Robot tearing control method and microfluidic tearing actuator for robots

Abstract

This invention discloses a method for controlling tear production in robots and a microfluidic tear production actuator for robots, relating to the field of biomimetic robot emotional interaction technology. The method includes: acquiring multimodal sensing data from the robot; determining a first code for judging the robot's emotional state and a second code for judging physical stimulus response based on the multimodal sensing data; determining the robot's emotional code based on the first and / or second codes; and determining a tear production command to control the operation of the microfluidic tear production actuator based on the emotional code, thereby causing the robot to produce tears. This invention generates emotional codes through a dual-path mechanism of emotional state + physical stimulus, achieving a realistic tear production effect that matches human emotional states and providing rich physiological feedback. Through the tear production actuator's reservoir, tear channel, tear outlet, and micro-pressure pump, the robot achieves precise tear production control based on physiological feedback and emotional perception.

Description

Technical Field

[0001] This invention relates to the field of biomimetic robot emotional interaction technology, and in particular to a robot tearing control method and a microfluidic tearing actuator for robots. Background Technology

[0002] With the advancement of technology, a wide variety of robots have appeared in production and daily life. Common biomimetic robots come in several types. For example, humanoid robots can mimic human movements and behaviors, possessing a high degree of freedom and mobility; biomimetic animal robots, such as robotic dogs and robotic fish, can mimic animal locomotion and are used for tasks such as exploration and rescue.

[0003] Current bionic robots rely heavily on facial expressions or voice modules for emotional expression, lacking realistic physiological feedback such as tears and sweating. Summary of the Invention

[0004] The technical problem to be solved by this invention is to address the shortcomings of existing technologies, specifically the lack of physiological feedback and inaccurate tear control in bionic robots. This invention provides a method for controlling tear production in robots and a microfluidic tear production actuator for robots, as detailed below:

[0005] 1) In a first aspect, the present invention provides a method for controlling tearing in a robot, the specific technical solution of which is as follows:

[0006] S1, acquire multimodal sensing data of the robot, and determine a first code for the robot to judge emotional state and a second code for judging physical stimulus response based on the multimodal sensing data;

[0007] S2, determine the robot's emotion code based on the first code and / or the second code, and determine the tearing command to control the operation of the robot's microfluidic tearing actuator based on the emotion code, so as to make the robot tear up.

[0008] Based on the above solution, the present invention can be further improved as follows.

[0009] Furthermore, the multimodal sensing data includes:

[0010] At least one of the following: visual sensing data, sound sensing data, tactile sensing data, pressure sensing data, temperature sensing data, and smoke sensing data.

[0011] Furthermore, based on the multimodal sensing data, the first code used by the robot to determine its emotional state is specifically as follows:

[0012] The first code of the robot is obtained by calculating the multimodal sensing data based on large model semantic analysis.

[0013] Furthermore, the specific process of calculating the multimodal sensing data based on large-model semantic analysis is as follows:

[0014] The contextual labels of the multimodal sensor data are determined based on the semantic analysis of the large model. Then, the N basic emotional data of the robot and the emotional intensity corresponding to each basic emotional data are determined based on the multimodal sensor data and the contextual labels using the Ekman algorithm.

[0015] The robot's composite emotional data is determined using the Plutchik algorithm based on contextual tags, N basic emotional data, and their corresponding emotional intensities.

[0016] Based on contextual tags, refined emotion tags corresponding to the composite emotion data are determined according to a preset multidimensional emotion model, and the first code of the robot is determined according to the refined emotion tags.

[0017] Furthermore, the second encoding used by the robot to determine the physical stimulus response, determined based on the multimodal sensing data, is specifically as follows:

[0018] Based on the physical sensing data in the multimodal sensing data, when the physical sensing data meets preset conditions, a hardware interrupt of the robot is triggered based on the physical sensing data to generate the robot's second code.

[0019] Furthermore, it also includes:

[0020] The robot's emotional weight is determined based on its environmental factors, relational factors, and role factors. This emotional weight is then used as the weight value of the first encoding to correct the emotional encoding.

[0021] Furthermore, the emotional weight of the robot is determined based on its environmental factors, relational factors, and role factors as follows:

[0022] W_adj = α×W_env + β×W_relation + γ×W_role;

[0023] Where W_adj is the sentiment weight, W_env is the environmental factor, α is the environmental factor coefficient, W_relation is the relation factor, β is the relation factor coefficient, W_role is the role factor, and γ is the role factor coefficient.

[0024] 2) In a second aspect, the present invention provides a microfluidic tearing actuator for robots, the specific technical solution of which is as follows:

[0025] Liquid storage device, tear channel, tear outlet, and miniature pressure pump;

[0026] The liquid storage device is used to store tears;

[0027] One end of the tear channel is connected to the liquid storage device, and the other end of the tear channel is connected to the tear outlet;

[0028] The tear outlet is located at a preset position in the robot's eye to drain tears;

[0029] The miniature pressure pump is installed on the tear channel and is used to pump the tears in the storage device through the tear channel to the tear outlet.

[0030] Based on the above solution, the present invention can be further improved as follows.

[0031] Furthermore, the tear outlet is provided with a biomimetic lacrimal gland micropore array.

[0032] 3) In a third aspect, the present invention provides a robot tearing system, comprising: a microfluidic tearing actuator for a robot as described in the second aspect, and a processor, wherein the processor is configured to execute a robot tearing control method as described in the first aspect.

