Immersive interactive training system and method for computer-simulated psychotherapy scenarios

Abstract

The application relates to the field of psychological treatment, and discloses an immersive interactive training system for computer simulation of a psychological treatment scene, which comprises: a storage unit for storing a core belief vector, an emotional state vector and a simulated physiological state vector representing the internal state of a virtual visitor; a user intervention analysis module for receiving intervention input of a user and analyzing the intervention input into an intervention vector; and a state evolution module connected with the storage unit and the user intervention analysis module, for calculating and updating the core belief vector, the emotional state vector and the simulated physiological state vector at a current time according to the intervention vector and the three state vectors at a previous time stored in the storage unit. By setting the core belief vector, the emotional state vector and the simulated physiological state vector, the internal state of the virtual visitor is mathematically represented from three different levels of dimensions, and the defects of traditional virtual human models, such as over-flattening and singleness in psychological depiction, are overcome.

Description

Technical Field

[0001] This invention relates to the field of psychotherapy, specifically to an immersive interactive training system and method for computer-simulated psychotherapy scenarios. Background Technology

[0002] This invention relates to the fields of computer virtual reality and psychotherapy technology, specifically to an immersive interactive training system and method for computer-simulated psychotherapy scenarios applied to skills training of mental health professionals. This type of system aims to provide a safe and efficient practice platform for mental health counselors, therapists, and students in related fields by constructing a controllable and repeatable simulated environment, thereby training and improving their professional skills in clinical diagnosis, intervention, and communication.

[0003] However, in the process of realizing this invention, the inventors discovered that existing technologies still face several unresolved technical bottlenecks in constructing virtual clients capable of deeply simulating the complex psychological dynamics of humans. Many existing systems rely excessively on large language models for role-playing and dialogue generation, which often results in virtual clients lacking a stable and calculable internal psychological core. Their behavior thus appears superficial, prone to forgetting or contradicting their role settings over long-term interactions, and lacking internal consistency in their response logic. Simultaneously, the reaction patterns of these systems are typically rigid; their response logic to specific interventions is pre-set and fixed, failing to take into account the virtual client's own cognitive structure and immediate physiological and psychological state, leading to a lack of dynamic realism in the interaction. A deeper problem lies in the difficulty of simulating the long-term and cumulative effects of the therapeutic process; virtual clients rarely demonstrate the deep evolution and growth of their core beliefs in cross-conversational interactions, limiting the value of training in promoting fundamental changes in related skills. In addition, a particularly prominent shortcoming is that most current models follow smooth and continuous state evolution, which makes them inadequate for simulating sudden and nonlinear key clinical events such as flashbacks of psychological trauma, greatly reducing the complexity and challenge of the training scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an immersive interactive training system and method for computer-simulated psychotherapy scenarios. It solves the problems of inconsistent behavioral logic and rigid response patterns caused by the lack of a calculable internal psychological kernel in virtual visitors, as well as the problems that existing models cannot reflect the long-term evolution of core beliefs and are difficult to simulate sudden nonlinear psychological events such as traumatic flashbacks.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an immersive interactive training system for computer-simulated psychotherapy scenarios, comprising: The storage unit is used to store the core belief vector, emotional state vector, and simulated physiological state vector that represent the internal state of the virtual visitor. The user intervention parsing module is used to receive user intervention input and parse it into intervention vectors. The state evolution module, connected to the storage unit and the user intervention parsing module, is used to calculate and update the core belief vector, emotional state vector and simulated physiological state vector at the current moment based on the intervention vector and the three state vectors stored in the storage unit at the previous moment. The response generation module, connected to the state evolution module, is used to generate a system response based on the three updated state vectors at the current moment using a large language model.

[0006] Preferably, the state evolution module is specifically used for: A state transition matrix is ​​generated, and the current emotional state vector is calculated based on the intervention vector, the emotional state vector of the previous moment, and the state transition matrix.

[0007] Preferably, the state evolution module is configured as follows: The state transition matrix is ​​dynamically generated based on the current core belief vector and simulated physiological state vector using a preset mapping function.

[0008] Preferably, the state evolution module is further specifically used for: Based on the emotional state vector of the previous moment and the preset emotional-physiological mapping matrix, the simulated physiological state vector of the current moment is updated and obtained.

