A scheduling method in a multi-user virtual reality scene

Abstract

The application relates to the technical field of virtual reality, in particular to a scheduling method in a multi-user virtual reality scene, which comprises the following steps: obtaining the motion state of each user and the target position corresponding to each virtual role; predicting whether a preset conflict condition is met under the condition that the current corresponding relationship between each user and each virtual role is kept unchanged, wherein the current corresponding relationship is a one-to-one corresponding relationship formed by the virtual roles currently driven by each user; and switching the current corresponding relationship to a target corresponding relationship which makes the preset conflict condition no longer met, so that each user drives the virtual role corresponding to the target corresponding relationship, when the preset conflict condition is met; and the target corresponding relationship is another set of one-to-one corresponding relationship between each user and each virtual role, which is different from the current corresponding relationship.

Description

Technical Field

[0001] This invention relates to the field of virtual reality technology, and more specifically, to a scheduling method in a multi-user virtual reality scene. Background Technology

[0002] In a multi-user co-located virtual reality scenario, multiple real users move within the same physical space, each driving a virtual character in the virtual scene from a first-person perspective. Constrained by the physical space area and safe distance, each user's physical movement range is limited. However, the narrative goals undertaken by each character in the virtual scene may require users to move in any direction. Therefore, conflicts may occur between the physical movement trajectories of different users, such as when the distance between any two users is less than the safe distance, or when they meet head-on in a narrow passage.

[0003] To address the aforementioned conflicts, one approach in existing technologies is to apply imperceptible geometric gains (such as rotation gain, translation gain, and bending gain) to the geometric mapping between the virtual scene and the physical site, guiding the user to circumvent the conflicts without their awareness. However, the effectiveness of this approach is limited by a perceptible threshold for the geometric gain: when the conflict cannot be resolved within this threshold (e.g., when passing in opposite directions within a narrow channel), the geometric gain channel becomes ineffective.

[0004] Another approach in existing technologies involves resetting the user's current movement and guiding them to a different direction when a conflict is imminent. Resetting interrupts the user's virtual experience, disrupting the rhythm of the exercise or task. In multi-user scenarios with frequent conflicts, frequent resets significantly reduce the continuity and immersion of the experience.

[0005] Therefore, how to resolve physical conflicts in multi-user virtual reality scenarios without disrupting the user experience when geometric gains cannot eliminate the conflicts is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] Therefore, it is necessary to provide a seamless scheduling method based on identity mapping permutation in multi-user virtual reality scenarios to address the above-mentioned technical problems. This method can resolve multi-user physical conflicts that cannot be resolved by geometric gains without interrupting the user experience.

[0007] This application provides a scheduling method for a multi-user virtual reality scene, applicable to multiple users in the same physical location, where each user corresponds one-to-one with multiple virtual characters in the virtual scene, and each user drives the virtual character corresponding to themselves; the method includes: Obtain the individual movement status of each user and the target location of each virtual character; Based on the movement state of each user and the target position of each virtual character, predict whether the preset conflict condition is met while keeping the current correspondence between each user and each virtual character unchanged; the current correspondence is a one-to-one correspondence formed by the virtual characters currently driven by each user; the preset conflict condition is used to characterize the conflict that will occur in the physical field when each user drives the corresponding virtual character to move to the corresponding target position according to the current correspondence. If the preset conflict conditions are met, the current correspondence will be switched to the target correspondence that makes the preset conflict conditions no longer met, and each user will be changed to the virtual role corresponding to themselves under the target correspondence. The target correspondence is another one-to-one correspondence between each user and each virtual role, which is different from the current correspondence. The switching is performed during the visual suppression period of the relevant users among the users; the relevant users are the users whose virtual characters are changed during the switching, and the users in the area where the virtual characters' field of view coverage changes.

[0008] The technical advantages of this application are as follows: In the aforementioned identity mapping permutation-based imperceptible scheduling method for multi-user virtual reality scenarios, the switching of correspondences is performed during the visual inhibition period of each user involved in the switch. The switching transient is masked by physiological visual inhibition and is not perceived by the user. The switch only changes the identity binding and narrative target assignment between the user and the virtual character. The physical pose, first-person viewpoint, and pose and movement of the characters at each position of each user remain continuous without interrupting user movement or performing a reset. Geometric gain is prioritized to resolve conflicts, and permutation is quantified by the perceptual budget and is only activated when necessary, ensuring the long-term sustainable operation of the solution. Compared with existing reset schemes, this application can resolve physical conflicts that geometric gain cannot resolve without interrupting the user experience, maintaining the coherence and immersion of the multi-user virtual reality scene. Attached Figure Description

[0009] Figure 1 This is a schematic diagram showing the comparison of identity replacement before and after an embodiment of this application. Figure 2 This is a schematic diagram illustrating the application environment of a scheduling method in a multi-user virtual reality scene provided in an embodiment of this application. Detailed Implementation

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

[0011] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0012] It should also be noted that, in order to ensure strict consistency in the names and symbols of the same technical feature throughout this application, the following terminology and symbol conventions are adopted. (User) Indicates the first One real user The set of all users is denoted as Virtual characters Indicates the first in the virtual scene A virtual character The set of all virtual characters is denoted as In the embodiments of this application, Identity mapping From user set To the character collection One-to-one mapping Indicates user Current driving role .exist Under the agreement, the collection With sets By their element subscripts and index sets respectively Natural equivalence ( Corresponding indicators , Corresponding indicators Identity mapping Corresponding indicator set A permutation on the above, by which this application will Consider as a symmetric group The elements in the text. Substitution. This indicates a change to the identity mapping. After the replacement is executed, the identity mapping is handled by... Become (i.e., apply first) Then apply it to the result. (composite mapping); where The symbol for this type is the rotation symbol, indicating the interchange of the indices within the parentheses. It is important to emphasize that the symbol... In this application, it is only used to indicate permutation; wherever statistical quantities are involved, they are all expressed in words as "standard deviation" and do not occupy symbols.

[0013] Geometric gain Represents the rotation gain, translation gain, and bending gain of the virtual-physical geometry mapping. Among these, the rotation gain... With translation gain Dimensionless ratio ( , (Corresponding to no scaling); bending gain The curvature of the arc traveled is used as a measure (curvature is the reciprocal of the radius of the arc, and its dimension is the reciprocal of the length). (Corresponding to straight-line walking). The range of values ​​for geometric gain that are statistically imperceptible to the user is called the gain threshold, i.e. , , (That is, the radius of the walking arc corresponding to the bending gain is not less than 22 meters); where "statistically imperceptible" means that there is no statistical difference between the detection rate of user gain manipulation and the baseline detection rate when no manipulation is applied. All detection thresholds in this application are defined according to this statistical caliber. It should be noted that the above gain threshold values ​​are calibration values ​​obtained under classical experimental conditions. Subsequent studies have shown that they vary with task, movement speed, and adaptation status. During implementation, they should be recalibrated according to the above statistical caliber based on specific implementation conditions. This application does not rely on the universality of this set of values. It should also be noted that... , , The three variables have different dimensions and are used as constraint variables with independent box constraints in the geometric solvability determination (step S102). The box constraints are applied independently by component and it is not required that the dimensions of each component be the same. The three variables do not constitute or need to constitute vectors in the same metric space.