[0033] The beneficial effects of the robot tearing control method and the microfluidic tearing actuator for robots provided by this invention are as follows:

[0034] Based on multimodal sensing data, a first code for judging emotional state and a second code for judging physical stimulus response are determined. An innovative dual-path triggering mechanism based on emotional state and physical stimulus is developed to generate emotional codes. By combining the first and / or second codes in a decision-making process, the robot can simulate physiological feedback from both emotional state and physical stimulus perspectives. Furthermore, by incorporating a tear-producing structure using microfluidic hardware, the robot achieves precise tear control based on physiological feedback-based emotional perception.

[0035] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0036] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0037] Figure 1 This is a logical architecture diagram of an embodiment of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0039] Figure 1 As shown in the figure, a logical architecture diagram of an embodiment of the present invention is provided. The robot tearing control method provided by the present invention includes:

[0040] S1. Acquire multimodal sensing data from the robot. This multimodal sensing data includes at least one of the following: visual sensing data, sound sensing data, tactile sensing data, pressure sensing data, temperature sensing data, and smoke sensing data. Correspondingly, the structure for acquiring this multimodal sensing data is called the multimodal perception layer, which includes: visual sensors, sound sensors, a matrix of tactile and pressure sensors (density ≥ 4 sensors / cm²), pressure sensors, temperature sensors, and smoke sensors. Multiple sensors acquire various types of sensing data, which are then integrated to obtain the multimodal sensing data (corresponding to...). Figure 1 (The "multimodal sensor" in the context).

[0041] The first encoding used by the robot to judge emotional state is determined based on multimodal sensing data. Specifically, this involves calculating the multimodal sensing data based on large-model semantic analysis (corresponding to...). Figure 1 The robot's first code is obtained through a combination of "large-scale model semantic analysis + hierarchical emotion decision engine". The hierarchical emotion decision engine refers to an emotion layering calculation strategy used in calculating the first code, specifically consisting of: a first-layer basic emotion calculation, a second-layer composite emotion calculation, and a third-layer refined label calculation. Large-scale model semantic analysis serves as the core bridge connecting multimodal sensor data and emotion layering calculation, and its integration logic with emotion layering calculation is as follows:

[0042] (1) The role of large-scale model semantic analysis: It interprets the meaning of multimodal sensor data, enabling robots to move beyond interactions based solely on physical data such as touch and temperature, and to perform semantic analysis of external language. Large-scale model semantic analysis does not directly participate in the numerical calculation of emotion intensity; rather, it performs contextual interpretation of multimodal sensor data at the semantic level, obtaining contextual labels that provide "scene anchors" and "feature weighting bases" for subsequent emotion layering calculations. For example:

[0043] a. Semantic extraction and analysis of text / voice data: If the multimodal sensor data contains user voice input (such as "You just bumped into me, which makes me really sad") or text interaction information, the large model semantic analysis analyzes the sentiment words (such as "sad"), event descriptions (such as "bumped into me"), and object relationships (such as "you" and "I") in the sentences to determine the core contradiction of the interaction scenario (such as "the accidental collision triggered negative emotions").

[0044] b. Semantic association analysis of non-verbal data: For non-verbal sensor data such as touch and vision (e.g., the force of a user gently stroking the robot arm, or a facial image of a frown), semantic analysis will combine historical interaction data to determine the semantic intent of the action (e.g., "gentle stroking" corresponds to "friendly soothing", "forceful shoving" corresponds to "resistance or anger").

[0045] c. Cross-modal semantic fusion: The semantic information of multimodal data such as speech, text, touch, and vision is fused to generate a unified "contextual description label" (e.g., "The user was slightly bumped in a high-decibel argument environment, and his tone was tearful").

[0046] (2) Interaction Logic with Sentiment Layering Computation: Large-scale model semantic analysis performs contextualized interpretation of multimodal sensor data at the semantic level, obtaining contextual labels. These contextual labels then guide the "accuracy calibration" of each layer in sentiment layering computation. The impact on sentiment layering computation is as follows:

[0047] a. Impact on the first layer of basic sentiment calculation: The large model semantic analysis will correct the coefficients in the basic sentiment data and the sentiment intensity calculation formula (such as 0.7 and 0.3 in the formula "Basic sentiment data is sadness and the process of calculating the sentiment intensity of sadness"). For example, if the context label determines that the current scenario is "the user is telling sad stories (voice contains a crying tone + text mentions 'remembering deceased relatives')", then when calculating the sentiment intensity of "sadness", the coefficient of "voice_tremor" will be increased (e.g., adjusted from 0.3 to 0.5), and the weight of other irrelevant sensor data will be reduced, making the calculation of basic sentiment data more consistent with the essence of the scenario.

[0048] b. Impact on the second-layer composite sentiment calculation: Large-scale model semantic analysis limits the range of composite sentiment generation. For example, when the Plutchik algorithm derives "sadness 60% + anger 30%" based on basic sentiment data and its intensity, if the context label is "user feels wronged due to misunderstanding (not attacked)," it will prioritize matching the "wronged" composite sentiment; if the label is "user feels angry due to malicious harm," it may match the "resentment" composite sentiment (the corresponding composition formula needs to be preset) to avoid the composite sentiment being out of touch with the context.