[0009] Preferably, the state evolution module is further specifically used for: When the difference between the current emotional state vector and the previous emotional state vector is detected to meet the preset positive progress pattern, the core belief vector is updated.

[0010] Preferably, the system further includes: The burst memory association module is used to store at least one triplet of data containing a text trigger, memory content, and state impulse vector.

[0011] Preferably, the state evolution module is further specifically used for: When the user intervention parsing module identifies the text trigger, the state impact vector is applied to the update calculation of the emotional state vector and the simulated physiological state vector.

[0012] Preferably, the user intervention parsing module integrates a natural language processing model to convert the user's text input into the intervention vector.

[0013] Preferably, the response generation module is specifically used for: The updated core belief vector, emotional state vector, and simulated physiological state vector at the current moment are combined into an instructional prompt and input into the large language model to generate the system response.

[0014] A method based on the above-mentioned interactive training system includes the following steps: Obtain user intervention input and parse it into intervention vectors; Based on the intervention vector, and the pre-stored core belief vector, emotional state vector, and simulated physiological state vector from the previous moment, calculate and update the core belief vector, emotional state vector, and simulated physiological state vector for the current moment. Based on the updated core belief vector, emotional state vector, and simulated physiological state vector at the current moment, a system response is generated through a large language model.

[0015] This invention provides an immersive interactive training system and method for computer-simulated psychotherapy scenarios. It has the following beneficial effects: 1. This invention mathematically represents the internal state of a virtual visitor from three different levels and with different rates of change—cognitive, emotional, and physiological—by setting core belief vectors, emotional state vectors, and simulated physiological state vectors. This overcomes the shortcomings of traditional virtual human models, which are overly flat and simplistic in their psychological portrayal. This multi-layered coupled representation allows the virtual visitor to not only exhibit momentary emotional fluctuations but also maintain a long-term stable core cognitive structure consistent with its background setting. Its behavior and responses thus possess unprecedented psychological depth and internal logic, greatly enhancing the realism of the simulated interaction.

[0016] 2. This invention designs a dynamic state transition matrix generation mechanism jointly determined by a core belief vector and a simulated physiological state vector. This makes the emotional response pattern of virtual visitors to external interventions no longer fixed, but rather subject to real-time adjustment based on their deep cognition and current physiological arousal level. This means that, in response to the same intervention statement from a user, virtual visitors in different psychological and physiological states will produce differentiated, context-appropriate responses. This effectively avoids the mechanical and repetitive response patterns of traditional question-and-answer systems, making the interaction process more dynamic, natural, and varied.

[0017] 3. This invention introduces a sudden memory association module containing text triggers, memory content, and state impact vectors, enabling the system to simulate nonlinear psychological phenomena common in real-world psychological activities, such as traumatic flashbacks or emotional outbursts triggered by specific cues. When the user's speech touches a preset sensitive area, this module can drive a dramatic, abrupt shift in the virtual visitor's internal state and generate a language response related to specific memory content. This provides trainees with a simulated training scenario for handling highly stressful and crisis situations, which is impossible to achieve with traditional smooth evolutionary models.

[0018] 4. This invention fundamentally solves the common problems of logical inconsistency, forgotten role settings, and uncontrollable behavior when directly applying large language models for role-playing by functionally decoupling the internal core model responsible for state calculation from the large language model responsible for language generation, and using the latter as a "language shell" strictly constrained by the former. This design ensures that every utterance of the virtual visitor has a traceable and interpretable internal state vector as its basis, guaranteeing a high degree of consistency and predictability of interactive behavior throughout the process, and significantly improving the system's reliability and rigor as a training tool.

[0019] 5. This invention successfully simulates the long-term and gradual nature of core cognitive change in psychotherapy by designing a slow-change update mechanism for the core belief vector that makes minor adjustments only when a positive progress pattern is detected. This mechanism allows the virtual client's fundamental "personality" or "worldview" to undergo beneficial and cumulative evolution with continuous and effective positive interactions. This enables the system to not only simulate a single session but also to demonstrate a therapeutic process with a growth arc that spans multiple sessions, providing trainees with a valuable platform to observe and learn how to facilitate profound change. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a diagram illustrating the architecture of the user intervention parsing module of the present invention. Figure 3 This is a diagram of the storage unit architecture of the present invention; Figure 4 This is a diagram of the architecture of the sudden memory association module of the present invention; Figure 5 This is a diagram of the response generation module architecture of the present invention; Figure 6 This is a diagram of the state evolution module architecture of the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: Please see the appendix Figure 1 -Appendix Figure 6 This invention provides an immersive interactive training system for computer-simulated psychotherapy scenarios, comprising: User Intervention Analysis Module In this embodiment, the user intervention parsing module is the core hub of the system of the present invention, enabling human-computer interaction and transforming the qualitative language input of external users into internally computable quantitative data. The fundamental purpose of this module is to act as a bridge between the external natural language world and the system's internal mathematical mental model, ensuring the accuracy and effectiveness of subsequent state evolution calculations.