[0014] blind window Indicates user The time interval during which visual perception is physiologically suppressed includes the saccade suppression interval and the blink suppression interval. The duration of saccade itself is approximately 20 to 200 milliseconds, and its duration increases with the amplitude of the saccade according to the principal sequence relationship (the principal sequence relationship refers to the deterministic empirical relationship that the duration of the saccade and the peak velocity increase monotonically with the amplitude of the saccade, where the peak velocity tends to saturate under large amplitude saccades, which is common knowledge in the field of eye tracking). Under typical amplitude, the duration of the saccade is approximately 20 to 60 milliseconds; the saccade suppression interval starts about 50 milliseconds before the saccade occurs and ends about 50 milliseconds after the saccade ends, and its total length is approximately equal to the duration of the saccade plus about 100 milliseconds, that is, about 120 to 300 milliseconds. The duration of the blink suppression interval is approximately 100 to 150 milliseconds, while the natural blink interval is approximately 4 to 6 seconds under normal conditions. It should be noted that the blink rate may decrease in head-mounted display viewing scenarios (e.g., to approximately 5 to 7 times per minute), correspondingly lengthening the natural blink interval. Therefore, the waiting delay budget for the blink blind window should be determined based on the measured distribution (see the backup path in step S106 for details). It should also be noted that the 100 to 150 milliseconds is a conservative estimate, corresponding to the duration of core visual interruption caused by eyelid occlusion—the generalization central inhibition of blinking may, in the literature, begin slightly earlier than eyelid closure and end slightly later than eyelid opening. This application uses the above conservative value for blind window timing budgeting, which only compresses the available window and does not misjudge moments outside the window as within it. Therefore, this value is sufficient for the security of gating submissions. Furthermore, users in a natural viewing state will continuously generate spontaneous saccades, and the spontaneous saccade rate (denoted as...) The typical value for spontaneous saccade suppression is about 2 to 3 times per second, so the spontaneous saccade suppression interval naturally appears at the expected interval in the sub-second range. Blind window The term "blind window" refers to the combined duration of the saccade inhibition interval and the blink inhibition interval. Its shortest duration is the smaller of the shortest durations of the two mechanisms: the shortest duration of the saccade inhibition interval is approximately 120 milliseconds, and the shortest duration of the blink inhibition interval is approximately 100 milliseconds. The minimum duration is approximately 100 milliseconds (corresponding to the blink suppression interval); this value is used for timing budgeting that needs to adapt to any type of blind window, and taking a smaller value is conservative (it only compresses the available submission window and will not misjudge moments outside the window as moments inside the window). Blind window induction stimulation refers to transient brightness or transient motion stimulation applied to the periphery of the user's visual field to reflexively induce a saccade, thereby artificially creating a blind window at the planned moment. The latency period for the stimulus-onset of reflexive saccades is approximately 150 to 250 milliseconds. It should be noted that the blind-window induced stimulus itself is a visible transient event presented in the user's field of vision. Its function is to reflexively induce saccades and mask the switching transients submitted by gating with the inhibition interval of the saccades. The induced stimulus itself is designed as a natural transient event compatible with the scene content, and its perceptibility is recorded in the perception budget account according to the concealment operation (see the accounting rules in step S107). This application does not rely on the assumption that the induced stimulus itself is invisible.

[0015] Perceived Budget It represents the user At any moment A scalar state variable representing the cumulative detectability of covert operations, with a value range of [value range missing]. ,in For users The upper limit of the perception budget. Detecting risks. Indicates user The probability of detecting a stealth operation is used for perception budget. It is a monotonically increasing function.

[0016] The identity mapping table is a data structure maintained locally on each client, recording which user drives each virtual character and the identity bindings (binding relationships between appearance parameters, titles, and narrative targets) of each rendered instance. It employs a double-buffered structure: one copy of the identity mapping table is stored in the current buffer and another in the backup buffer, with the active pointer pointing to the currently active copy; this is only used in the blind window. The method of submitting the active pointer of the inverted identity mapping table is called gated submission at moments of perceived discontinuity, hereinafter referred to as gated submission. It should be noted that this application distinguishes between two concepts: virtual role. The virtual character (VCR) is a narrative entity that carries identity, title, and narrative goal. A rendering instance is the presentation unit of each virtual character on the client, driven by the physical pose of the currently driving user, according to its position binding method. Displacement changes the identity binding between the virtual character and the rendering instance, but does not change the pose of the rendering instance driving the user. Shadow representation is the pre-assembled data of the displacement target state on the client, including the appearance resources of the target identity, title and interface element binding parameters, and narrative goal binding parameters. The displacement schedule table is a data structure generated by the scheduler, recording displacement pairs, displacement modes, and schedule submission times. With extension limit ; Extend the upper limit This represents the maximum allowable delay for the planned submission time. The identity mapping table employs a double-buffered structure: the current buffer stores the currently active identity mappings, with the active pointer pointing to the current buffer; the spare buffer pre-assembles shadow representations, i.e., identity mappings that replace the target state; and in the blind window... The active pointer is flipped from the current buffer to the standby buffer, thus completing the gated commit (sub-millisecond operation).

[0017] The rendering and viewpoint model adopted in this application is a position-bound model: each virtual character acts as a body proxy for its driving user, and its rendering pose is determined by the physical pose of the currently driving user through the user's virtual-physical geometry mapping; each user observes the virtual scene from the viewpoint of the virtual character they drive in a first-person perspective, that is, each user's first-person viewpoint is always located at the rendering position corresponding to their own physical pose. Correspondingly, the reference eye height of the user's viewpoint is determined by the set eye height of the virtual character they drive, and the relative movement of the user's head is superimposed on this reference eye height in real time by the positioning unit. Based on this model, the permutation is performed in a position-bound manner: at the moment of submission, the physical pose and first-person viewpoint of each user remain unchanged, and each client updates the identity binding of the permutation-related rendering instance according to the target correspondence. Since the same rendering instance before and after the permutation is still driven by the same user's physical pose, the pose and movement of the characters at each position seen by each observer remain continuous; each permutation also maintains continuity with the user's own first-person viewpoint and body movement; the permutation only changes the identity binding and narrative target assignment. It should be noted that the viewpoint in position rigging refers to the rendered viewpoint determined by the head-mounted display's tracking coordinate system of the positioning unit; interpupillary distance adjustment and individual differences in the relative position of the eyes and the head-mounted display are routine corrections at the binocular display level and do not affect the aforementioned position rigging semantics. Specifically: the real user stands on the physical ground, the head-mounted display determines the rendered viewpoint, and drives the virtual character currently being driven by the user through virtual-physical geometry mapping. (Right now The rendering pose of the indicated role is always bound to the user's physical pose. The physical walking trajectory and the virtual perception trajectory may differ due to geometric gain, but the user's first-person viewpoint always follows the physical pose. This is the semantics of position binding - the permutation only changes the identity binding, not the pose and viewpoint.

[0018] There are two permutation modes. Permutation mode one (including identity permutation with appearance permutation) is suitable for scenarios where differences in the appearance of each character are allowed, but can be made indistinguishable: at the moment of submission, the permutation swaps the appearance parameters of the corresponding rendered instances of both characters according to the target correspondence; the premise is that the difference in appearance parameters (including set eye height) between the two characters has converged to the static discrimination threshold. The following conditions ensure that even when observed side-by-side, the objects are indistinguishable (in this application, all detection thresholds are defined according to statistical standards; "indistinguishable" means that the observer's discrimination detection rate has no statistical difference from the baseline level), so that the replacement of appearance parameters at each location does not produce perceptible static differences, and the transient state is then masked by a blind window; among which... This is the static discrimination threshold for the appearance of two characters under normal viewing conditions, which is different from the frame-by-frame change detection threshold used for progressive adjustment. The convergence process is achieved through frame-by-frame adjustments (the adjustment amount per frame does not exceed a certain limit). Therefore, the convergence process itself is statistically imperceptible. Permutation Mode Two (Appearance-Unified Identity Permutation) is suitable for scenarios where the appearance of each character is pre-unified (e.g., uniformed characters): the appearance parameters and eye height settings of each identity are uniformly configured. At the moment of submission, the appearance parameters and baseline eye height of each rendered instance do not need to change. Only the identity identifier and narrative target assignment are re-bound according to the target correspondence. Therefore, the permutation does not produce any changes in character appearance, pose, or geometry, nor does it change the baseline eye height for the user. It should be noted that if the identity identifier is presented as a shared visual element (e.g., a nameplate above the head, a teammate list, or a minimap marker), such elements are updated along with the identity binding. Since pointer flipping on each client is completed within its local user's blind window, the update of such visual elements also takes effect within the blind windows of the corresponding observers. Therefore, the permutation does not produce any perceptible visible changes for any observer. Changes to interface elements visible to the user, such as titles and target hints, are also executed within their blind windows, and the update of target hints is a common narrative event in this application.

[0019] like Figure 1 The diagram shows a comparison of identity swapping before and after, and a sequence of gating submissions. Figure 1 The left side is before the replacement, user Towards the right target Forward, User Towards the left target Moving forward, the two meet facing each other in the passage; there is no geometric solution. Figure 1 On the right is the scene after the swap, where the two characters switch roles. drive (Original (identity) and turned back along the way they came. drive (Original (identifying their identities) and then turning back along the path they came from, their paths no longer intersecting, the conflict resolved, and the experience uninterrupted; in Constantly apply blind window guidance stimulation to the user's field of vision, scanning the area around the user's eyes. This occurs at any given time. After detecting a saphenous scan, each client flips the active pointer of its local identity mapping table within the saphenous scan suppression interval (blind window), the switch takes effect, and all submissions are made no later than [date missing]. Completed, moment of conflict materialization Located at the right end of the timeline.

[0020] A consistency proxy is a software entity that intercepts interactions directed to the replaced role during the replacement transition period. (Verification observation period) This is the time interval used to verify whether the submission has been detected after the gating submission is completed, with values ​​ranging from 2 to 5 seconds. Assignment misalignment conflicts refer to conflicts that satisfy the following two conditions: the set of narrative target positions for each virtual character exactly constitutes a permutation of user target assignments under the current identity mapping (set condition), and there exists a non-identical permutation. This allows each user to autonomously move according to their newly assigned target and reach the target position of the replaced role without trajectories conflicting with each other (trajectory conditions).