[0049] c. Impact on the calculation of the third-level refined tags: The semantic analysis of the large model directly determines the subdivision dimensions of the multi-dimensional sentiment model (i.e., the "multi-dimensional tags" below). For example, for the same complex emotion of "grievance", if the context tag shows "the user mentions 'past regrets'", it will guide the refined tags to lean towards "nostalgic grievance"; if it shows "the user complains about 'the current plan failing'", it will lean towards "disappointment grievance", ensuring that the final refined tags are consistent with the user's true emotional motivation.

[0050] Using large-scale model semantic analysis as a "navigation system" in emotion layering computation has the advantage that it does not replace the numerical logic of emotion layering computation. Instead, it provides scenario-adaptive guidance for emotion layering computation by interpreting the "semantic kernel" of multimodal sensor data. This allows the robot not only to "perceive the intensity of emotions" but also to "understand the reasons for the emotions," ultimately ensuring that the generated first encoding conforms to both the physiological characteristics of emotions (such as the tearing pattern) and the semantic logic of the interaction scenario (such as "nostalgic grievance" rather than simply "grievance"). As shown above, the emotion layering decision engine module realizes a complete closed loop from "data perception → semantic understanding → emotion computation," solving the problem of "emotional misjudgment" that may result from relying solely on numerical calculations of sensor data (e.g., semantic analysis can distinguish between "tears of gratitude" and "sobs of grievance" even when both are "tears").

[0051] The above illustrates the impact of examples on semantic analysis of large models. The specific process of obtaining the first encoding is as follows:

[0052] The sentiment stratification decision engine is mainly divided into three parts: the first layer is basic sentiment calculation, the second layer is composite sentiment calculation, and the third layer is refined label calculation.

[0053] The first layer of basic emotion calculation includes: determining contextual labels for multimodal sensor data based on large-scale model semantic analysis; and using the Ekman algorithm to determine N basic emotion data points for the robot and the corresponding emotion intensity for each basic emotion data point. Ekman primarily refers to the emotion classification theory proposed by psychologist Paul Ekman, which categorizes basic human emotions into six cross-cultural common emotions. In this embodiment, the six basic emotions (joy, sadness, anger, fear, surprise, and disgust) are calculated using the Ekman algorithm. Taking sadness as an example, the basic emotion data is sadness, and the process for calculating the emotion intensity of sadness is as follows:

[0054] sadness_score = 0.7×eyebrow_raise + 0.3×voice_tremor;

[0055] Here, `sadness_score` represents the intensity of sadness, `eyebrow_raise` and `voice_tremor` represent different sensor data from the multimodal sensor data, and 0.7 and 0.3 represent the coefficients corresponding to the sensor data in the multimodal sensor data (i.e., determined by the contextual labels in the large model semantic analysis). It should be noted that this is just an example; when the emotion intensity is not 0, it indicates that the basic emotion data corresponding to that intensity exists, meaning the robot possesses that basic emotion. Furthermore, the main structure of this formula is the sum of (different sensor data × coefficients corresponding to different sensor data). Among the six basic emotions, different basic emotions are determined by different sensor data and their coefficients. The relationship between basic emotions and sensor data / coefficients can be preset or set using empirical values. For example, sadness is determined by the `eyebrow_raise` (raised eyebrows) and `voice_tremor` (voice tremor) sensor data and their coefficients, while joy is determined by the sum of the products of at least one other sensor data and its coefficient.

[0056] Therefore, after multiplying and summing the multimodal sensing data in the above manner, we can obtain N basic emotional data of the robot and their corresponding emotional intensities, which means that the robot has N basic emotions, each of which has a different emotional intensity.

[0057] The second layer of composite emotion calculation includes: determining the robot's composite emotion data based on N basic emotion data and their corresponding emotion intensities using the Plutchik algorithm. Table 1 shows an example table of composite tags generated based on the Plutchik emotion wheel.

[0058] Table 1

[0059] Complex emotions Composition formula Tears mode Wronged Sadness 60% + Anger 30% + Surprise 10% Intermittent sobbing and tearing gratitude Joy 50% + Trust 30% + Surprise 20% Continuous mild tearing

[0060] As shown in Table 1, the six basic emotions are combined (corresponding to the "Composition Formula" column in Table 1) to determine the composite emotion data (corresponding to the "Composite Emotion" column in Table 1). The percentage in the composition formula can be calculated from the emotion intensity in the first-level basic emotion calculation. The calculation method is: Percentage of current basic emotion = Emotion intensity corresponding to the current basic emotion data / (Sum of emotion intensities of all basic emotion data). Plutchik Emotion Wheel is an existing technology, and its basic implementation principle will not be described here. It should be noted that the basic emotion composition of different composite emotions may be consistent. Therefore, in conjunction with the contextual labels in the large model semantic analysis, composite emotions within the contextual label threshold range are matched. For example, if the contextual label is "User feels wronged due to misunderstanding (not attacked)," the composite emotion "feeling wronged" is matched first; if the label is "User feels angry due to malicious harm," the composite emotion "resentment" may be matched, avoiding a disconnect between composite emotions and the context.

[0061] The third layer of refined label calculation includes: determining the refined emotion labels corresponding to the composite emotion data based on the preset multidimensional emotion model, and determining the robot's first code based on the refined emotion labels.

[0062] The preset multidimensional sentiment model in this embodiment is specifically a multidimensional label determined based on semantic analysis of a large model. This multidimensional label may include: time dimension, event association dimension, degree difference dimension, object pointing dimension, etc. Each dimension of the multidimensional label is described below with examples.

