policy learning and reconstruction, it has the ability to provide personalized adaptive learning. Current technologies usually use fixed parameters, which cannot adapt to different user preferences. This application constructs a closed-loop mechanism of "state-action-reward" through policy gradient reinforcement learning. Based on the user's historical feedback (such as rejection probability and response delay), it dynamically reconstructs decision parameters (target distance, speed factor) to evolve exclusive interaction strategies that are adapted to specific users, which solves the problem that general strategies cannot meet personalized needs. As the usage time increases, the quality of companionship of the robot pet will continue to improve.

[0103] Reference Figure 14This application also discloses a robotic pet 1400, which includes a memory 1410, a processor 1420, and a computer program stored in the memory 1410 and executable on the processor 1420. When the processor 1420 executes the computer program, it can implement the robotic pet control method as described in any of the preceding embodiments applied to the robotic pet.

[0104] One embodiment of this application also provides a computer-readable storage medium storing computer-executable instructions for performing the control method for a robotic pet as provided in any embodiment of this application.

[0105] An embodiment of this application also provides a computer program product, including a computer program or computer instructions, which are stored in a computer-readable storage medium. The processor of the robot pet reads the computer program or computer instructions from the computer-readable storage medium and executes the computer program or computer instructions, causing the robot pet to perform the robot pet control method in any of the preceding embodiments.

[0106] The system architecture and application scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. Those skilled in the art will know that as the system architecture evolves and new application scenarios emerge, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0107] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0108] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0109] The terms “component,” “module,” “system,” etc., used in this specification are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process or execution thread, and components may be located on a single computer or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, or a network, such as the Internet interacting with other systems via signals).

Claims

1. A method for controlling a robotic pet, comprising: The distance information between the user and the robotic pet, the orientation information of the user relative to the robotic pet, and the user's micro-motion information are obtained, wherein the micro-motion information represents the user's non-displacement subtle limb movements or signal changes caused by physiological activities. Based on the distance information, the orientation information, and the micro-motion information, construct human dynamic trend feature information; The user's interaction state is identified based on the human body dynamic trend feature information, and the target behavior information of the robot pet is determined based on the interaction state. The interactive behavior of the robotic pet is controlled based on the target behavior information.

2. The control method according to claim 1, characterized in that, The step of constructing human dynamic trend feature information based on the distance information, the orientation information, and the micro-motion information includes: The micro-motion energy level used to characterize the user's activity intensity is determined based on the micro-motion information; Based on the distance information, distance trend information is determined to characterize the change in distance between the user and the robotic pet; The orientation stability used to characterize whether the user is continuously facing the robotic pet is determined based on the orientation information; Based on the micro-motion information, high-frequency disturbance intensity information is determined to characterize whether the user has a tendency to actively interact. Human dynamic trend feature information is obtained by combining the micro-motion energy level, the distance trend information, the orientation stability, and the high-frequency disturbance intensity information.

3. The control method according to claim 2, characterized in that, The step of determining the micro-motion energy level, used to characterize the user's activity intensity, based on the micro-motion information includes: The first average micro-motion energy within a preset time period is determined based on the micro-motion information. The micro-motion energy level used to characterize the user's activity intensity is determined based on the first average micro-motion energy.

4. The control method according to claim 2, characterized in that, The step of determining distance trend information to characterize the change in distance between the user and the robotic pet based on the distance information includes: The radial velocities of multiple users within a preset time period are determined based on the distance information; Distance trend information, characterizing the change in distance between the user and the robotic pet, is determined based on multiple user radial velocities.

5. The control method according to claim 2, characterized in that, The step of determining the orientation stability based on the orientation information to characterize whether the user is continuously facing the robotic pet includes: Determine the variance of the location information within a preset time period; The variance is used to determine the orientation stability that characterizes whether the user continues to face the robotic pet.

6. The control method according to claim 2, characterized in that, The step of determining the high-frequency disturbance intensity information used to characterize whether the user has a tendency to actively interact based on the micro-motion information includes: The instantaneous micro-motion energy is determined based on the micro-motion information; The high-frequency disturbance intensity information used to characterize whether a user has a tendency to actively interact is determined based on the instantaneous micro-motion energy.

7. The control method according to claim 2, characterized in that, The process of combining features based on the micro-motion energy level, the distance trend information, the orientation stability, and the high-frequency disturbance intensity information to obtain human dynamic trend feature information includes: Time smoothing is performed on the micro-motion energy level, the distance trend information, the orientation stability, and the high-frequency disturbance intensity information, respectively. Human body dynamic trend feature information is obtained by performing any one of the following processes on the current timestamp, the time-smoothed micro-motion energy level, the distance trend information, the orientation stability, and the high-frequency disturbance intensity information: feature splicing, weighted fusion, and model prediction.