[0021] In addition, the remaining symbols used in this application are uniformly agreed upon as follows: prediction window (Within which time window is the forward push of each user trajectory to determine geometric solvability); Conflict materialization moment (The actual time of the conflict, estimated by trajectory projection); safety distance Any two roles within the group , The difference in the presentation parameter vectors (Frame-by-frame change detection threshold) Static discrimination threshold (As defined above); Single-frame rendering parameter adjustment amount ; Movement style parameters (a parameter vector composed of periodic movement characteristics such as stride frequency, stride length, arm swing amplitude, and phase; the character presentation parameter vector in this application is composed of appearance parameters and movement style parameters combined). , , , The convergence criterion applies to the merged rendering parameter vector; switching related users (users whose virtual characters are changed during the permutation, and observer users in the local rendering view involved in the permutation-related rendering instances); frame interval ; Reduction required for appearance convergence Upper limit of appearance convergence time Perceive budget increments (of which the first) The increment corresponding to the next concealment operation is denoted as... The time of its occurrence is recorded as Perceived budget decay time constant); ( ); Detect the location parameters of the risk logic function With scale parameters ( , and (same dimensions); where the position parameter With the cumulative number of covert operations Move right, denoted as Its base value before habituation is denoted as ( That is, the perception budget level corresponding to a detection rate of 50% under baseline conditions; the base value of the upper limit of the perception budget. ( The implementation must meet the design requirements. (i.e., the budget ceiling base value is not higher than 50% of the baseline detection point); the calibration constant of the habitual saturation function. and ( And it is dimensionless. (and is counted in terms of the number of operations); risk threshold for detecting risk. Mean of individualized saccadic latency for each user (Estimated based on the mean of the user's reflexive saccade latency distribution); Spontaneous saccade rate for each user (See definition above); the set of narrative target locations of the set of characters involved in the conflict. ; and current identity mapping The set of physical target locations that will inevitably be generated by each user according to their role objectives. .

[0022] This application provides a scheduling method for multi-user virtual reality scenarios, which can be applied to the following: Figure 2 The application environment shown includes server 101 and... There is one client 102. The server 101 communicates with the client 102 through a communication network, which can be a wired network or a wireless network. The scheduler is deployed on the server 101. Client 102 and individual users One-to-one correspondence, meaning client 102 and the user using that client. One-to-one correspondence. Each client has 102 connections: a head-mounted display unit for presenting virtual scenes to the user; an eye-tracking unit with a sampling rate of no less than 120 Hz for detecting the user's saccades and blinks; and a positioning unit for tracking the user's head posture.

[0023] The requirement for an eye-tracking unit sampling rate of at least 120 Hz is due to timing budget considerations for gating submission: the submission action must be completed after the client detects a saccade and before the end of the saccade suppression interval. The shortest duration of the saccade suppression interval is approximately 120 milliseconds (corresponding to a saccade duration of approximately 20 milliseconds, starting approximately 50 milliseconds before the saccade occurs); 120 Hz sampling corresponds to a sampling interval of approximately 8.3 milliseconds. In practice, 2 to 3 consecutive sampling points exceeding the threshold are used to confirm the occurrence of a saccade, which can reliably identify the saccade within approximately 17 to 25 milliseconds after its start. It should be noted that gated submission can only occur after saccade detection is completed. The initial portion of the suppression interval before saccade occurs is unavailable for triggered submission. Therefore, the timing budget is always based on the remaining window after detection: after deducting approximately 50 milliseconds for the initial saccade and approximately 17 to 25 milliseconds for the detection delay, approximately 45 to 53 milliseconds of usable submission window are still retained between the completion of detection and the end of the suppression interval, which is sufficient to provide ample time margin for sub-millisecond activity pointer flipping and local verification. If the sampling rate is significantly lower than this level, the saccade detection delay will further occupy the suppression interval, compress the submission margin, and increase the exposure risk. It should be noted that the 120 Hz mentioned above refers to the effective sampling rate: in implementation, the detection granularity should be calculated based on the measured effective sampling rate of the eye-tracking unit (when the measured effective sampling rate is lower than the nominal value, the detection granularity should be coarsened accordingly based on the measured value), and the end-to-end delay of the eye-tracking link should be included in the saccade detection delay budget. The timing constraint that "the detection completion time is earlier than the end time of the saccade suppression interval" should be reconfirmed based on the approved detection delay. Optionally, to further increase the temporal margin, the blind window induction stimulus is applied to an off-center position in the periphery of the visual field so that the induced saccade amplitude is not less than about 10 degrees; according to the saccade master sequence relationship, the saccade duration corresponding to this amplitude is more than 40 milliseconds, and the corresponding saccade inhibition interval is more than 140 milliseconds.

[0024] Optionally, server 101 can be an independent physical server, server cluster, distributed system or cloud server; the terminal where client 102 is located can be a desktop computer, laptop computer, portable wearable device, etc., and this application does not limit it.

[0025] In some embodiments, the scheduling system includes a scheduler and There are 102 clients. The scheduler includes modules for conflict prediction, conflict identification, and plan generation. The clients include modules for commit preparation, gated commit, and consistency maintenance. Each client also maintains a local double-buffered identity mapping table. The accounting function for covert operations is integrated into the consistency maintenance module, while the intra-group preparation function, which runs continuously during the session, is handled by the preparation module. The scheduler-side modules handle global decision-making functions such as conflict prediction, conflict identification, and plan generation, while the client-side modules handle local execution functions such as shadow representation assembly, commit readiness checks, blind window detection, active pointer flipping, and perception verification. The two are connected through the distribution of the replacement plan table and status reporting. Specifically, the scheduler (server 101 side) includes three main modules: conflict prediction, conflict identification, and plan generation. Each client 102 includes three main modules: commit preparation, gated commit, and consistency maintenance, and maintains a local double-buffered identity mapping table. Each client connects to a head-mounted display unit, an eye-tracking unit, and a positioning unit. The scheduler and each client collaborate through the distribution of the replacement plan table and status reporting.

[0026] This application provides a scheduling method for a multi-user virtual reality scene, which can be executed by server 101, client 102, or jointly by both. For example, steps requiring global information, such as conflict prediction and permutation decision-making, are executed by server 101, while steps requiring millisecond-level local responses, such as saccade detection and active pointer toggling, are executed by client 102. The two entities collaborate to complete all steps. This application does not limit the executing entity for each step in the scheduling method.

[0027] In some embodiments, a scheduling method is provided for a multi-user virtual reality scene, which is executed collaboratively by a scheduler and the client corresponding to each user. In this embodiment, the method includes the following steps: Step S102, Conflict prediction and geometric solvability determination: within the prediction window The trajectory of each user is pushed forward to determine whether there is a geometric gain trajectory that satisfies the gain threshold constraint, so that all users maintain a physical distance of no less than a safe distance between each other. And it does not go beyond the physical boundaries of the site.

[0028] Step S103, Conflict type identification: If the geometric solvability is determined to be infeasible, identify whether the conflict is an assignment misalignment type conflict and check whether there is a permutation pair that meets the legality conditions.

[0029] Step S104, Replacement plan table generation: If the conflict is an assignment misalignment type conflict and there are valid replacement pairs, generate a table recording the replacement pairs, replacement mode, and plan submission time. With extension limit Replacement plan.

[0030] Step S105, Shadow Representation Assembly and Submission Readiness Check: Each relevant client assembles the shadow representation according to the replacement plan, and confirms that the shadow representation assembly is complete and that each switching-related user is in a state where blind window induction stimulation can be applied.

[0031] Step S106, Blind Window Induced Alignment and Gated Submission: Before the planned submission time, based on the average individualized saccade latency of each observer. A blind window induction stimulus is applied, and each client flips the active pointer of its local double-buffered identity mapping table within the blind window after detecting a saccade.

[0032] Step S107, Transition Period Consistency Maintenance, Post-Commit Verification and Accounting: Maintain the interaction consistency during the transition period from the occurrence of the first local commit. Verify whether the commit has been detected during the verification observation period after the commit is completed, and convert the intensity of each hidden operation in this transaction into the perceived budget increment and write it into the perceived budget account.