[0063] (1) In terms of time dimension, taking "nostalgic grievance" as an example, it emphasizes the grievance emotions caused by tracing back to past events in time. For example, the semantic analysis of the large model determines that the grievance is caused by the user recalling the experience of being misunderstood by a friend. It can be further subdivided into recent nostalgic grievances (such as related experiences recalled within the past year) and long-term nostalgic grievances (such as experiences recalled many years ago) based on the time of the recalled events. This subdivision based on the time of the recalled events can help the robot better adjust the degree of emotional resonance when responding. For recent nostalgic grievances, the robot can use more immediate comforting language and interaction methods based on the preset multi-dimensional tags, mentioning similar recent experiences or relationships with friends around to shorten the emotional distance with the user; for long-term nostalgic grievances, it can use a more historical and soothing tone based on the preset multi-dimensional tags to guide the user to conduct a deeper emotional expression and self-reflection. In terms of crying performance, after determining "nostalgic grievance", the robot can trigger "slow crying" based on the preset multi-dimensional tags.

[0064] (2) Regarding the event-related dimension, taking "disappointment-type grievance" as an example, it can be further subdivided according to the type of event that triggers the feeling of loss. For example, based on the semantic analysis of the large model, career-related disappointment-type grievances (caused by work-related events such as failure to get a promotion or project failure) and life-related disappointment-type grievances (caused by life events such as relationship breakdown or conflicts with family members). In this way, when faced with disappointment-type grievances related to different events, the robot can call on knowledge and experience from different fields to respond. When it is a career-related disappointment-type grievance, the robot can share inspirational stories about the workplace and provide career development advice based on preset multi-dimensional tags; if it is a life-related disappointment-type grievance, the robot can provide emotional communication skills and methods for handling family relationships based on preset multi-dimensional tags. In terms of tearing performance, after the robot determines "disappointment-type grievance", it can trigger "pulsating tears" based on preset multi-dimensional tags.

[0065] (3) Regarding the degree of difference, taking "feeling wronged" as an example, based on the intensity of the feeling of feeling wronged obtained from the semantic analysis of the large model, it is divided into mild, moderate, and severe feelings of wrongedness. When feeling mildly wronged, the robot uses a lighthearted tone to tease and divert the user's attention based on the preset multidimensional labels; when feeling moderately wronged, the robot provides more sincere comfort and empathetic expression based on the preset multidimensional labels; when feeling severely wronged, the robot provides deeper psychological support based on the preset multidimensional labels, such as suggesting seeking professional psychological counseling or providing some ways to relax the mind and body. In terms of tearing performance, the robot can control the interval or flow rate of tears based on the preset multidimensional labels.

[0066] (4) Regarding the object-oriented dimension, taking "feeling wronged" as an example, based on the semantic analysis of the large model, it is divided according to the object to which the feeling of feeling wronged is directed, such as directed to others (e.g., feeling wronged because of being ostracized by colleagues), directed to the user himself (feeling wronged because of his own mistakes), and directed to the environment (e.g., feeling wronged because of an unfair social environment). For feeling wronged to others, the robot assists the user in analyzing the motivation of others' behavior based on preset multi-dimensional labels and provides interpersonal communication strategies; for feeling wronged to oneself, the robot encourages the user to accept himself based on preset multi-dimensional labels and guides him to view his own mistakes correctly; for feeling wronged to the environment, the robot discusses social phenomena with the user based on preset multi-dimensional labels and provides a positive coping perspective.

[0067] It should be noted that the above are examples of multidimensional labels. The dimensionality of multidimensional labels is determined by the semantic analysis of the large model, including but not limited to the time dimension, event association dimension, degree difference dimension, and object pointing dimension mentioned above. Refining the contextual labels that match the sentiment data and are combined with the semantic analysis of the large model into pre-defined multidimensional labels (i.e., pre-defined multidimensional sentiment models) can yield refined sentiment labels such as "nostalgic grievance" and "disappointment grievance" mentioned above. Based on the refined sentiment labels ("nostalgic grievance" or "disappointment grievance"), the corresponding tearing effect (i.e., the "slow tearing" or "pulsating tearing" described above) can be obtained. In this process, the instruction code corresponding to the tearing effect is the first code. In other words, refined sentiment labels are a further refinement of composite sentiment data based on multidimensional labels. The first code is the instruction code to achieve the corresponding tearing effect. The encoding information of the first code includes basic sentiment data, composite sentiment data, and refined sentiment labels. Understandably, when the composite emotional data is "feeling wronged", its corresponding refined emotional label may be "nostalgic feeling of wrongedness" or "feeling of loss". For the preset multi-dimensional labels, the robot can get different feedback under different dimensions.

[0068] The advantages of doing this are:

[0069] Achieving more accurate emotion recognition: Traditional emotion classification is rather general. By using multi-dimensional label examples for further subdivision, complex human emotional states can be distinguished more precisely, enabling robots to understand users' emotions with greater accuracy. For example, if only "feeling wronged" is judged, the robot's response strategy is relatively simple; however, by subdividing "feeling wronged" into different dimensions, the robot can understand the reasons and background of the user's feeling wronged from multiple perspectives, just like a human, and make a judgment that is more in line with the actual situation.