[0023] In a preferred embodiment, the module is configured to process user voice input, and its internal components may be further divided into a voice recognition unit and a treatment technology parsing unit.

[0024] The speech recognition unit is responsible for receiving the continuous speech stream from the user, captured by the microphone audio acquisition device. This unit can employ mature Automatic Speech Recognition (ASR) technology to convert the speech stream into a text string in real-time or near real-time. To ensure smooth interaction, a streaming speech recognition scheme is preferred to reduce the latency between the user's speech and the system's response. The output of this unit, i.e., the text string, forms the basis for subsequent semantic analysis.

[0025] The therapeutic technology parsing unit receives the text string generated by the speech recognition unit and performs in-depth semantic analysis and intent recognition on it. This process is not a simple keyword matching, but rather, based on an understanding of professional knowledge in the field of psychotherapy, it maps unstructured natural language into a standardized, multi-dimensional feature space.

[0026] To implement this mapping, the system predefines a set containing... This is a category library of therapeutic techniques. This library is summarized and categorized based on mainstream psychotherapy theories (cognitive behavioral therapy, motivational interviewing), and includes a series of therapeutic techniques with clearly defined operational aspects, such as "empathic understanding and feedback," "open-ended questioning," "cognitive restructuring challenges," and "summarizing and clarifying." This predefined category library provides a unified and professionally significant benchmark for subsequent quantitative analysis.

[0027] Building upon this foundation, the core of this treatment technique analysis unit is a pre-trained multi-label text classification model. This model is trained on a professionally labeled corpus, where each sample is a therapist's utterance labeled with one or more treatment technique categories. Through this training process, the model learns the complex mapping relationship between natural language expressions and specific treatment techniques.

[0028] During system runtime, when this unit receives a text string, it uses this pre-trained model for inference. The model's output is the core output of this module. An intervention vector of dimension, at time step Time is recorded as Its mathematical expression is: ; In this vector, each component The value range of this component is normalized to the interval [0,1]. The value of this component represents the number of occurrences of the first element in the current user input text as evaluated by the model. A confidence score or intensity level for a predefined treatment technique. A higher confidence score indicates a higher level of confidence. Value (assumption) The term "empathic understanding and feedback" indicates that the user's statement was highly recognized by the model as an empathic behavior. This vectorized representation, compared to simple binary classification, can capture the various technical components and their relative strengths contained in a single user intervention more delicately and comprehensively, thus providing richer and more accurate input information for subsequent state evolution calculations.

[0029] Ultimately, the user intervention analysis module will generate intervention vectors. This vector is passed to the state evolution module. It is the core external driving force triggering changes in the virtual visitor's internal state and will directly participate in the subsequent state transition matrix. In applied computing, the emotional state vector of virtual visitors The update plays a decisive role, thus forming an indispensable link in the causal chain of the entire system.

[0030] Storage unit and state vector In this embodiment, the storage unit and state vectors constitute the core data carrier of the virtual visitor's internal psychological world in the system of the present invention, and are the fundamental basis for realizing its dynamism, consistency, and interpretability. The storage unit is not a simple static data warehouse, but rather a set of mathematical descriptions representing the multi-level psychological states of the virtual visitor, namely, state vectors, used to maintain and update in real time. These vectors are assigned initial values ​​during system initialization and are iteratively updated by the state evolution module in each interaction round, thus forming a continuous, memory-based "stream of consciousness" of the virtual visitor.

[0031] Specifically, the internal state is defined by three distinct yet deeply coupled state vectors, which depict the psychological landscape of the virtual visitor from different dimensions.