[0033] It should be noted that, in addition to the sequentially connected steps described above, the method also includes step S101, namely, in-group preparation. Step S101 is a background step that runs continuously during the session: its output (appearance compatibility status of roles within the group, measurement results of body size and body shape parameters of each user, individualized blind window induction parameters for each user, and the raised perceptual budget ceiling) is a prerequisite for the legality check in step S103 and the blind window induction alignment in step S106. Therefore, step S101 does not have a sequential relationship with steps S102 to S107, but rather continuously provides them with prerequisite parameters. The logical dependencies of steps S102 to S107 are as follows: the negative result of the geometric solvability determination in step S102 is a prerequisite for entering step S103; the positive determination of the conflict type in step S103 is a prerequisite for generating the replacement plan table in step S104; the plan submission time in the replacement plan table... The substitution pair is a common input for both step S105 (assembly) and step S106 (gating submission); the readiness confirmation in step S105 is a prerequisite for proceeding to step S106; the occurrence of the first local submission in step S106 is the initiation condition for transitional consistency maintenance in step S107, and the submission completion event in step S106 is the trigger condition for verification and posting in step S107. If any step is omitted, subsequent steps will lose input or their validity basis.

[0034] The implementation methods for each step are explained in detail below.

[0035] First, let's explain step S101 (intra-group preparation). Step S103, which selects legitimate permutation pairs, requires three prerequisites: both parties in the permutation pair must be compatible in body type and appearance; the blind window induction parameters for both parties must be calibrated; and the perceptual budget upper limit for both parties and relevant observers must be raised to a level sufficient to cover a single submission. Appearance compatibility can be achieved through active preparation, blind window induction parameters can be obtained through calibration, and the perceptual budget upper limit can be raised through habituation. Establishing these three conditions requires cumulative time measured in minutes. If preparation is only initiated when a conflict occurs, it will be too late. Therefore, preparation must be continuously carried out in preparation groups during the session. Body type compatibility depends on the actual body size of the users; the preparation loop can only measure and group these dimensions, not change them. Close interaction is an instantaneous state, and both are determined instantly by step S103 during decision-making. The scheduler divides users into preparation groups, and three loops are continuously executed within each group.

[0036] In some embodiments, the first loop is an appearance convergence loop: the appearance parameters and motion style parameters of the characters driven by users in the same group converge at a subthreshold rate (the adjustment direction is to make the parameter vector difference between each pair of characters in the group appear as a result). (The direction of monotonically decreasing). This is based on the statistical imperceptibility of gradual changes: In this application, all detection thresholds are defined according to statistical standards—when the amplitude of a single change is lower than the frame-by-frame change detection threshold, the observer's detection rate of this change is not statistically different from the baseline detection rate when no change is applied; at this point, the change is said to be statistically imperceptible. As supporting evidence, the false negative rate of gradual appearance changes in the blindness experiment reached 73% to 85% (i.e., even for gradual changes with discernible amplitude, the observer did not detect them in most trials). Let any two roles within the group... , The difference between the presentation parameter vectors is The frame-by-frame change detection threshold is (The adjustment amount applied to each character per frame satisfies) ; The frame-by-frame changes of each role are measured separately. Therefore, when the permutation is adjusted simultaneously for both parties, the frame-by-frame changes of each party remain subthreshold and will not improve perceptibility. The static discrimination threshold is... ;when When, the required reduction is denoted as .because Constraints are applied frame-by-frame by character: when only one character adjusts at the maximum subthreshold rate, The maximum reduction per frame The exchange applies to both parties at a rate not exceeding [amount missing]. When the maximum speed is adjusted towards each other (i.e., "moving closer"), The maximum reduction per frame Therefore, the number of frames required for convergence. Satisfying bilateral inequalities

[0037] in This indicates rounding down; "+1" is used to ensure convergence. Strictly less than Rather than exactly equal to (when When the value is an integer multiple of the corresponding rate, there is exactly one extra frame to ensure it is strictly less than the given rate; when the value is a non-integer, the formula is rounded up. Taking the most conservative one-sided adjustment case, the upper bound of the appearance convergence time is...

[0038] in The frame interval is [missing information]; when both sides move closer to each other, the convergence time is in the asymptotic sense (i.e., [missing information]). Much larger (Time) can be shortened to approximately When the required reduction is small, this ratio approaches 1. That is, the above formula gives an upper bound estimate of the time required to guarantee convergence (this upper bound is based on the premise that all parties adjust at the maximum subthreshold rate; if the actual adjustment rate is lower than the maximum rate, the convergence time will be extended accordingly, and the scheduling should be recalculated based on the actual rate). In the implementation of adjusting at the maximum subthreshold rate, the actual convergence time is no later than this value, and the scheduler uses... Scheduling is performed at the latest possible moment after convergence to ensure conservatism and safety in the scheduling. After this convergence process, roles within the group achieve appearance compatibility in pairs (i.e., the difference in appearance parameters is less than...). (And, when observed statically, they are statistically indistinguishable). The convergence objects include appearance parameters and set eye height. This loop transforms "permutation pairs must be appearance compatible" from a priori screening condition into a state variable that can be actively manufactured, which is a prerequisite for the availability of permutation mode one.

[0039] In some embodiments, the second loop is a blind window calibration loop: probing peripheral transient stimuli (presented as transient events compatible with the scene content) of varying intensities are applied to each user at different times; the stimulus intensity threshold and latency distribution for each user to produce reflexive saccades are recorded to form individualized calibration parameters, wherein the mean of the latency distribution for each user is the individualized saccade latency mean. The individualized calibration parameter is the one that step S106 can use to schedule the submission time. The basis for determining the timing of stimulus application in reverse order of the target: if the user The mean incubation period is The blind window induction stimulus should be in Apply it constantly, only by scanning can you... It happens constantly, and the saccade inhibition zone is self-contained. Coverage starts from approximately milliseconds and the subsequent submission time.

[0040] In some embodiments, the third loop is a habituation loop: performing small, covert operations during low-risk periods. The physiological basis for this is the adaptive drift of the detection threshold: repeated exposure to similar sub-threshold manipulations temporarily raises the user's detection threshold; experiments have documented that the gain detection threshold increases with exposure. In the language of perceptual budget, this is expressed as the upper limit of the perceptual budget. With the cumulative number of covert operations Ascending according to the saturation function:

[0041] in The upper limit of the budget before habituation ( ), , For calibration constants ( And it is dimensionless. (And counted by the number of operations). In , , Under the constraints, Follow Strictly monotonically increasing and saturating ,and This is strictly consistent with the loop objective of "raising the upper limit of the perception budget." The corresponding manifestation of raising the detection threshold in the detection risk model is that the position parameter of the detection risk logic function changes with... Shift to the right according to the same saturation state, that is

[0042] in This represents the location parameters before habituation (the perceptual budget level corresponding to a detection rate of 50% in the baseline state). Therefore, for the same perceptual budget level... Risk detection The perception budget decreases strictly with increasing habituation, which is strictly consistent with the physiological mechanism that "an increased detection threshold makes operations of the same intensity less noticeable"; The increase in threshold value characterizes the expansion of the cumulative operation volume that the user can tolerate; both, together and in the same saturation form, characterize the adaptive drift of the detection threshold. It should be noted that the implementation must meet the design conditions. (The upper limit of the perceived budget is no higher than 50% of the baseline detection point); under this condition, as habituation deepens, the domain boundary is allowed to ( The risk of detection on [the surface] does not increase—specifically, when [the risk of detection on the surface] does not increase. Time boundary risk Strict decline, when The boundary risk is always 50% of the detection point value, and the expansion of the habitual loop does not come at the cost of increasing the boundary risk. The effect of this loop is to raise the upper limit of the perception budget, that is, to expand the amount of available operations for blind window induction and gating submission in step S106.

[0043] The outputs of the three loops mentioned above together constitute the legality prerequisites for subsequent steps: appearance compatibility status and body shape parameters are used for the legality condition check in step S103, and the individualized latency mean is also included. The increased perceptual budget upper limit is used for the legality condition check in step S103 and for the reverse stimulus application time in step S106. It is used for the budget balance check in step S103 and the budget consumption in step S106. Therefore, although step S101 is independent of the main path that connects steps S102 to S107 in the process, it is a preparatory step for the legal execution of each step in the main path.

[0044] Step S102 (Conflict Prediction and Geometric Solvability Determination) answers the question "Is it necessary to use substitution?" The reason for answering this question first is that the execution of substitution consumes the perceived budget (its consumption is recorded according to the accounting rules of step S107), and the perceived budget is a finite resource. Therefore, substitution should not be used for conflicts that can be resolved by geometric gain.