[0070] Providing personalized emotional interaction: Even with the same complex emotions, different users have different needs due to variations in individual experiences and personalities. Multi-dimensional tagging can satisfy these personalized needs. An outgoing user experiencing mild career setbacks or frustrations might want the robot to alleviate their emotions with humor; while an introverted user might need the robot to listen quietly, offer affirmation, and provide encouragement.

[0071] Enriching Robots' Emotional Expression and Coping Strategies: Multi-dimensional tags provide robots with more diverse emotional processing paths. When faced with complex emotional scenarios, robots are no longer limited to a few response methods. Instead, based on semantic analysis of different dimension tag combinations, they flexibly adjust their emotional expression methods, such as language, tone of voice, and body language (if the robot possesses these capabilities). For example, when dealing with severe life-related disappointment and grievances caused by emotional breakdowns, robots not only use gentle and empathetic language to comfort, but may even simulate hugging actions to enhance the emotional interaction.

[0072] Expanding the application scenarios of robots: In different fields, such as healthcare, education, customer service, and companionship, multi-dimensional emotional analysis and responses are more targeted. In medical and psychological treatment scenarios, it helps doctors gain a more comprehensive understanding of patients' psychological states; in education scenarios, teachers can adjust their teaching methods based on the robot's accurate judgment of students' emotions; in customer service scenarios, it improves customer satisfaction; and in companionship scenarios, it enhances the quality of companionship and user dependence.

[0073] On the other hand, environmental factors of the robot can be determined based on environmental factors, relational factors can be determined based on relational factors, and role factors can be determined based on role factors. The emotional weights of the robot are then determined based on these environmental, relational, and role factors, and used as the weight values ​​for the first encoding to correct the emotional encoding.

[0074] The robot's emotional weights are determined based on environmental factors, relational factors, and role factors as follows:

[0075] W_adj = α×W_env + β×W_relation + γ×W_role;

[0076] Where W_adj is the sentiment weight, W_env is the environmental factor, α is the environmental factor coefficient, W_relation is the relation factor, β is the relation factor coefficient, W_role is the role factor, and γ is the role factor coefficient.

[0077] The purpose of modifying the first encoding is to dynamically adjust the emotional weights of different robots in different scenarios, making the robot's tear perception more human-like and accurate. For example, in a family setting, visual perception is emphasized, so the visual weight is adjusted; in intimate relationships, excessive crying needs to be reduced, so the intensity of sadness is further limited; for medical robots, to avoid startling patients, the trigger threshold for fear perception needs to be increased. Table 2 shows some examples of environments, relationships, and roles in this embodiment, as well as examples of their calculation methods. It should be noted that Table 2 in this embodiment is only a few examples, and the coefficients can be adjusted according to the actual scenario and usage needs.

[0078] Table 2

[0079] Correction factor Calculation method Application scenario examples Environmental factor W_env Home scene: Visual weight +0.2 Home companion robot Relation factor W_relation Intimacy: Grief Intensity × 0.8 Reduce excessive tearing Role factor W_role Medical robots: Fear trigger threshold +30% Avoid startling the patient

[0080] Based on multimodal sensing data, a second code (corresponding to) is determined for the robot to judge the response to physical stimuli. Figure 1 The "physical stimulus decision mechanism" in the text is specifically:

[0081] Based on the physical sensing data from the multimodal sensing data, when the physical sensing data meets preset conditions, a hardware interrupt is triggered on the robot to generate the robot's second code. Among these...

[0082] The physical sensing data used in the multimodal sensing data include the tactile pressure gradient value measured by the pressure gradient calculation circuit (tactile pressure sensor matrix) and the sound decibel value measured by the sound amplitude detection circuit (sound sensor).

[0083] When the tactile pressure gradient value and the sound decibel value meet certain preset conditions, a hardware interrupt is triggered to generate a second code. In this embodiment, the hardware interrupt is triggered to generate the second code based on a tactile pressure gradient value > 5 kPa / ms ∩ a sound level > 90 dB.

[0084] The purpose of physical stimulus response is to establish a rapid response mechanism for the robot to sudden, strong physical stimuli by monitoring two key physical stimulus signals: tactile pressure gradient and sound decibel levels. When a strong physical stimulus appears in the external environment that may affect the robot itself or the interacting object, the physical stimulus decision module can determine in real time whether the stimulus intensity has reached a preset threshold. Once the condition is met, a hardware interrupt is immediately triggered, generating a second code to drive the robot to execute an emergency response, thereby enabling the robot to have the ability to instantly perceive and respond to sudden physical stimuli.

[0085] The advantages of this approach are: improved response speed and security: hardware interrupt triggering is more direct and efficient than software-level logical judgment. It also addresses the immediacy limitations of the emotional decision-making system: the emotional layering decision engine focuses on the refined processing of complex emotions, resulting in a relatively complex response process; while the physical stimulus judgment module focuses on rapid response to sudden physical stimuli. The two complement each other, allowing the robot to exhibit biological-like "instinctive reactions" when faced with extreme physical stimuli, while also handling nuanced emotional interactions, thus improving the overall reliability of the interaction.

