A barrier-free seat multi-modal interaction control system

Abstract

The application relates to the technical field of interactive control, in particular to a barrier-free seat multi-modal interactive control system. The system comprises the following steps: acquiring a voice recognition instruction value, an APP control instruction value and a physical key level value, merging a multi-modal instruction set, aligning with a vehicle safety state value to generate a concurrent interaction matrix; analyzing a control source attribute to obtain a permission set, building an arbitration decision tree, substituting a priority filtering algorithm into an execution screening operation to generate an effective instruction set; converting the effective instruction set into a target position set, calculating a difference value to generate a pose deviation set, extracting a non-zero deviation amount to convert into a target driving sequence; converting the non-zero deviation amount into an error input value, inversely calculating an occupation ratio adjustment value, generating a motor control order for physical adjustment. The application builds a closed-loop control architecture covering perception, decoupling, positioning and inverse calculation compensation, prevents potential risks caused by blind execution of physical interference in the passenger cabin space, and guarantees smooth circulation of the overall in-out cabin interactive link.

Description

Technical Field

[0001] This invention relates to the field of interactive control technology, specifically to a multimodal interactive control system for accessible seating. Background Technology

[0002] In the context of intelligent transformation of vehicle accessibility facilities, the coordinated judgment and closed-loop control of the execution actions and vehicle status across multi-dimensional physical interaction channels is an important technological application direction in the field of intelligent cockpit age-friendly manufacturing. Existing technologies generally adopt a control method that combines a single hard-wired direct drive with mechanical switch limits. That is, by monitoring the analog level changes of a single physical node (such as the outer side of the seat armrest, the inner side of the door trim panel, or the wired handle), when the seat posture indicators such as rotation and unfolding or translation and descent reach a specific travel threshold, within a pre-defined single independent power supply range, the motor action that needs to be cut off is identified and determined by manual vision or basic limit switches. This is currently the mainstream technical means to realize the physical trajectory control and entry / exit assistance of accessible seats.

[0003] However, existing technical solutions have significant shortcomings in terms of the decoupling accuracy of complex cross-channel interactive signals and the continuity of cabin physical motion control. Firstly, their data decision-making models for system safety tracing suffer from isolation and dimensionality gaps. Existing systems are mostly based on the delineation of independent data islands for each external component. When faced with transient speed changes and vehicle posture shifts that objectively exist between occupant interaction and the underlying chassis, they lack the ability to effectively integrate the underlying state data of the entire vehicle network, leading to serious deviations in the location of hazardous intervention and constraint nodes. Secondly, the system only performs passive tracing—an "open-loop power-off" approach—after a local execution module experiences mechanical overload. This is insufficient in situations with highly concurrent operation commands from multiple sources, such as physical button input and remote control via mobile applications. In this state, it is difficult to distinguish the true authority level of each control terminal's independent operation actions to the final physical response, which easily leads to misjudgments caused by interference or masking between accidental input commands and dangerous control signals. Furthermore, the output of maintenance results suffers from homogenization and poor equipment applicability; that is, it only provides a coarse power cut-off or uniform relay fuse protection to the motor drive circuit that is judged to be abnormal, failing to incorporate self-diagnosis and transformation into underlying degradation mechanisms for heterogeneous hidden dangers such as complex usage cycles or hardware and software wear and tear. This results in fault alarms fed back to the user terminal lacking remedial strategic upgrade significance and direct source tracing effectiveness. Finally, the low recognition of concurrent interaction signals and the inability to link underlying vehicle body status in existing technologies make it difficult to solve the systemic logic deadlock problem in the complex interaction scenarios of accessible facilities, easily causing physical interference or unsteady-state jerking in the passenger compartment due to blind control.