[0045] In some embodiments, the scheduler takes the current motion state of each user (including physical pose and velocity) and the narrative target position of each virtual character as input, and performs prediction within a prediction window. The inner forward trajectory is determined, and the following feasibility problem is solved: Does a gain trajectory satisfy the gain threshold constraint exist? This allows all users to be within the prediction window. Maintain a physical distance of no less than the safe distance between each pair of objects. And it must not exceed the physical site boundaries. This problem is a feasible trajectory determination problem with box constraints, which can be solved using conventional feasibility checks of model predictive control. The determination is performed on a rolling basis during the session; when a decision is made infeasible, the conflict entityization time is obtained simultaneously by trajectory forward calculation. The estimated value is used in step S104 to determine the planned submission time. If it is determined to be feasible, only geometric gain is applied to resolve conflicts, and the transaction ends; if it is determined to be infeasible, it means that the geometric channel has been exhausted, which is the only entry point to step S103, thus ensuring that the replacement is only enabled when necessary, realizing the scheduling principle of "geometric priority". It should be noted that the geometric gain is applied within the statistically imperceptible range defined by the gain threshold, and its corresponding perceived budget increment is counted as zero. Therefore, the geometric channel does not consume perceived budget (see step S107 for the accounting rules).

[0046] Step S103 (conflict type identification) answers the question, "Does a geometrically unsolvable conflict belong to the type that can be resolved by substitution?" Identification is necessary because substitution changes the assignment relationship of "which body goes to which target." Therefore, substitution is only effective when the conflict stems from a misalignment in the assignment relationship itself, rather than from no one going to the target location.

[0047] In some embodiments, the determination criterion is: let the set of narrative target locations of the set of characters involved in the conflict be denoted as... (in (Number of roles involved in the conflict), record the current identity mapping. The set of physical target locations that will inevitably be generated by each user according to their role objectives is as follows: If and only if there exists a non-identical permutation. This allows each user, after the replacement, to reach their respective destinations via physical trajectories generated by their newly assigned targets. The physical location of the corresponding element, and the distance between any two users at any time is not less than the safe distance. When the physical trajectories do not conflict in space and time, the conflict is classified as an assignment misalignment conflict. For ease of implementation, this determination can be broken down into two steps: First, verify the set conditions— Is the set of physical corresponding locations exactly the same? The first step involves reassigning an element; this set condition is a necessary but not sufficient condition for the above judgment criteria. If the set condition is not met, misassignment conflicts can be directly excluded. The second step, under the premise that the set condition is met, involves forwarding the physical trajectory of each user after each candidate non-identity permutation, and verifying the spatiotemporal non-conflict condition of the trajectory (the distance between any two users at any time is not less than a certain value). A positive determination of an assignment misalignment conflict is made only when a non-identical permutation that passes this test exists. It is important to emphasize that situations where only the set condition is met but the trajectory condition is not (e.g., the trajectories of two users still intersect within the channel after reassignment, or the time interval at a certain moment is less than...) are not considered valid. This should not be classified as an assignment misalignment conflict, but should be treated as a non-assignment misalignment conflict.

[0048] A typical example is target order reversal: user , The target characters are located behind the opponent's physical body (i.e.) The target is located On the side of the way they came from, The target is located (On the side of the path from which they came), under the constraints of channel width and safe distance, the two people must cross each other to reach their respective goals, which is geometrically infeasible; therefore, performing a permutation... (The rotation symbol indicates the exchange of indicator 1 and indicator 2) After that, the new goals of the two are located behind each other. They can reach them by turning around and walking in a straight line. The physical trajectories no longer intersect, and the conflict is resolved without imposing any trajectory manipulation on the user. The turning is a movement initiated by the user according to their new narrative goal, not a reset imposed by the system.

[0049] If the conflict is determined to be an assignment misalignment type and a valid permutation pair exists (legality conditions are described in step S103.1 below), then proceed to step S104; otherwise, if the conflict is a non-assignment misalignment type or a valid permutation pair does not exist, the existing reset operation is degraded. Since the degradation path is consistent with the processing method of the prior art, the lower bound of the capability of the solution provided in this application is therefore not lower than that of the prior art.

[0050] In some embodiments, step S103 further includes step S103.1, namely, a valid substitution pair condition check. A valid substitution pair must simultaneously meet the following conditions: (i) the difference between the set eye height and body shape parameters of the corresponding identities of both parties in the substitution pair does not exceed a compatibility threshold, which is preferably 10% (in terms of relative difference); wherein the set eye height and body shape parameters of each identity are set according to the body size of its original driving user, so this condition is equivalent to the height difference and arm span difference between the two users in the substitution pair not exceeding the compatibility threshold; the basis is that: after substitution, the presentation ratio of the user's torso and limbs in the virtual scene and the interactive reach should match their body, and a large height difference or arm span difference will lead to inconsistency between proprioception and visual presentation and be noticed; as for the benchmark eye height, the set eye height of each identity in Mode 2 has been unified, and the difference between the set eye heights of both parties in Mode 1 has converged with the appearance and is lower than the static discrimination threshold. Therefore, no perceptible change in viewpoint height occurs at the moment of submission; the exact value of this threshold is determined by a controlled experiment, that is, users with controlled differences in body size switch driving identities within the blind window, and the threshold is determined by subjective detection rate, which is a conventional experimental method in this field; (ii) the current perceptual budget balance of the substitution for both parties and relevant observers is sufficient to cover one submission, that is, the predicted perceptual budget after submission does not exceed the upper limit of the perceptual budget, and the detection risk is lower than the risk threshold (see step S107 for the accounting rule); (iii) the two parties are not in close interaction with each other, based on the fact that subtle behavioral abnormalities of the other party are most easily detected in close interaction; (iv) the blind window induction parameters of the substitution for both parties and relevant observers (mean of individualized saccade latency). The calibration has been completed in step S101. Optionally, for permutation mode one, a valid permutation pair must also satisfy the following: the difference in appearance parameters between the roles driven by both parties in the permutation pair has converged to the static discrimination threshold. The following conditions apply: juxtaposed static observations are statistically indistinguishable; for substitution mode two, a valid substitution pair must also satisfy the requirement that the appearance parameters and eye height of the characters driven by both parties are pre-defined and uniform. If any of the above conditions are not met, there is no valid substitution pair, and the reset operation is performed according to the previous degenerate procedure.

[0051] Step S104 (Generation of Replacement Plan Table) is executed after step S103 has confirmed that the conflict is an assignment misalignment conflict and that a valid replacement pair exists. The scheduler generates a replacement plan table. The replacement plan table contains four elements: replacement pair. (in , (To replace user metrics for each user); Replacement mode (Mode 1 or Mode 2); Planned submission time ; Extend the upper limit .

[0052] Among them, the planned submission time The selection satisfies a two-way constraint: on the one hand, No earlier than the moment when the shadow representation assembly and submission readiness check may be completed; on the other hand... No later than the moment the conflict materializes Subtracting the safety margin, which covers the time required for the replacement to allow the user to receive a new narrative objective and adjust their actions (e.g., change direction) after submission, the validation observation period, and the time required to perform partial rollback actions (including waiting for their next blind window) on suspected vigilant users, where... The trajectory is estimated by forward extrapolation from step S102. If in If not all preconditions for time-gated submission are met (e.g., a blind window for an observer has not yet been formed), then... Postponed; The continuation and backup path of step S106 share the same time budget. That is, starting from the originally planned submission time, regardless of whether the submission is delayed due to waiting for the preconditions to be met or due to a failed induction and switching to a backup path, all submissions by users involved in the switch must be delayed by no more than [a certain number] days from the originally planned submission time. The task must be completed before the specified time; if it is not completed before the budget is reached, the transaction is aborted and a reset operation is performed. This bidirectional constraint determines that step S104 must generate the plan before step S105, and step S105 must be ready before step S106. The order of steps is thus uniquely determined by physical causality. (Advanced limit) The value should be matched to the source of latency in the backup path.

[0053] After the displacement plan is generated, it is sent to the relevant clients, which then assemble shadow representations accordingly. These relevant clients include the clients corresponding to both users in the displacement pair, and the clients corresponding to the observer users whose local rendering views involve the displacement-related rendering instances. The composition of the shadow representations is as defined above. It is important to emphasize that whether the displacement can be seamless depends on whether the rendering scene before and after the instant of submission is continuous. This is precisely the object to be prepared and checked in step S105. The continuity criterion is position-binding semantics: the physical pose and first-person viewpoint of each user remain unchanged before and after submission, the pose and motion of each rendering instance remain unchanged, and only the identity binding is updated.