[0032] First, in order to characterize the deep-seated and relatively stable personality traits and cognitive patterns of virtual visitors, this invention defines a core belief vector (BeliefVector). This vector is a A real vector of dimension 1 has the following mathematical expression: ; Each component of the vector The value range of is normalized to [0,1], representing the strength or centrality of a predefined core belief ("I am worthless", "others are untrustworthy") in the virtual visitor's cognitive structure. A higher ... The value signifies that the corresponding belief is deeply ingrained and difficult to sway by external information. This vector is slowly variable, updating much less frequently than other state vectors, typically undergoing only minor adjustments when the system determines that the interaction has made significant and positive progress. This design aims to simulate the difficulty and gradual nature of changing core beliefs in real psychotherapy. In the system, It plays a fundamental regulatory role. As one of the key inputs for calculating the state transition matrix in the state evolution module, it fundamentally determines the inherent tendencies and response patterns of virtual visitors in response to external interventions.

[0033] Secondly, in order to capture the rapidly changing emotional responses of virtual visitors during the interaction, this invention defines an Emotion Vector. This vector is a A real vector of dimension 1 has the following mathematical expression: ; Each component of the vector The value range is also within the [0,1] interval, quantitatively representing the intensity of a predefined basic emotion (anxiety, sadness, anger, hope) at the current moment. Unlike the stability of the core belief vector, the emotional state vector is rapidly changing, recalculated in each interaction round based on the user's intervention. This vector is a direct product of the core calculation in the state evolution module, and also serves as a key information source driving physiological state changes and generating the final language response. Its real-time changes intuitively reflect the effective impact of user intervention on the virtual visitor's immediate mood.

[0034] Furthermore, in order to construct a more complete and realistic psychophysiological unity, this invention introduces a simulated physiological state vector (Physiological Vector). This vector is a A real vector of dimension 1 has the following mathematical expression: ; Each component in the vector The value of takes the range [0,1], representing a simulated level of a normalized physiological indicator related to emotion (heart rate, skin conductance, respiratory rate). The introduction of this vector is not merely as an auxiliary output of emotion; its key role is to form a dynamic feedback loop with the cognitive level. On the one hand, The state is influenced by the emotional state of the previous moment. The influence it has on the physiological expression of emotions; more importantly, it is related to the core belief vector. Together, they serve as input to the state transition matrix generating function, modulating the emotional reactivity at the current moment. This design means that a virtual visitor in a state of high physiological stress, even if their core beliefs remain unchanged, will experience a significant change in their emotional response pattern to the same intervention.

[0035] In summary, these three vectors do not exist in isolation, but rather together constitute an interacting dynamic system. Core Belief Vector The "tone" of the response pattern is set, and the emotional state vector is used. It is the "melody" of real-time interaction, while simulating physiological state vectors. This provides the "physiological harmony" that influences the expression of melody. This multi-layered state vector representation method is stored together in the storage unit and is called and updated collaboratively by various modules of the system, providing the necessary and sufficient data foundation for realizing a virtual visitor with depth, consistency and complex dynamics.

[0036] Sudden Memory Association Module In this embodiment, the Salient Event Memory Module (SEMM) is a key component in the system of this invention used to realize nonlinear, event-driven psychological responses. Its fundamental purpose is to break the continuity and smoothness of conventional state evolution models, introducing a mechanism that can simulate the suddenness and discontinuity of phenomena such as traumatic flashbacks and emotional outbursts in real-world psychological activities, thereby greatly enhancing the realism and complexity of the virtual visitor's behavioral patterns.

[0037] This module does not function as a dynamic computing unit within the system, but rather as a structured, static database pre-configured during system initialization. This database stores several preset memory entries with high emotional load, each designed to define a specific "psychological trigger point" and the series of chain reactions it triggers.

[0038] In a preferred embodiment, the database is composed of Composed of triplets, its overall structure can be represented as follows: ; Each index is The triplet contains the following three core elements: First, there is the text trigger ( This element defines the external conditions that activate the memory entry. It is not limited to a single keyword, but can be a set of related words, a specific phrase, or a specific semantic pattern identified through semantic analysis methods. A trigger can be set to input containing any one of the words "father," "strict," or "punishment." This text trigger acts like a key, specifically used to unlock a long-forgotten, meaningful memory.