[0004] To address this, a multimodal interactive control system for barrier-free seating is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a multimodal interactive control system for accessible seating to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multimodal interactive control system for barrier-free seating, comprising:

[0007] The multimodal perception module acquires voice commands, APP commands, and physical button signals, and merges them into a concurrent command set; acquires vehicle status signals; and performs timing alignment operations on the concurrent command set and vehicle status signals to generate an interactive array. The priority arbitration module parses the control source attributes contained in the interaction array to obtain the permission set; it converts the concurrent instruction set into nodes, and the permission set and vehicle status signals into constraints to build an arbitration decision tree; it substitutes the interaction array into the arbitration decision tree and substitutes the priority filtering algorithm into the arbitration decision tree to perform filtering operations and generate a valid instruction set. The trajectory localization module converts the valid instruction set into preset travel parameters to generate a target position set; it acquires the current pose parameters, performs a difference operation on the current pose parameters and the target position set to generate a pose deviation set; it extracts the non-zero values ​​in the pose deviation set to generate a non-zero deviation amount, and converts the non-zero deviation amount into a target driving sequence. The inverse calculation closed-loop module converts the numerical values ​​bound to the target drive sequence within the non-zero deviation into error input values, substitutes the error input values ​​into the proportional-integral-derivative algorithm to inversely calculate the duty cycle adjustment value, generates a motor control command containing the duty cycle adjustment value, and transmits the motor control command to the actuator for physical adjustment.

[0008] Preferably, the specific generation process of the interactive array includes: receiving voice commands, APP commands, and physical button signals, and merging them into a concurrent command set; receiving vehicle status signals; extracting the channel identifier code and receiving timestamp carried by the concurrent command set; extracting the status message identifier and message timestamp carried by the vehicle status signal; performing a subtraction operation on the receiving timestamp and the message timestamp to obtain the absolute difference, and generating a synchronization time window; extracting the concurrent command set and vehicle status signal within the synchronization time window and merging them to generate an associated command group; merging the receiving timestamp and the message timestamp to construct a trigger timing chain; and performing a recombination and alignment operation on the associated command group along the trigger timing chain to generate the interactive array.

[0009] Preferably, the specific generation process of the permission set includes: extracting control source attributes contained in the interaction array; obtaining preset priority sorting constants to generate a permission mapping vector library; obtaining vehicle speed and gear linkage restriction items to generate a safety intervention vector library; substituting the control source attributes into the permission mapping vector library to perform a matching query operation to generate a basic priority identifier; performing a joint judgment operation on the basic priority identifier and the safety intervention vector library to generate a safety state quantity; extracting activation feature values ​​from the safety state quantity; performing a logical AND operation on the basic priority identifier and the activation feature values ​​to generate a highest intervention mask word; and merging the basic priority identifier, the safety state quantity, and the highest intervention mask word to generate a permission set.

[0010] Preferably, the specific generation process of the effective instruction set includes: converting the concurrent instruction set into nodes, converting the permission set and vehicle status signals into constraints, and building an arbitration decision tree; substituting the interaction array into the arbitration decision tree to extract node status attributes; substituting the priority filtering algorithm into the node status attributes to perform classification operations, generating retained feature items and interference feature items; performing a cut-off operation on the transmission path pointing to the interference feature items; converting the retained feature items into activation values; calculating the safety constraint distribution function of the vehicle status signal under the activation value condition; extracting the parking expectation parameter within the safety constraint distribution function; performing a fusion operation on the parking expectation parameter and the activation value to generate the functional net effect parameter; summarizing the functional net effect parameter and the retained feature items to generate the effective instruction set.

[0011] Preferably, the specific generation process of the pose deviation set includes: converting the effective instruction set into preset travel parameters; performing decomposition operations on the preset travel parameters to generate target rotation parameters, target elevation parameters, and target translation parameters, and merging them into a target position set; obtaining the current pose parameters; performing decomposition operations on the current pose parameters to generate current rotation parameters, current elevation parameters, and current translation parameters; performing subtraction operations on the target rotation parameters and the current rotation parameters to generate rotational motion deviation; performing subtraction operations on the target elevation parameters and the current elevation parameters to generate elevation motion deviation; performing subtraction operations on the target translation parameters and the current translation parameters to generate translational motion deviation; and combining the rotational motion deviation, elevation motion deviation, and translational motion deviation to generate the pose deviation set.