[0054] Step S105 (Shadow Representation Assembly and Submission Readiness Check) is executed after step S104, resolving local readiness issues before submission. Specifically, each relevant client preloads the appearance resources, title and interface element binding parameters, and narrative target binding parameters corresponding to the target identity indicated by the replacement schedule into the spare buffer of the local identity mapping table, forming a shadow representation; then, a submission readiness check is performed: confirming that the shadow representation assembly is complete and that each switching-related user is currently in a state where blind window induced stimuli can be applied (e.g., not currently blinking or making large head movements). Since the position binding semantics ensure that the submission itself does not cause any discontinuity in pose or viewpoint, passing the submission readiness check means that all rendering prerequisites for submission have been met, and the only remaining timing issue—the submission time must fall within the blind windows of each observer—is resolved by step S106.

[0055] Step S106 (Blind Window Induced Alignment and Gated Submission) is executed after the shadow representation assembly and readiness checks have passed, addressing the submission timing issue: submission must be performed within each observer's blind window, but the appearance of natural blind windows is random and asynchronous for each observer. If passively waiting for all observers' natural blind windows to appear one by one results in a transition period of several seconds between the first and subsequent submitters, during which the first submitter has already interacted with the replaced role, while the subsequent submitter still sees the unreplaced role, leading to causal misalignment in the interaction semantics. This problem is inherent in multi-observer scenarios where single-observer masking solutions do not exist. This step compresses this transition period by actively creating blind windows.

[0056] The physical basis for actively creating blind windows is the inducibility of reflexive saccades: transient brightness or motion stimuli in the periphery of the visual field can induce a reflexive saccade with a latency of approximately 150 to 250 milliseconds; once the saccade occurs, its inhibition interval (from approximately 50 milliseconds before the saccade occurs to approximately 50 milliseconds after the saccade ends) forms a window in which visual perception is physiologically inhibited. The switching transient submitted by the gating falls within this window without producing a perceptible visual transient. As mentioned earlier, the inducing stimulus itself is a visible transient event; this step does not rely on the assumption that "the stimulus is invisible." The object of the blind window is the switching transient of the identity mapping table, not the inducing stimulus itself. From a timeline perspective, the blind window inducing stimulus in... The process is applied at any time, inducing saccades with a latency of approximately 150 to 250 milliseconds. The saccade inhibition interval covers the period from approximately 50 milliseconds before the saccade occurs to approximately 50 milliseconds after the saccade ends (approximately 120 to 300 milliseconds). The gating submission (active pointer flip) is executed after the saccade is detected and before the end of the saccade inhibition interval. The switching transient falls within the window where visual perception is physiologically inhibited, and the perceptual discontinuity is masked by the blind window.

[0057] In some embodiments, the specific action is as follows: the scheduler calculates the individualized latency mean of each observer. (As determined by step S101), in Blind window induced stimuli are applied to the periphery of each observer's field of vision at constant intervals. Upon detecting a saccade, each client immediately flips the active pointer of its local identity mapping table from the current buffer to the backup buffer (containing the pre-assembled shadow representation), completing the local submission. The active pointer flipping is a sub-millisecond operation, negligible compared to the available submission window after detection (approximately 45 to 53 milliseconds in the shortest saccade case, see the aforementioned timing budget). Therefore, the submission action itself does not constitute a timing bottleneck; the real timing constraint lies in the induced latency, which has already been determined by the calibration parameters in step S101. The reverse pre-compensation. Since each observer's stimulus is compensated according to their individual mean latency, each observer's saccades are ideally aligned to... The actual scanning time was around There is intra-individual latency jitter (typically on the order of tens of milliseconds), therefore the actual submission time of each client revolves around... Adding the temporal distribution of the respective scanning detection delays, the distribution range is the main source of the transition period (see also step S107).

[0058] Optionally, for observers whose induction fails (users do not produce saccades), a backup path is executed: wait for their first natural blind window—including the inhibition interval of spontaneous saccades and the blink inhibition interval (blink inhibition duration 100 to 150 milliseconds, blink interval approximately 4 to 6 seconds in normal natural state; users in a natural viewing state continuously produce spontaneous saccades, and the spontaneous saccade rate... The typical value is about 2 to 3 times per second, so the expected latency for waiting for the spontaneous scanning blind window is in the sub-second range, significantly shorter than the expected latency for waiting for the blinking blind window. The commit is completed within this natural blind window, and its commit is later than the planned commit time. The resulting short transition period is covered by the consistency proxy (see step S107). Backup path and Delayed sharing of the same time budget (See step S104) If the task is not completed even after exceeding the budget, the current transaction will be suspended and a reset operation will be performed. The value of should match the latency of the object being waited for: when the spontaneous scanning blind window is the primary object being waited for, the latency should be sub-second to second. This can cover the expected waiting time; if an observer's backup path can only rely on the blink blind window, the worst-case waiting time needs to be determined based on the measured blink interval distribution (approximately 6 seconds in normal natural conditions; in head-mounted display viewing scenarios, the blink rate can drop to approximately 5 to 7 times per minute, with the worst-case reaching approximately 10 seconds), and the waiting time should be increased accordingly. Alternatively, the transaction can be aborted during the commit readiness check phase, and a reset operation can be performed to ensure that the commit time falls within the true blind window under any circumstances.

[0059] In step S107 (Transitional Consistency Maintenance, Post-Commit Verification and Accounting), transitional consistency maintenance begins when the first local commit occurs, and post-commit verification and accounting are performed after the commit is completed. Regarding the closed loop after the commit is completed, two issues must be addressed: First, if an observer commits late via a backup path, interactions pointing to the replaced role during the transition period must be handled correctly; second, each gating commit is predicated on "the blind window truly exists and is detected correctly," and this predicate itself needs to be verified.

[0060] Regarding the first point, from the first local commit to the last local commit (or transaction abort), the consistency proxy intercepts interactions directed to the replaced role and records them in the interaction log. During the interception period, responses are generated according to the behavior model of the identity before replacement (the behavior model can be implemented using state machines or scripted interaction strategies, etc., which are conventional implementations in this field and are pre-configured by the application content). The interaction commitment constituted by this response is also recorded in the interaction log. When the transaction is closed, it is handled according to the interaction semantic type: environment-oriented actions are directly inherited; dialogue and commitment-type interactions are renegotiation by the consistency proxy with the context of the replaced identity; exclusive actions bound to the identity are invalidated and a neutral reason is fed back. Since the transition period in this application has been compressed to within a hundred milliseconds by step S106 (except for inducing failures, in which case the duration of the transition period is determined by the actual waiting delay of the backup path and is extended to the upper limit). (where the upper bound is the threshold), the burden of the consistency agent is much lower than that of the pure passive waiting scheme, which reflects the pressure reduction effect of step S106 on step S107.

[0061] Regarding the second point: the verification observation period after submission. Within the period, each client collects perceptual continuity indicators when the user's gaze is next switched to another character, including: the deviation of gaze duration from the user's baseline, the occurrence of reverse large-angle rescanning, and the rate of change in pupil diameter; if the weighted sum of the three indicators exceeds a threshold, it is considered a suspected detection. Verification observation period. The value can be, for example, 2 to 5 seconds. If a suspected detection is identified during the observation period, a partial rollback is triggered for that observer: the activity pointer is reversed within the next blind window and the transaction is resubmitted; detection results exceeding the observation period are logged in the audit log and progressively patched in subsequent sessions. Patching is allowed instead of just rolling back because once a detection occurs, it cannot be erased; a rollback can only prevent further exposure, not reverse the already occurred perception. Therefore, the commitment of this step is statistically verifiable and repairable, rather than absolutely undetectable.

[0062] Regarding accounting: Each concealed operation is immediately written into the corresponding user's perceived budget account upon occurrence—each minor manipulation in the preparation loop of step S101 is recorded immediately upon occurrence, and blind window induced stimuli and gating submissions in this transaction are converted and recorded upon completion of submission; step S107 summarizes, verifies, and supplements all concealed operations in this transaction when the transaction is closed. Therefore, the account balance used in step S103 to check the budget balance before submission is always the accurate accumulation of already recorded operations. The perceived budget increment of concealed operations is converted according to intensity as follows: Among them, the first The increment corresponding to the next concealment operation is denoted as... The time of its occurrence is recorded as Geometric gain applied within the statistically imperceptible range defined by the gain threshold has its corresponding perceived budget increment counted as zero; therefore, geometric channels do not consume perceived budget. The dynamics of the perceived budget account are as follows: the contribution of each operation changes over time according to a time constant. Exponential decay, i.e.

[0063] The summation is only performed at the time of occurrence. No later than the current moment Each covert operation is performed; risk detection is performed using logical functions.

[0064] in For positional parameters, depending on the habituation state according to Move to the right; For scale parameters ( ,and (same dimensions) and Fitted from calibration experiments under baseline conditions, The rightward shift pattern is determined by repeated labeling under different habituation states. In the habituation state... When fixed, for It is a strictly monotonically increasing function. This modeling approach is analogous to the motion sickness dosage model, but the modeling object is the perceived risk of covert operations rather than discomfort.