[0039] Secondly, the content to be memorized ( This element is a descriptive text string that vividly recounts a specific traumatic or pivotal event triggered by the trigger. For example, if the trigger is about "failing an exam," the memory would be a text describing "the feeling of utter devastation when seeing the report card alone after that crucial exam." This text, once activated, does not participate in mathematical calculations but serves as core narrative material directly used in the subsequent response generation phase, providing a concrete and understandable context for the virtual visitor's sudden transformation.

[0040] Furthermore, it is the state impulse vector ( This element is a quantitative tool for achieving instantaneous shifts in mental states; it is a carefully pre-designed, high-dimensional numerical vector. It consists of two parts: one dimension... Emotional impact vector and a dimension Physiological impact vector This vector is designed to simulate the intense emotional and physiological shock an individual experiences when a painful memory is triggered. The value of "fear" and "shame" is significantly positive, while the value of "hope" is significantly negative. The values ​​have large positive values ​​in the dimensions of simulating "heart rate" and "respiratory rate".

[0041] This module operates in parallel with the user intervention parsing module. Upon receiving the user's raw text input, the system immediately compares that text with all text triggers in the SEMM database. The system performs a match. Once a successful match is detected, the system immediately retrieves the complete memory content corresponding to the trigger from the database. and state impact vector This data is then passed to subsequent processing steps.

[0042] The introduction of this module has profound implications for the entire system on two levels. Firstly, its output state impulse vector... In the state evolution module, it will be directly superimposed on the smooth state calculated by the conventional evolution path, thus creating the emotion vector. and physiological state vector A dramatic shift. On the other hand, its output is the text of the memory content. The response generation module will forcibly inject this into the imperative prompts that generate the final reply. This allows the output of the large language model to not only express an abstract, intense emotion, but also to generate language directly related to the specific memory content, thus accurately simulating the psychological phenomenon of "flashback" at the dialogue level.

[0043] In this way, the sudden memory association module endows the system of this invention with the ability to process highly stressed and traumatic responses, enabling it to go beyond simple logical dialogue and exhibit more profound and realistic psychological dynamics, thus providing a more clinically valuable simulated scenario for interactive training.

[0044] State Evolution Module In this embodiment, the state evolution module constitutes the computational core and dynamic engine of the system of the present invention. The fundamental purpose of this module is to simulate a complex psychological evolution process influenced by multiple factors. It receives intervention information from external users and the previous state of the system, and through a series of rigorous and interlocked operational calculations, ultimately outputs a completely new and complete internal psychological state of the virtual visitor at the current moment. This process abandons the simple "input-output" response logic and instead implements a computational model with an inherent causal chain and dynamic evolutionary laws.

[0045] In a preferred embodiment, the module operates in a discrete interaction time step. Internally, a carefully designed, phased process is followed to update the status. This process aims to address psychological changes at different levels and timeframes separately, and ultimately integrate them organically into a unified whole.

[0046] First, this module performs coupled physiological-emotional updates, aiming to reflect the persistent impact of emotions on physiological states.

[0047] In real-world psychological activities, emotions do not arise and dissipate out of thin air; they are often accompanied by corresponding changes in the physiological system, and this influence has a certain lag and cumulative effect. To simulate this phenomenon, the module first uses the emotional state vector from the previous moment stored in the storage unit. and physiological state vector To calculate the baseline physiological state at the current moment. Its calculation follows the formula below: ; Here It is a preset dimension. The emotion-physiology mapping matrix defines the weight coefficients within which different emotions (anxiety) affect different physiological indicators (heart rate) to what extent. It is an element-wise nonlinear activation function, specifically the Sigmoid function, used to constrain the calculation results within a reasonable range of [0,1]. The output of this step... Not only will it form the basis of the current physiological state, but more importantly, it will serve as a key dynamic context in subsequent calculations of emotional responses.

[0048] Next, the module enters its most innovative phase: the generation of the dynamic state transition matrix.