[0012] Preferably, the specific generation process of the target driving sequence includes: obtaining the seat positioning reference and the allowable fluctuation threshold; extracting the rotational motion deviation, lifting motion deviation, and translational motion deviation from the pose deviation set; performing an absolute difference operation on the rotational motion deviation, lifting motion deviation, and translational motion deviation relative to the seat positioning reference to generate an absolute difference set; performing a comparison operation on the absolute difference set and the allowable fluctuation threshold, and extracting values ​​greater than the allowable fluctuation threshold to generate non-zero deviations; performing a mapping operation on the non-zero deviations to generate a motion execution weight set; extracting the peak values ​​in the motion execution weight set and converting them into primary weight ratios; substituting the primary weight ratios into the non-zero deviations to perform a reverse retrieval operation to match the target execution field; and converting the target execution field into a target driving sequence.

[0013] Preferably, the specific process for generating the duty cycle adjustment value includes: extracting the numerical values ​​bound to the target drive sequence within the non-zero deviation and converting them into error input values; obtaining a preset gain constant; performing proportional-integral-differential operations on the error input value and the preset gain constant to generate a reference adjustment base; extracting the channel attribute features contained in the target drive sequence; obtaining the current monitoring safety current limit value associated with the channel attribute features; and substituting the reference adjustment base into the current monitoring safety current limit value to perform a truncation operation to generate the duty cycle adjustment value.

[0014] Preferably, the specific generation process of the motor control command includes: extracting the duty cycle adjustment value and the target drive sequence; obtaining the channel addressing sequence number in the preset hardware configuration file; constructing a data packet structure containing control header code, register address code, channel addressing sequence number and execution parameter tail string; writing the duty cycle adjustment value into the execution parameter tail string; converting the target drive sequence into internal register addressing code and writing it into the register address code; combining the control header code, register address code, channel addressing sequence number and execution parameter tail string to generate the motor control command; and delivering the motor control command to the actuator for physical adjustment.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. After establishing the arbitration transmission network, the concurrent interaction matrix is ​​substituted into it. A strong preemption isolation algorithm is used to isolate conflicting and cascading terms, generating a decoupling matrix. Accessible seating applications are complex, with voice commands, app controls, physical buttons, and vehicle-level states intertwined, creating implicit concurrent interference effects. By employing a strong preemption isolation mechanism, transmission paths pointing to low-authority features are severed, effectively filtering out interference from accidental touches and dangerous situations. This allows for the analysis of the true controlling intent of a single control source on the seat's actions, thus clarifying a clear priority transmission path. This provides a logically sound basis for identifying the source of interaction deadlocks and helps improve control transparency in multimodal concurrent input environments.

[0016] 2. Based on the priority set, the independent contribution of the decoupling matrix to the target seat pose is calculated, generating a command contribution set and a relative weight set. Then, the peak weight ratio within the relative weight set is extracted, and the command contribution set is used for reverse retrieval to output the effective command set. By rigorously quantifying the independent priority weight of each concurrent command, the true driving source currently affecting physical execution is identified. Dynamic targeting based on peak weights ensures the uniqueness and high specificity of subsequent physical adjustments, enabling control commands to focus on the highest priority demands for restrained action mapping. This avoids the risk of interference superposition caused by simultaneous command issuance from multiple control terminals, contributing to the long-term maintenance of a stable operational state in the crew cabin's physical response.