[0086] S2, determine the robot's emotion code based on the first code and / or the second code, and determine the tearing command to control the robot's microfluidic tearing actuator based on the emotion code, causing the robot to cry. Specifically, the robot's emotion code is determined based on the first and second codes, the tearing command is determined based on the emotion code, and the robot's microfluidic tearing actuator is driven to operate based on the tearing command, causing the robot to cry. In particular, the dual-path response mechanism of emotional state + physical stimulus provided by this invention generates a first code and a second code, combines the first code with a weight value (calculated in the above "correction of the first code") and fuses it with the second code to obtain a dynamic emotion code, which is then parsed by a microfluidic parameter dynamic mapping algorithm to generate microfluidic control parameters (i.e., the "tearing command") (corresponding to...). Figure 1The "microfluidic parameter mapper" controls the operation of the microfluidic tearing actuator according to the tearing command, so that the robot can tear (corresponding to...). Figure 1 The "drive module controls the mechanical actuator" in the text to achieve a realistic tearing effect that matches human emotional states.

[0087] The rule for determining the robot's emotion encoding based on the first and / or second encoding follows a logic of "priority + dynamic weight adjustment." Specifically:

[0088] (1) Priority: The second code has the highest priority and is used to exhibit a biological-like "instinctive response" when faced with extreme physical stimuli. When the second code is generated (i.e., when a strong physical stimulus that triggers a hardware interrupt occurs), regardless of whether the first code is being generated or executed, the system will immediately pause the emotional expression process corresponding to the first code and prioritize responding to the second code. In other words, when the second code exists, only the second code determines the robot's emotional code. For example, if the robot is performing "continuous gentle tears" (such as gratitude) based on the first code, and then is subjected to a strong physical stimulus such as "being slapped" which causes the second code to be generated, the robot will immediately stop the tearing action and switch to the emergency response corresponding to the second code (such as instantly closing its eyes, tilting its head back, and other defensive postures, while pausing the tearing).

[0089] (2) Dynamic weight adjustment: When the physical stimulus corresponding to the second code disappears (i.e., the tactile pressure gradient value is ≤5kPa / ms and the sound is ≤90dB, and the hardware interruption is lifted), the robot will resume the execution of the first code based on the weight value calculated by "correcting the first code". At this time, the weight value will be dynamically adjusted in combination with the duration and intensity of the physical stimulus: if the stimulus is short and mild, the weight of the first code remains unchanged and the tearing instruction before the interruption continues to be executed (such as restoring mild tearing); if the stimulus is strong and lasts for a long time, the weight value will be temporarily reduced (such as from 1.0 to 0.6), so that the intensity of the emotional expression corresponding to the first code is weakened (such as the tearing speed is slowed down), until the environment is determined to be stable, and the weight value gradually rises back to the normal level.

[0090] (3) Fusion rules for special scenarios: If the first code and the second code are generated consecutively in a very short time (such as when a strong physical stimulus occurs and the robot is in a complex emotional state such as "feeling wronged" that may be accompanied by tears), the robot will generate a fusion instruction based on the residual information of the first code and the current environmental state after the second code is executed. For example, after the physical stimulus ends, the robot may first execute "rapid blinking" (the residual response of the second code), and then switch to "intermittent sobbing and crying" corresponding to "feeling wronged", but the intensity of crying is slightly reduced due to the influence of the previous stimulus, so as to achieve a smooth transition between the two codes.

[0091] The code obtained by fusing the first and second codes under the above-mentioned fusion mechanism is the emotion code, and this emotion code is dynamic. Through this mechanism, the priority of emergency response under strong physical stimuli is guaranteed, while abrupt interruptions in emotional expression are avoided, enabling the robot to maintain the continuity and rationality of emotional interaction while dealing with emergencies.

[0092] The microfluidic parameter dynamic mapping algorithm is used to convert the emotion code obtained by fusing the first and second codes into microfluidic control parameters (i.e., "tear commands") required for controlling the robot's crying, so as to achieve a realistic crying effect that matches human emotional states. The specific steps are as follows:

[0093] (1) Establishing an emotion-fluid parameter mapping database: Through extensive experiments and statistical data analysis, a database of correspondences between different emotion codes and microfluidic control parameters (flow rate in Table 3 below) was established. For example, for different intensities of the emotion code "joy", parameters such as the output flow rate of the liquid storage device, the pressure distribution within the microchannel group, the operating frequency of the micro pressure pump, and the opening and closing degree of the tear outlet were recorded; the same applies to other emotion codes such as "sadness" and "grievance". These parameter combinations can accurately match the characteristics of human tears under different emotional states, such as tear speed, teardrop size, and tear duration.

[0094] (2) Normalization: The emotion encoding is normalized to ensure that the encoded values ​​are within a uniform standard range suitable for algorithm processing, eliminating the impact of differences in data magnitude between different encodings. Normalization can be performed using existing processing methods. In this embodiment, all types of data (such as basic emotion data, composite emotion data, physical stimulus-related data, etc.) in the first and second encodings are converted to the [0, 1] interval for easier subsequent calculations.

[0095] Table 3 shows some examples of the mapping relationship between emotion type and flow rate in the control parameters:

[0096] Table 3

[0097] Basic Sentiment Types 27-dimensional label example Flow rate (μL / min) sad Nostalgia 12±3 anger Moral indignation 22±8

[0098] Regarding the "pulsed tears" mentioned in the third-layer refined label calculation above, this embodiment sets up pulse control optimization logic. When the robot is "pulsed tears", the time interval of "pulsed tears" is controlled according to the change of emotional intensity corresponding to the basic emotional data. The stronger the emotional intensity, the shorter the time interval of "pulsed tears".