8. The control method according to claim 2, characterized in that, The step of identifying the user's interaction state based on the human body dynamic trend feature information includes: The user's interaction status is obtained by identifying and mapping based on the micro-motion energy level, the distance trend information, the orientation stability, and the high-frequency disturbance intensity information.

9. The control method according to claim 1, characterized in that, Determining the target behavior information of the robotic pet based on the interaction state includes: The corresponding behavioral intent and direction of movement are determined based on the interaction state. Based on the stated behavioral intent and the stated direction of movement, the target behavioral information of the robotic pet is determined.

10. The control method according to claim 1, characterized in that, The acquisition of distance information between the user and the robotic pet, the user's orientation relative to the robotic pet, and the user's micro-movement information includes: Continuously transmit millimeter-wave sensing signals to the surrounding environment; Continuously receive the echo signal of the millimeter-wave sensing signal; Environmental point cloud data is generated based on the echo signal; The distance information between the user and the robotic pet, the user's orientation relative to the robotic pet, and the user's micro-motion information are extracted from the environmental point cloud data.

11. The control method according to claim 10, characterized in that, The step of extracting distance information between the user and the robotic pet, the user's orientation relative to the robotic pet, and the user's micro-motion information from the environmental point cloud data includes: Static background information is identified in the environmental point cloud data, and the static background information is filtered out to obtain dynamic object point cloud data. Initial user point cloud data is identified from the dynamic object point cloud data, and the distance information between the user and the robotic pet, the user's orientation information relative to the robotic pet, and the user's micro-motion information are determined based on the initial user point cloud data.

12. The control method according to claim 11, characterized in that, The step of identifying initial user point cloud data in the dynamic object point cloud data includes: Multiple target point cloud clusters were identified in the dynamic object point cloud data; Obtain the second average micro-motion energy for each of the target point cloud clusters; Among all the target point cloud clusters, the target point cloud clusters whose second average micro-motion energy is higher than a preset energy threshold during a preset time period and whose corresponding point cloud cluster size meets a preset size threshold are determined as the initial user point cloud data.

13. The control method according to claim 11, characterized in that, The step of determining the distance information between the user and the robotic pet, the user's orientation information relative to the robotic pet, and the user's micro-motion information based on the initial user point cloud data includes: The user's position relative to the robotic pet is determined based on the initial user point cloud data; Based on the azimuth information, the millimeter-wave sensing signal is emitted towards the user, and point cloud data of the target user is continuously acquired; The distance information between the user and the robotic pet, as well as the user's micro-movement information, are determined based on the target user's point cloud data.

14. The control method according to claim 1, characterized in that, The step of controlling the interactive behavior of the robotic pet based on the target behavior information includes: Generate corresponding motion parameters and interaction distance based on the target behavior information; The interactive behavior of the robotic pet is controlled based on the motion parameters and the interaction distance.

15. The control method according to claim 1, characterized in that, The method further includes: Obtain information about the interaction process between the robotic pet and the user; Determine the user's personalized interaction preference information based on the interaction process information; Candidate adjustment actions are generated based on the personalized interaction preference information; The control strategy parameters of the robotic pet are updated based on the candidate adjustment actions.

16. The control method according to claim 15, characterized in that, The personalized interaction preference information includes at least one of the following: Comfortable distance range, engagement level, response speed, and aversion sensitivity.

17. The control method according to claim 1, characterized in that, The method further includes at least one of the following: If the environmental point cloud data detected by the robotic pet is abnormal or lost, stop the robotic pet's movement behavior and switch the robotic pet's current behavior to silent companionship behavior. If the robot pet's sensor malfunctions for a preset duration, the robot pet's current behavior state will be switched to standby mode. If the robotic pet receives conflicting or non-interactive commands from the user, it shall remain in a standby state. If the target behavior information is determined to be unreasonable, the robot pet's interaction behavior is controlled by adopting the robot pet's default behavior. If it is determined that the target behavior information poses a security risk, the current behavior state of the robot pet will be switched to standby mode. If it is determined that the robot pet is currently in a low battery or abnormal power consumption state, control the robot pet to move to the vicinity of the charging dock and control the robot pet to issue a charging prompt message; If it is detected that the current space is not suitable for large-scale movement, the robot pet's current behavior state will be switched to a stationary interaction state.

18. A robotic pet, characterized in that, include: At least one processor; At least one memory for storing at least one program; The method for controlling a robotic pet as described in any one of claims 1 to 17 is implemented when at least one of the programs is executed by at least one of the processors.

19. A computer-readable storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are used to execute the control method for the robotic pet as described in any one of claims 1 to 17.

20. A computer program product, comprising a computer program or computer instructions, characterized in that, The computer program or the computer instructions are stored in a computer-readable storage medium. The processor of the robotic pet reads the computer program or the computer instructions from the computer-readable storage medium and executes the computer program or the computer instructions, causing the robotic pet to perform the control method of the robotic pet according to any one of claims 1 to 17.

Images