[0017] 3. The independent contribution values ​​bound to the effective instruction set are converted into pose deviation values. The compensation duty cycle value is then calculated using a proportional-integral-derivative (PID) algorithm, generating a motor control command containing the duty cycle adjustment value. This command is then transmitted to the actuator controller bound to the effective instruction set for physical adjustment. By establishing a dimensional conversion bridge using PID, the abstract multimodal motion intent is reverse-analyzed into specific microscopic spatial differences and motor drive adjustment values. This streamlines the transmission logic from upper-level interactive commands to lower-level hardware actions, unifying complex systemic pose displacement tasks into a single actuator's duty cycle fine-tuning. This ensures the drive strategy is smoothly implemented within the current physical trajectory, effectively mitigating dynamic jerking during occupant entry and exit without significantly altering the existing mechanical transmission mode.

[0018] 4. By constructing a globally collaborative closed-loop control architecture encompassing perception, decoupling, positioning, and back-calculation compensation, the complex interplay of variables in multi-channel interactions of accessible seats is effectively addressed from a holistic perspective. By acquiring voice recognition command values, APP control command values, and physical button level values ​​and merging them into a multimodal command set, and aligning it with vehicle safety status values ​​to generate a concurrent interaction matrix, the originally physically isolated input terminals are integrated into the macro-data view of the entire cockpit. Relying on an arbitration transmission network, the intricate permission relationships between commands are elucidated. Priority sets and decoupling matrices are used to establish a rigorous causal mapping between multi-source interaction intentions and underlying hardware indicators. This achieves in-depth communication of cross-channel control signals. By outputting targeted device control commands with deductive justification, deep-seated concurrent interference is systematically resolved. From a macro-level perspective, this prevents potential risks of physical conflicts or systemic deadlocks in the passenger cabin space caused by blind execution, ensuring the smooth flow of the overall entry and exit interaction link. Attached Figure Description

[0019] Figure 1 This is a flowchart of a multimodal interactive control system for accessible seating proposed in an embodiment of this invention application; Figure 2 This is a schematic diagram of the priority arbitration logic proposed in an embodiment of this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1-2 The present invention provides a multimodal interactive control system for barrier-free seats, the specific modules of which are as follows: The multimodal perception module acquires voice commands, APP commands, and physical button signals, and merges them into a concurrent command set; acquires vehicle status signals; and performs timing alignment operations on the concurrent command set and vehicle status signals to generate an interactive array. The priority arbitration module parses the control source attributes contained in the interaction array to obtain the permission set; it converts the concurrent instruction set into nodes, and the permission set and vehicle status signals into constraints to build an arbitration decision tree; it substitutes the interaction array into the arbitration decision tree and substitutes the priority filtering algorithm into the arbitration decision tree to perform filtering operations and generate a valid instruction set. The trajectory localization module converts the valid instruction set into preset travel parameters to generate a target position set; it acquires the current pose parameters, performs a difference operation on the current pose parameters and the target position set to generate a pose deviation set; it extracts the non-zero values ​​in the pose deviation set to generate a non-zero deviation amount, and converts the non-zero deviation amount into a target driving sequence. The inverse calculation closed-loop module converts the numerical values ​​bound to the target drive sequence within the non-zero deviation into error input values, substitutes the error input values ​​into the proportional-integral-derivative algorithm to inversely calculate the duty cycle adjustment value, generates a motor control command containing the duty cycle adjustment value, and transmits the motor control command to the actuator for physical adjustment.