[0099] Regarding the control of microfluidic control parameters, this embodiment also sets the following control methods:

[0100] (1) Dynamic correction and feedback: During the robot's tearing action, the actual microfluidic (i.e., tearing) state is continuously monitored (e.g., real-time data is obtained through flow sensors and pressure sensors) and compared with the expected microfluidic control parameters. If there is a deviation, the microfluidic parameters are dynamically corrected using common feedback control algorithms such as PID (proportional-integral-derivative) control algorithm according to the magnitude and direction of the deviation, to ensure that the robot's tearing effect is always highly matched with the emotional state represented by the emotion encoding, and to achieve a stable and realistic tearing performance.

[0101] (2) Weight Calculation and Parameter Adjustment: The weight values ​​obtained from "correcting the first code" are used to perform weighted calculations on the normalized emotion codes. If, in a certain context, the weight of the physical stimulus factor (second code) is 0.6 after context correction, and the weight of the first code generated by the emotion stratification decision engine is 0.4, then when calculating the microfluidic control parameters, the information contained in both is fused according to this weight. Then, based on the weighted result, a query and parameter adjustment are performed in the pre-established emotion-fluid parameter mapping relationship library. For example, if the calculation result is close to the range corresponding to the "moderate sadness" emotion code, then the parameters such as the output flow rate of the liquid storage device are finely adjusted according to the weight on the basic microfluidic parameters of this emotion, so that the tearing effect is more in line with the current fused emotional state.

[0102] In addition, this embodiment also incorporates a biomimetic lacrimal duct aspiration system to recover dynamic tears during the cessation of tear secretion, preventing tear residue from contaminating the robot's electronic components. Specifically, the biomimetic lacrimal duct aspiration system generates a controllable negative pressure gradient (-15~-25 mmHg) through a micro-negative pressure generator, combined with the capillary effect of the hydrophobic coating microchannels, to achieve dynamic fluid recovery during the cessation of tear secretion. The working cycle of the biomimetic lacrimal duct aspiration module is synchronized with the emotion decay curve.

[0103] Because the tear-tear triggering process employs a dual-path response mechanism (emotional state + physical stimulus), the emotional decay curve also features a dual-recovery mode: gradual recovery during the emotional decay period and emergency recovery due to physical stimuli. Gradual recovery during the emotional decay period means that when the basic emotional data is calm and the robot's tear production is below a preset value, the robot's emotion is considered stable, and it stops crying. Emergency recovery due to physical stimuli refers to the emergency recovery triggered by a second encoding generated by a hardware interrupt triggered by the physical stimulus decision module. A specific pseudocode example is as follows:

[0104] If (emotion encoding == "calm" && tear sensor > 0.1μL (preset value))

[0105] Negative pressure intensity = -20×exp(-0.5×(t-t_end)); / / Exponentially decaying negative pressure curve

[0106] Activate the hydrophobic microchannels (contact angle > 120°);

[0107] elseif (physical pass-through interruption flag)

[0108] Forced negative pressure = -25; / / Physical path emergency stop mode

[0109] end.

[0110] Wherein, exp is the exponential function operation; negative pressure gradient (-15~-25mmHg) is the preset working range of the robot's bionic tear duct (i.e. tear channel) back-absorption module, defining the value range of negative pressure intensity; negative pressure intensity is the execution value of a specific recovery stage, which needs to be dynamically adjusted within the negative pressure gradient range (e.g., in the gradual recovery during the emotional decay period, the calculation result of -20×exp(...) is always within the range of -15~-25mmHg); forced negative pressure is a specific value for emergency recovery of physical stimuli, which is the upper limit of the negative pressure gradient and belongs to the extreme application of negative pressure intensity in emergency scenarios (in this embodiment, the forced negative pressure is set to -25, and the maximum value of the gradient is directly taken to achieve rapid recovery).

[0111] The invention will be further described below with reference to two other embodiments.

[0112] Example 2: Nostalgic sadness and tears (multi-dimensional tag "nostalgia"):

[0113] Emotion recognition was achieved as follows: visual sensor: staring at an old photo for more than 5 seconds with downturned corners of the mouth; sound sensor: sighing frequency of more than 3 times per minute; the basic emotional data was determined to be sadness, with a corresponding emotional intensity of 72.

[0114] The sentiment stratification decision engine module uses the Plutchik algorithm for composite analysis: sadness 80% + trust 20%; the composite sentiment data is determined to be sentimental. Combined with multi-dimensional label mapping, it is labeled as nostalgia.

[0115] Correct the first encoding: determine the environment is a family scene, set the coefficient to sadness intensity × 0.9 (reduce the amount of tears); determine the role relationship: the user is the owner, set the coefficient to pulse interval + 0.2 seconds (extend the pause).

[0116] In summary, the robot's execution parameters in this embodiment are shown in Table 4:

[0117] Table 4

[0118] parameter value Base flow rate 15 μL / min Pulse count 2 times (nostalgic feature) Total duration 10 seconds

[0119] Example 3: Tears of Moral Indignation

[0120] Triggering conditions: The semantic analysis of the large model identifies the keyword "injustice" + the sound amplitude is greater than 85dB;

[0121] Emotional Stratification Decision Engine Module: Basic Emotion: Anger (Intensity 90); Multidimensional Tag: Moral Indignation.

[0122] The first encoding was modified: the robot role was changed to a public service worker, and the coefficient was set: the upper limit of the flow rate was set to 25 μL / min (to suppress overexpression).