[0049] Traditional methods typically define behavioral patterns using static rules. This invention recognizes that an individual's stress response patterns are not static but are jointly regulated by their deep-seated beliefs and current physiological arousal levels. To achieve this, the module dynamically generates a state transition matrix for a specific moment in time. The generating function for this matrix is: ; The function here It is a non-linear mapping function, and in a preferred implementation, it can be a small, pre-trained feedforward neural network. The function receives two key inputs: a core belief vector representing the long-term, stable cognitive structure. and the baseline physiological state vector representing the current immediate physiological arousal level This function maps these two high-dimensional vectors to a single vector with dimension 1. matrix Each element of the matrix It is precisely defined as: the first [action] imposed by the user in the current core beliefs and physiological state. The first intervention technology for virtual visitors This dynamic generation mechanism allows for the determination of the direction and intensity of an emotion's influence. It also imbues the virtual visitor's response patterns with a high degree of situational dependence and psychological realism.

[0050] The module then uses this dynamically generated transition matrix to perform a smooth emotional state update calculation.

[0051] This step is the core of the regular interaction logic; it transforms external stimuli (user intervention) into changes in emotion through internal states (transition matrix). The calculation formula is as follows: ; This formula indicates that the emotional state is smoothly updated. It was the emotion of the previous moment. Based on this, plus intervention from the current user After the state transition matrix The modulated effect.

[0052] Next, the module handles the impact of sudden events to achieve nonlinear state transitions.

[0053] The module checks whether the Burst Memory Association Module (SEMM) is activated in the current interaction. If activated, it means a dramatic "flashback" reaction needs to be simulated. At this point, the module will retrieve the state impulse vector from the SEMM ( and The current state is directly superimposed onto the already calculated smoothed state and baseline physiological state to obtain the final state at the current moment: ; ; If SEMM is not activated, then the two impulse vectors mentioned above are zero vectors, and the final state is... and This is equivalent to a smoothly updated state. This dual-path design ensures that the system can handle both regular, smooth dialogue evolution and sudden, disruptive psychological events.

[0054] Finally, this module is responsible for handling the evolution of core beliefs, which is key to simulating long-term treatment effects.

[0055] Changing core beliefs is difficult and slow. Therefore, the module does not update the core belief vector in every round. Instead, it will first assess from arrive Net change in emotional state at any given time The module will only fine-tune the core belief vector when this change conforms to a predefined pattern representing "positive treatment progress" (a significant decrease in negative emotional components such as anxiety and defensiveness, and a significant increase in positive emotional components such as hope and security). The update formula is: ; Here It is a very small positive number, namely the learning rate, used to ensure that the change of belief is gradual. It is a gating function that maps positive sentiment changes to a targeted belief update vector. If the sentiment change does not meet the conditions for positive progress, Output the zero vector, such that That is, the core beliefs remain unchanged.

[0056] Through the series of interconnected calculation steps described above, the state evolution module finally outputs the current time step. The set of state vectors that have been fully updated This data is then passed to the response generation module, providing a solid, computable data foundation for generating highly contextualized and psychologically consistent system responses.

[0057] Response generation module In this embodiment, the response generation module is the final output of the system of the present invention. Its core responsibility is to transform the highly structured and mathematical psychological state derived from complex calculations within the system into natural and fluent language and behavioral descriptions that can be perceived and understood by external users. The design concept of this module is to transform the Large Language Model (LLM) from an uncontrolled, black-box decision-making entity into a high-quality "language shell" or "behavior generator" constrained by precise instructions. This ensures the naturalness of interaction while fundamentally guaranteeing the logical consistency, interpretability, and controllability of the virtual visitor's behavior.

[0058] In a preferred embodiment, the workflow of this module can be divided into two main stages: the dynamic instruction construction and synthesis stage, and the instruction-based language model invocation and response acquisition stage.

[0059] First, this module constructs and synthesizes dynamic instructions. The goal of this stage is to create a comprehensive, structured prompt that will serve as the sole constraint and guide for the large language model to generate content. The module first receives its current time step from the state evolution module. The final calculation output, namely the updated three core state vectors: core belief vector Emotional state vector and simulated physiological state vector .

[0060] In order for these numerical vectors to be understood by the language model, the module first needs to "translate" them. It transforms the key components and their values ​​in these vectors into a descriptive natural language.

[0061] Crucially, this module also checks whether the Sudden Memory Association Module (SEMM) is activated during this round of interaction. If the SEMM is activated, the module will obtain the specific text of the memory content transmitted by the SEMM. This text will be unconditionally and with high priority forced into the instructions being built for rendering.