[0022] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0023] Example 1 This application discloses a multimodal interactive control system for accessible seating, see below. Figure 1The specific steps proposed in this invention include: a multimodal perception module, which acquires voice commands, APP commands, and physical button signals, and merges them into a concurrent command set; acquires vehicle status signals; performs timing alignment operations on the concurrent command set and vehicle status signals to generate an interaction array; a priority arbitration module, which parses the control source attributes contained in the interaction array to obtain a permission set; converts the concurrent command set into nodes, and converts the permission set and vehicle status signals into constraints to build an arbitration decision tree; substitutes the interaction array into the arbitration decision tree, and substitutes the priority filtering algorithm into the arbitration decision tree to perform filtering operations to generate a valid command set; The trajectory positioning module converts the valid command set into preset travel parameters to generate a target position set; it acquires the current pose parameters, performs a difference operation on the current pose parameters and the target position set to generate a pose deviation set; it extracts non-zero values ​​from the pose deviation set to generate non-zero deviation quantities, and converts the non-zero deviation quantities into a target drive sequence; the inverse calculation closed-loop module converts the values ​​bound to the target drive sequence within the non-zero deviation quantities into error input values, substitutes the error input values ​​into the proportional-integral-derivative algorithm to inversely calculate the duty cycle adjustment value, generates a motor control command containing the duty cycle adjustment value, and transmits the motor control command to the actuator for physical adjustment.

[0024] Furthermore, voice commands, APP commands, and physical button signals are acquired and merged into a concurrent command set; vehicle status signals are acquired; timing alignment operations are performed on the concurrent command set and vehicle status signals to generate an interaction array; corresponding to the multimodal perception module; the specific implementation process includes: The system receives voice commands, APP commands, and physical button signals, and merges them into a concurrent command set; it receives vehicle status signals; it extracts the channel identifier and receiving timestamp carried in the concurrent command set; it extracts the status message identifier and message timestamp carried in the vehicle status signal; it performs a subtraction operation on the receiving timestamp and the message timestamp to obtain the absolute difference, and determines whether the absolute difference falls within a preset synchronization time window; it extracts the concurrent command set and vehicle status signal that successfully fall within the synchronization time window and merges them to generate an associated command group; it merges the receiving timestamp and the message timestamp to construct a trigger timing chain; and it performs a recombination and alignment operation along the trigger timing chain to generate an interactive array.

[0025] Specifically, the generation process of the interactive array is as follows: The state association perception module needs to capture three key physical input data from the heterogeneous hardware terminals and machine bus in the passenger cabin: In the voice acquisition stage, a microphone array is deployed on the cabin ceiling to locate and capture sound sources from multiple directions. The sampling frequency parameter is usually set between 16,000 and 48,000 Hz, and the selection of the frequency parameter is strictly limited based on the distribution range of human voice frequency bands and the computing bandwidth of the digital signal processing chip; the noise reduction depth parameter is usually set between 20 and 40 dB, and the selection of the depth parameter is based on a comprehensive evaluation of tire noise index and air conditioning blower noise during high-speed vehicle operation; the offline wake-up response latency is usually limited to within 200 to 500 milliseconds, and the selection of the time parameter is based on a comprehensive evaluation of smooth interactive experience and the computing power limit of the dedicated neural network acceleration chip. Generally, the more complex the background noise, the more computing power is required for the filter, and each feature extraction requires an additional 10 to 15 milliseconds of processing time to ensure recognition accuracy. The voice recognition chip transmits the actually extracted control features to the receiving port, and the value is the voice recognition command value.

[0026] In the mobile device connectivity stage, the wireless communication module establishes a low-level data link with the smartphone, receiving adjustment commands from the remote end via a radio frequency antenna. During this process, the Bluetooth communication frequency band is typically set to 2.4 GHz. The selection of this frequency band is rigorously determined by comprehensively considering the radio frequency transmission power of the in-vehicle equipment, the electromagnetic penetration capability within the cabin, and resistance to multipath fading. The standard data packet encryption level requires a high-level encryption standard of 256 bits. The encryption parameters are selected based on automotive-grade information security protocols, replay attack prevention requirements, and user privacy protection standards. The protocol stack parser built into the communication module sends the calculated valid command message to the receiving port, generating APP control command values. In the physical contact interaction stage, the microswitch contacts close, transmitting the analog voltage signal to the analog-to-digital converter via a hard-wired loop. The voltage sampling threshold is typically set between 3.3 and 5.0 volts. The specific parameter selection is based on dynamic evaluation and setting of the fluctuation range of the current in-vehicle low-voltage power supply network, the wiring harness impedance voltage drop, and the tolerance margin to prevent electrostatic discharge. The analog voltage signal is processed into digital logic levels by the analog-to-digital converter circuit and sent to the receiving port to generate the physical button level values. Three heterogeneous data sources are received by the communication interface, then concatenated in the cache memory to form a multimodal instruction set.