[0065] The phrase "perceived budget balance is sufficient to cover one commit" in step S103.1 refers to the predicted budget balance after the commit. No more than ,and Below the risk threshold This check covers the permutation for both parties and relevant observers. As a result, the frequency of permutation use is subject to quantitative constraints, which is why the "geometric priority" principle in step S102 and the "habitual expansion" loop in step S101 can be connected to form a closed loop: geometric gain saves budget (the budget increment of in-threshold geometric gain is counted as zero), budget constrains the permutation frequency, and habitual expansion raises the budget ceiling. The three constitute a complete resource cycle.

[0066] It should be noted that this application only provides the functional form and calibration method for the perceived budget account and related model parameters, without providing fictitious measured values. The reason is that the individual distribution of such parameters needs to be obtained by the implementer through actual measurement. The calibration method is a psychophysical experimental paradigm well-known to those in the art: recruiting subjects, applying covert operations at random times and intensities in a controlled scenario, recording the subjects' detection responses using two-way forced choice (2AFC) or subjective reporting, and fitting a logistic function to the operation intensity using the detection rate to obtain the location and scale parameters (the baseline fitting provides the base values). With scale parameters The time constant is obtained by repeatedly applying the same operation at different time intervals and measuring the recovery curve of the detection rate. By fitting the data with the cumulative number of operations on the horizontal axis and the detection threshold drift on the vertical axis, the habituation constant can be obtained. and —The detection threshold drift is the location parameter. The right shift amount can be measured by repeatedly fitting the above logistic function under habituated states with different cumulative operation counts. The rightward shift pattern and The lifting patterns share the same batch of habitual calibration data and the same set of constants. , Both are strictly derived from the same source at the calibration level. Similarly, the frame-by-frame change detection threshold... The calibration method is as follows: the character appearance parameters are gradually changed with different amplitudes, and the subthreshold stride is determined by the observer detection rate; the static discrimination threshold... The calibration method is as follows: two roles with controlled differences in appearance parameters are presented side-by-side, and the indistinguishability boundary is determined by the observer's discrimination detection rate; the compatibility threshold (height difference and arm span difference, preferably 10%) is calibrated by having users with controlled differences in body size switch driving identities within a blind window, and the threshold is determined by the subjective detection rate; the weighting and judgment threshold used for verification are calibrated by collecting three perceptual continuity indicators in control trials with known and no switching, and fitting weights and thresholds with the false alarm rate not exceeding a given level as a constraint. All the above calibration experiments are conventional psychophysical methods, which can be reproduced by those skilled in the art without inventive effort; therefore, this application meets the requirement of sufficient disclosure.

[0067] The following examples illustrate how the above steps are connected.

[0068] Example 1: Reversal of target order in a narrow corridor without resetting (main path). The scenario is a 400-square-meter shared site. User, tactical exercise application, the role is a uniformed character with a unified appearance (both Mode 1 and Mode 2 are applicable, this example uses Mode 2, the appearance parameters of each rendered instance are pre-unified). , The narrative target is located exactly behind the other person's physical body (target order reversed), and the two will meet in a passageway with a net width of 1.2 meters within the venue; the moment when the two meet, i.e., the moment when the conflict materializes, is recorded as... .

[0069] Preparation period (session minutes 0 to 3, corresponding to step S101): , They belong to the same preparation group. The blind window calibration loop has measured the mean latency of the two users and the three observers expected to gaze at this channel segment (out of eight users, excluding the permutation pair, whose local rendered views involve this channel segment), for example... Milliseconds; the habituation loop has performed two small concealment operations. and The perception budget caps for all three observers have been increased according to the saturation function, and each manipulation is written into the corresponding budget account immediately upon occurrence; the appearance convergence loop continues to run among the other roles in the same group whose appearances are not unified (in this example, the permutation has unified the appearances of the roles, so there is no need to wait for convergence). Because the preparation period is completed, the subsequent legality check of step S103.1 can pass, and the causal relationship between preparation and use is reflected here.

[0070] Seconds (corresponding to steps S102 and S103): Geometric solvability determination (prediction window) performed by the conflict prediction module in a rolling manner. The first encounter between the two individuals in the channel segment at this moment (seconds) is deemed infeasible: there is no channel with a clear width of 1.2 meters that satisfies both the gain threshold and the safe distance. The opposing trajectories are then used to estimate the moment of conflict materialization. Then, conflict type identification is performed: the target location set is exactly the reverse order of the two assignments, and the set condition is satisfied; further, the trajectories of the two individuals after the substitution (each turning back along the coming path) are pushed forward, and the distance between them at any time is not less than [missing information]. The trajectory conditions are also met, and the conflict is determined to be an assignment misalignment type. Step S103.1 checks are passed (height difference 6%, less than the 10% compatibility threshold; sufficient budget balance for both parties and 3 observers in the replacement; the two parties are not in close interaction; blind window induction parameters for both parties have been calibrated). If any of the above conditions are not met, the system will degenerate and perform a reset; in this example, all conditions are met, so the replacement path is entered. It should be noted that the determination point in this step is naturally given by the rolling prediction: the conflict occurs when... The system enters the prediction window within seconds, makes a judgment immediately, and allows approximately 15 seconds for subsequent processing.

[0071] Seconds (corresponding to steps S104 and S105): The scheduler generates a replacement plan table: replacement pairs The replacement mode is mode two, and the planned submission time is... Seconds (with a safety margin of 13 seconds, covering the time required for the two individuals to turn around and disengage from the conflict after receiving a new target, the verification and observation period, and the possible time for partial rollback handling), extending to the upper limit. Seconds (to ensure submission no later than) Completed in seconds; this 1-second budget matches the sub-second expected latency of the backup path, which primarily waits for spontaneously scanned blind windows. The relevant client receives the displacement schedule (or shadow representation data related to its local rendering): in this example, the relevant clients are both sides of the displacement pair. , There are 5 clients in total, including the corresponding client and the clients corresponding to the 3 observers. The above 5 clients then complete the shadow representation assembly within sub-seconds (in this example, only the identity identifier and narrative target binding parameters are involved, and the appearance resources do not need to be changed; each observer client assembles the target state bound to the character identity involved in its local rendering view), and the submission readiness check passes.

[0072] Milliseconds (corresponding to step S106): Press The average individualized latency, such as milliseconds, is used by the scheduler to... , And three observers were given blind-window induced stimuli respectively; At any given moment, five users involved in the switch enter the saccade suppression zone sequentially within a latency jitter (on the order of tens of milliseconds) within their individual individual timeframes. Upon detecting a saccade, each client flips the activity pointer of its local identity mapping table, and the substitution takes effect. Semantic binding is applied based on location; each user's physical pose and first-person viewpoint remain unchanged at the moment of submission. The appearance, pose, and movement of the character at each location as seen by each observer are continuous (in this example, the appearance is uniform; the substitution does not change the character's appearance, pose, or geometry; only the visual elements of the identity identifier class are updated in relation to the narrative target). , The titles and target prompts in the interface are updated to the narrative target of the swapped identities within the blind window. This target update is presented in the form of narrative events commonly seen in tactical exercises. Whether the two individuals notice this is determined by the verification in step S107 using statistical methods (see the next paragraph). Subsequently, the two individuals autonomously turn according to the new target, return along their respective routes, and no longer enter the passage segment to cross in opposite directions. The conflict is resolved without any trajectory manipulation.

[0073] to Seconds (corresponding to step S107): In this example, there are no induced failures. The transition period is the distribution range of the actual submission time of each client (consisting of individual latency jitter and scan detection delay, less than 100 milliseconds). The consistency proxy does not intercept events. Verification observation period. Within seconds, perceived continuity indicators were collected from 5 users involved in the switchover (2 users and 3 observers for the replacement). The weighted average of all three indicators did not exceed the threshold, and the transaction was closed. The perceived budget increment calculated from blind-window induced stimuli and gating commits was also assessed. This is written into each individual's budget account (each operation during the preparation period is recorded immediately upon its occurrence). In contrast, the only means of existing technology in this scenario is... , A reset is performed, interrupting the drill for 5 to 10 seconds. The verifiable difference in this example is that the number of resets is reduced from 1 to 0, and the number of events identified as suspected to be detected during the verification is 0.

[0074] Based on the same inventive concept, this application also provides a scheduling device for a multi-user virtual reality scene. The solution provided by this device is similar to the method described above, so the specific limitations of the device embodiment can be found in the limitations of the scheduling method above, and will not be repeated here.