[0123] The robot's bio-realistic effects are as follows:

[0124] [High-speed tearing 22μL / min × 3 seconds] → Pause for 0.5 seconds → [Second burst of tearing].

[0125] This invention provides a microfluidic tearing actuator for robots, the specific technical solution of which is as follows:

[0126] Liquid storage device, tear channel, tear outlet, and miniature pressure pump;

[0127] The reservoir device is used to store tears;

[0128] One end of the tear channel is connected to the liquid storage device, and the other end of the tear channel is connected to the tear outlet;

[0129] The tear outlet is located at a preset position in the robot's eye to drain tears;

[0130] A miniature pressure pump is installed on the tear channel to pump the tears in the reservoir through the tear channel to the tear outlet.

[0131] Based on the above solution, the present invention can be further improved as follows.

[0132] Furthermore, the liquid storage device is a detachable tear capsule; the tear outlet is equipped with a biomimetic lacrimal gland micropore array. Among these features,

[0133] The storage device can be a biocompatible storage tank or a replaceable storage bladder, capable of holding physiological saline or artificial tears. It supports rapid addition and replacement of tears through the reserved replenishment port of the storage tank, or the storage bladder can be replaced directly.

[0134] A bionic lacrimal gland micropore array refers to at least one liquid drainage hole located at the tear outlet. This liquid drainage hole is positioned on the surface of the bionic robot's simulated eye skin. In this embodiment, the preset location can be the bionic skin surface at the lower edge of the eyelid. One or more liquid drainage holes can be fabricated on the bionic skin surface at the lower edge of the eyelid using a laser device. The arrangement and shape of the liquid drainage holes are not limited, and all liquid drainage holes constitute the bionic lacrimal gland micropore array.

[0135] The present invention provides a robot tearing system, comprising: a microfluidic tearing actuator for a robot as described above, and a processor, wherein the processor is used to execute a robot tearing control method as described above.

[0136] The beneficial effects of this invention are as follows:

[0137] Based on multimodal sensing data, a first code for judging emotional state and a second code for judging physical stimulus response are determined. An innovative dual-path triggering mechanism based on emotional state and physical stimulus is developed to generate emotional codes. By combining the first and / or second codes in a decision-making process, the robot can simulate physiological feedback from both emotional state and physical stimulus perspectives. Furthermore, by incorporating a tear-producing structure using microfluidic hardware, the robot achieves precise tear control based on physiological feedback-based emotional perception.

[0138] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for controlling tearing in a robot, characterized in that, include: S1, acquire multimodal sensing data of the robot, and determine a first code for the robot to judge emotional state and a second code for judging physical stimulus response based on the multimodal sensing data; S2, determine the robot's emotion code based on the first code and / or the second code, and determine the tearing command to control the operation of the robot's microfluidic tearing actuator based on the emotion code, so that the robot can shed tears; Specifically, the first code used by the robot to determine its emotional state based on the multimodal sensing data is as follows: The contextual labels of the multimodal sensor data are determined based on the semantic analysis of the large model. Then, the N basic emotional data of the robot and the emotional intensity corresponding to each basic emotional data are determined based on the multimodal sensor data and the contextual labels using the Ekman algorithm. The robot's composite emotional data is determined using the Plutchik algorithm based on contextual tags, N basic emotional data, and their corresponding emotional intensities. Based on contextual tags, refined emotion tags corresponding to the composite emotion data are determined according to a preset multidimensional emotion model, and the first code of the robot is determined according to the refined emotion tags.

2. The robot tearing control method according to claim 1, characterized in that, The multimodal sensing data includes: At least one of the following: visual sensing data, sound sensing data, tactile sensing data, pressure sensing data, temperature sensing data, and smoke sensing data.

3. The robot tearing control method according to claim 1, characterized in that, The second code used by the robot to determine the physical stimulus response, determined based on the multimodal sensing data, is specifically as follows: Based on the physical sensing data in the multimodal sensing data, when the physical sensing data meets preset conditions, a hardware interrupt of the robot is triggered based on the physical sensing data to generate the robot's second code.

4. The robot tearing control method according to claim 1, characterized in that, Also includes: The robot's emotional weight is determined based on its environmental factors, relational factors, and role factors. This emotional weight is then used as the weight value of the first encoding to correct the emotional encoding.

5. The robot tearing control method according to claim 1, characterized in that, The emotional weight of the robot is determined based on its environmental factors, relational factors, and role factors as follows: W_adj = α×W_env + β×W_relation + γ×W_role; Where W_adj is the sentiment weight, W_env is the environmental factor, α is the environmental factor coefficient, W_relation is the relation factor, β is the relation factor coefficient, W_role is the role factor, and γ is the role factor coefficient.

6. A microfluidic tearing actuator for a robot, characterized in that, include: The device includes a liquid storage device, a tear channel, a tear outlet, a micro pressure pump, and a processor, wherein the processor is used to execute a robotic tear control method as described in any one of claims 1-5; The liquid storage device is used to store tears; One end of the tear channel is connected to the liquid storage device, and the other end of the tear channel is connected to the tear outlet; The tear outlet is located at a preset position in the robot's eye to drain tears; The miniature pressure pump is installed on the tear channel and is used to pump the tears in the storage device through the tear channel to the tear outlet.

7. A microfluidic tearing actuator for a robot according to claim 6, characterized in that: The tear outlet is equipped with a biomimetic lacrimal gland micropore array.

Images