[0027] Synchronization: The receiving port receives vehicle safety status values ​​sent from the final test vehicle bus. These vehicle safety status values ​​serve as a benchmark for the current vehicle's driving limitations. It is important to note that the transmission of status messages is affected by the number of bus gateway routing nodes. Typically, each additional gateway forwarding link increases the communication latency of the underlying control messages by approximately 2 milliseconds. Therefore, message capture is performed in the final stage after the main controller at the bus end has completed parsing.

[0028] Because speech recognition, wireless communication, hard-wired levels, and bus polling often occur with intervals of several message slots or even seconds, strict source alignment logic is required. Extract the channel identifier code carried in the multimodal instruction set data packet header and record the reception timestamp indicating the end of the hardware terminal's receiving action. Synchronize and extract the status message identifier carried in the vehicle safety status value data packet header and the message timestamp from the testing phase.

[0029] Subsequently, a subtraction operation is performed on the received timestamp and the message timestamp to obtain the absolute difference, and it is determined whether this absolute difference falls within a preset synchronization time window. The two sets of data are considered to have temporal concurrency attributes if and only if the absolute difference is within the synchronization time window. Based on the temporal concurrency attributes, in the underlying relational database, the multimodal instruction sets and vehicle safety status values ​​within the synchronization time window are extracted and merged to generate a related instruction group. A strict triggering timing chain is constructed by combining the order of the received timestamp and the message timestamp; the dwell time difference of the related instruction group between two adjacent bus polls is calculated synchronously. If the difference exceeds a preset message drift tolerance window, the system will retrieve the device maintenance logs of the corresponding gateway within the time period; if the logs show that node disconnection and restart or parameter recalibration occurred during the period, the status set is marked as contaminated failed data, and potentially logically out-of-order dirty data is removed. The threshold parameter of the preset synchronization time window is typically set between 50 and 150 milliseconds. The specific selection method for this threshold parameter is based on a comprehensive evaluation of the maximum concurrent time difference of multimodal heterogeneous communication links and the red line of in-vehicle human-machine interaction haptic delay.

[0030] Finally, the cleaned and filtered group of related instructions is reorganized and aligned along the triggering sequence chain, with the row dimension representing independent interaction cycle samples and the column dimension representing different terminal variables, and the concurrent interaction matrix is ​​finally output.

[0031] By extracting the channel identifier code to determine the control source attribute, and combining the receiving timestamp and the message timestamp to construct the trigger timing chain, the scattered multimodal instruction set and vehicle safety status value are reorganized and aligned in time sequence, eliminating data silos between input terminals and providing objective and true underlying data support for subsequent permission intervention and stripping.

[0032] Furthermore, the control source attributes contained within the interaction array are analyzed to obtain the permission set; the concurrent instruction set is converted into nodes, and the permission set and vehicle status signals are converted into constraints to build an arbitration decision tree; the interaction array is substituted into the arbitration decision tree, and the priority filtering algorithm is substituted into the arbitration decision tree to perform filtering operations and generate a valid instruction set; corresponding to the priority arbitration module; see [reference]. Figure 2 The specific implementation process includes: Extract control source attributes contained in the interaction array; obtain preset priority sorting constants to generate a permission mapping vector library; obtain vehicle speed and gear linkage restrictions to generate a safety intervention vector library; substitute control source attributes into the permission mapping vector library to perform matching query operations and generate basic priority identifiers; perform joint judgment operations on the basic priority identifiers and the safety intervention vector library to generate safety state variables; extract activation feature values ​​from the safety state variables; perform logical AND operations on the basic priority identifiers and activation feature values ​​to generate the highest intervention mask word; merge the basic priority identifiers, safety state variables, and the highest intervention mask word to generate a permission set.