[0075] In some embodiments, a computer device (e.g., a terminal), a computer-readable storage medium, and a computer program product are also provided; the computer device stores a computer program, the computer-readable storage medium stores a computer program, and the computer program product includes a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0076] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A scheduling method in a multi-user virtual reality scene, characterized in that, The method applies to multiple users in the same physical location, where each user corresponds one-to-one with multiple virtual characters in a virtual scene, and each user drives their corresponding virtual character; the method includes: Obtain the motion state of each user and the target position corresponding to each virtual character; Based on the movement state of each user and the target position of each virtual character, it is predicted whether a preset conflict condition is met while keeping the current correspondence between each user and each virtual character unchanged; the current correspondence is a one-to-one correspondence formed by the virtual characters currently driven by each user; the preset conflict condition is used to characterize that a conflict will occur in the physical field when each user drives the corresponding virtual character to move to the corresponding target position according to the current correspondence. If the preset conflict condition is met, the current correspondence is switched to a target correspondence that makes the preset conflict condition no longer met, so that each user drives the virtual role corresponding to itself under the target correspondence; the target correspondence is another one-to-one correspondence between each user and each virtual role, which is different from the current correspondence. The switching is performed during the visual suppression period of the relevant user among the users; the relevant user is the user whose virtual character changes during the switching, and the user whose field of vision covers the area where the changed virtual character is located.

2. The method according to claim 1, characterized in that, The step of predicting whether a preset conflict condition is met, based on the movement state of each user and the target position corresponding to each virtual character, while keeping the current correspondence between each user and each virtual character unchanged, includes: Based on the individual movement status of each user, the expected movement trajectory of each user within the prediction period is extrapolated; the prediction period is a time estimated backward from the current moment. Determine whether there exists an adjustment method that satisfies preset geometric adjustment constraints, such that when each user moves along the extrapolated expected motion trajectory, the distance between any two users is not less than a preset safety distance, and none of the users cross the boundary of the physical site; the adjustment method is an adjustment method for the geometric mapping relationship between the virtual scene and the physical site; the preset geometric adjustment constraints are used to limit the adjustment range of the geometric mapping relationship to a range imperceptible to the user; If no adjustment method satisfies the preset geometric adjustment constraint, the preset conflict condition is determined to be satisfied.

3. The method according to claim 1, characterized in that, Before switching the current correspondence to a target correspondence that makes the preset conflict condition no longer satisfied, the method further includes the step of determining the target correspondence; The step of determining the target correspondence includes: Other one-to-one correspondences between each user and each virtual character that are different from the current correspondence are respectively regarded as candidate correspondences; For each of the candidate correspondences, it is determined whether each user can move to the target position of the driven virtual character when each user drives the corresponding virtual character according to the candidate correspondence, and whether the movement trajectories of each user do not conflict with each other. The candidate correspondences in which each user can move to the target position of the driven virtual character and the movement trajectories of each user do not conflict are determined as the target correspondences. In cases where there is no candidate correspondence where each user can move to the target position of the driven virtual character and the movement trajectories of each user do not conflict with each other, a reset operation is performed on each user; the reset operation is an operation that interrupts the user's current movement and guides the user to turn.

4. The method according to claim 1, characterized in that, The step of switching the current correspondence to a target correspondence that makes the preset conflict condition no longer satisfied includes: The users whose virtual roles change under the target correspondence are identified as replacement pairs; the replacement pairs include at least two of the users. Before switching, the appearance parameters and motion style parameters of the virtual characters driven by each user in the displacement pair are brought closer to each other at a rate that is less than perceptible. The switching is performed after the difference between the appearance parameters of the virtual characters driven by each user in the replacement pair is less than a preset static discrimination threshold; the preset static discrimination threshold is the upper limit of the difference in appearance between the two virtual characters that the observer cannot distinguish under normal viewing conditions.

5. The method according to claim 4, characterized in that, The switching is performed after the difference between the appearance parameters of the virtual characters driven by each user in the replacement pair is less than a preset static discrimination threshold, including: Based on the predicted time of conflict occurrence and the preset safety margin, the target time for performing the switch is determined; the target time is no later than the time obtained by subtracting the safety margin from the time of conflict occurrence. If the preconditions for the switching are not all met when the target time is reached, the target time will be postponed within a preset postponement limit; if the preconditions are still not all met after the preset postponement limit is exceeded, the switching will be terminated and a reset operation will be performed on the corresponding user; the reset operation is to interrupt the user's current movement and guide the user to turn. The switching is performed at the target time; Specifically, the delay duration of each observer is pre-calibrated; the delay duration is the time from when the observer receives a transient stimulus at the edge of their visual field to when their eyeballs rotate; the observer is the user whose current visual field covers the area where the virtual character driven by the displacement pair is located. For each observer, the delay time of the observer is advanced before the target time, and a transient stimulus is applied to the edge of the observer's visual field so that the visual inhibition period caused by the observer's eye movement induced by the transient stimulus covers the target time. The switching is performed when the visual inhibition period of each observer covers the target time.

6. The method according to claim 1, characterized in that, Each user corresponds to a client, and the client is used to present the virtual scene to the corresponding user; the step of switching the current correspondence to a target correspondence that makes the preset conflict condition no longer satisfied includes: Each client maintains a currently active driver relationship record and a backup driver relationship record locally; the driver relationship record is used to record the correspondence between each user and each virtual character. Before switching, the target mapping relationship is written into the backup driver relationship record of each client; During the visual suppression period, each client switches the effective pointer from the currently effective driving relationship record to the backup driving relationship record, making the backup driving relationship record the new currently effective driving relationship record.

7. The method according to claim 5, characterized in that, After applying a transient stimulus to the edge of the observer's visual field, the method further includes: For an observer who does not produce eye movement after being subjected to the transient stimulus, the switching is performed for that observer during the naturally occurring visual inhibition period; the naturally occurring visual inhibition period includes the visual inhibition period corresponding to the observer's spontaneous eye movement and the visual inhibition period corresponding to the observer's blink; During the transition period between the earliest and latest executed switch, intercept interaction requests directed to the switched virtual character; the switched virtual character is the virtual character that drives the user change during the switch, and handle them according to the type of the interaction request: environment-directing interaction requests are handed over to the switched virtual character for inheritance processing; dialogue commitment interaction requests are renegotiated according to the context of the switched virtual character; identity binding interaction requests are invalidated, and a neutral reason is fed back to the user who initiated the interaction request.

8. The method according to claim 1, characterized in that, After switching the current correspondence to a target correspondence that makes the preset conflict condition no longer satisfied, the method further includes: During the observation period after the switch, gaze behavior indicators are collected for each observer when they gaze at the switched virtual character; the switched virtual character is the virtual character that drives the user to change during the switch; the observer is the user whose field of vision covers the area where the switched virtual character is located; the gaze behavior indicators include the deviation of the gaze dwell time from the observer's historical baseline, whether there is a reverse rescan, and pupil changes. Based on each observer's gaze behavior index and their own historical baseline, it is determined whether the observer is aware of the switch; If it is determined that the observer has noticed the switch, during the period of visual inhibition of the observer, the correspondence seen by the observer is reversed from the target correspondence back to the current correspondence, and then the current correspondence is switched back to the target correspondence.

9. The method according to claim 3, characterized in that, The method further includes: Before switching the current correspondence to the target correspondence, the cumulative perceptibility of each user's covert operations is obtained; the covert operations include small manipulations performed on each user during the preparation process during the session, transient stimulation operations applied to the user, and correspondence switching operations; the cumulative perceptibility is the degree obtained by accumulating the perceptibility of the corresponding user's historical covert operations over time. Determine whether the cumulative detectability of each user related to the switch meets a preset budget condition; the preset budget condition is used to limit the cumulative detectability of the corresponding user after the switch is executed to not exceed the user's detectability limit, and the probability of the user detecting the switch is lower than a preset risk threshold; the users related to the switch are the users who drive the change of the virtual character in the switch, and the users whose field of vision covers the area where the changed virtual character is located. If the preset budget conditions are met, the switching is performed; if the preset budget conditions are not met, the reset operation is performed on the corresponding user.

10. The method according to claim 9, characterized in that, The method further includes a preparation step that is continuously performed during the session, the preparation step comprising: The appearance and motion style parameters of the virtual characters driven by each user within the same preparation group converge at a rate less than perceptible; the preparation group is a pre-divided user group among the multiple users. A tentative transient stimulus is applied to the edge of the visual field of each user, and the time from the time the user receives the tentative transient stimulus to the time the user's eyeballs rotates is recorded to obtain the delay time of each user. During low-risk periods, small-scale covert operations are performed on each user, such that the upper limit of the cumulative detectability allowed by the preset budget conditions increases with the cumulative number of times the covert operations are performed.

Images