[0062] Finally, the module organizes and encapsulates all the translated and integrated text fragments, along with necessary contextual information, the virtual visitor's basic background setting, and the complete dialogue history with the user, into a complete, structured instruction prompt. This prompt is designed to provide the large language model with a comprehensive and unambiguous set of instructions regarding "who I am," "how I feel," "what I am thinking," and "how I should respond."

[0063] Next, the module enters the instruction-based language model invocation and response retrieval phase.

[0064] At this stage, the module sends the carefully crafted instructional prompts from the previous step to a pre-selected large language model via an application programming interface (API). This large language model can be any mature model in the field with strong natural language understanding and generation capabilities.

[0065] Unlike traditional methods that directly allow large language models to engage in open-ended dialogues, the invocation method in this invention is instructional. The prompts received by the large language model are not simple questions, but tasks with strict "role-playing" instructions. It is explicitly told to generate responses strictly according to the descriptions of internal states (cognitive, emotional, physiological, and spontaneous memories) in the prompts. This constraint ensures that the generation process of the large language model is no longer an endless probability sampling, but rather takes place within a narrow semantic space precisely defined by the internal model.

[0066] After processing this instructional prompt, the large language model generates a natural language text response. This text is stylistically natural and fluent, conforming to human language habits, but in its core content, emotional tone, linguistic logic, and underlying intentions, it will accurately and faithfully reflect the complex and contradictory inner world described by the internal state vector.

[0067] Finally, the response generation module receives the text returned by the large language model and can choose to present it directly to the user or convert it into speech through the text-to-speech (TTS) unit for playback. Simultaneously, it drives the virtual avatar to perform corresponding nonverbal actions (changes in facial expressions and posture), thus completing a full, inside-out interaction. This generation method ensures that every system response experienced by the user is supported by a traceable and analyzable internal psychological state.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An immersive interactive training system for computer-simulated psychotherapy scenarios, characterized in that, include: The storage unit is used to store the core belief vector, emotional state vector, and simulated physiological state vector that represent the internal state of the virtual visitor. The user intervention parsing module is used to receive user intervention input and parse it into intervention vectors. The state evolution module, connected to the storage unit and the user intervention parsing module, is used to calculate and update the core belief vector, emotional state vector and simulated physiological state vector at the current moment based on the intervention vector and the three state vectors stored in the storage unit at the previous moment. The response generation module, connected to the state evolution module, is used to generate a system response based on the three updated state vectors at the current moment using a large language model.

2. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 1, characterized in that, The state evolution module is specifically used for: A state transition matrix is ​​generated, and the current emotional state vector is calculated based on the intervention vector, the emotional state vector of the previous moment, and the state transition matrix.

3. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 2, characterized in that, The state evolution module is configured as follows: The state transition matrix is ​​dynamically generated based on the current core belief vector and simulated physiological state vector using a preset mapping function.

4. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 1, characterized in that, The state evolution module is also specifically used for: Based on the emotional state vector of the previous moment and the preset emotional-physiological mapping matrix, the simulated physiological state vector of the current moment is updated and obtained.

5. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 1, characterized in that, The state evolution module is also specifically used for: When the difference between the current emotional state vector and the previous emotional state vector is detected to meet the preset positive progress pattern, the core belief vector is updated.

6. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 1, characterized in that, The system also includes: The burst memory association module is used to store at least one triplet of data containing a text trigger, memory content, and state impulse vector.

7. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 6, characterized in that, The state evolution module is also specifically used for: When the user intervention parsing module identifies the text trigger, the state impact vector is applied to the update calculation of the emotional state vector and the simulated physiological state vector.

8. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 1, characterized in that, The user intervention parsing module integrates a natural language processing model to convert the user's text input into the intervention vector.

9. The immersive interactive training system for computer-simulated psychotherapy scenarios according to claim 1, characterized in that, The response generation module is specifically used for: The updated core belief vector, emotional state vector, and simulated physiological state vector at the current moment are combined into an instructional prompt and input into the large language model to generate the system response.

10. A method based on the interactive training system of claim 1, characterized in that, Includes the following steps: Obtain user intervention input and parse it into intervention vectors; Based on the intervention vector, and the pre-stored core belief vector, emotional state vector, and simulated physiological state vector from the previous moment, calculate and update the core belief vector, emotional state vector, and simulated physiological state vector for the current moment. Based on the updated core belief vector, emotional state vector, and simulated physiological state vector at the current moment, a system response is generated through a large language model.

Images