[0033] Specifically, the process of generating permission sets includes: To accurately define the intervention permissions of different control commands, the system first extracts the control source attributes contained in the interaction array and splits them into two parts: extracting the hardware identifier of the command channel to establish a front-end mapping array, and obtaining the corresponding underlying communication protocol message header to establish a link tag array. A masking filtering algorithm is then used to process the front-end mapping array: extracting the characteristic bit segments and checksums of each column of identifiers, performing a bitwise XOR operation on each original value to remove redundant control bits, and uniformly mapping the three sets of data protocols with significantly different baud rates and clock frequencies—voice wake-up, remote connection, and physical hardwire—to a standard register address space of 16 bits with the high-order bits set to 0.

[0034] Based on the above array, a permission mapping vector library is generated by obtaining preset priority sorting constants, and a matching query operation is performed. To ensure that the vector library has absolutely strict mutual exclusion judgment capability, its memory addressing structure includes the following hierarchical settings: the input addressing layer is set to 3 matching base addresses, corresponding to the aforementioned three control sources: voice, remote, and physical. Subsequently, 4 lookup comparison hidden layers are connected, and their internal data unit capacities are set to 128 bytes, 64 bytes, 32 bytes, and 16 bytes respectively. The selection of these 4-level capacity parameters is based on a comprehensive setting of the vehicle microprocessor's L1 cache size and the access overhead of static random access memory. After each lookup comparison hidden layer, a hardware latch is cascaded to solve the problem of transient data loss on the bus by utilizing its holding characteristics. Between the 1st and 2nd hidden layers, a watchdog anti-lock layer with a timeout threshold of 50 milliseconds is inserted. The selection of this threshold parameter is based on the message delay fluctuation of the CAN bus under extreme operating conditions at the highest load rate, with the aim of performing a hot reset operation to prevent controller crashes when the matching gets stuck in an infinite loop. Finally, a result layer containing only one output register is connected to output a determined basic priority identifier.

[0035] During the computation phase of generating the safety intervention vector library by acquiring vehicle speed and gear linkage limitations, various state constraint parameters are set: the parking gear (P gear) level characteristic is set to logic high; the zero speed tolerance threshold is set to within 0.5 kilometers per hour. A vehicle safety linkage locking mechanism is introduced here. When the bus feedback indicates that the vehicle is not in P gear or the vehicle speed exceeds the zero speed tolerance threshold, the system determines that there is a risk of occupant fall or mechanical interference. The safety intervention vector library will directly output the highest-level intervention command, forcibly locking the issuance of all subsequent seat actions; simultaneously, a 20-millisecond interrupt preemption mechanism is introduced. When writing the safety intervention vector library to memory, a direct memory access controller is used, utilizing its data transfer mechanism, which is independent of central processor intervention, to dynamically update the underlying vehicle state set, accelerating the reporting of safety signals.

[0036] In each alternation between routine command issuance and safety status change, the specific calculation formula derivation process for the joint judgment operation is as follows: The actual physical value of the chassis speed is extracted from the safety intervention vector library, and the gear status value output by the main controller in a single analysis is extracted. Next, a comparison operation is performed between the physical speed value and the zero speed tolerance threshold to obtain the Boolean value between them. Then, this Boolean value is XORed with the gear status value to obtain the comprehensive safety feature value of the sample. Afterwards, for multiple sets of comprehensive safety feature values ​​issued by all bus nodes within the current time window, a cascaded bitwise AND operation is performed sequentially (ensuring all samples are in a safe state before outputting a high level), and the final calculation result is output as the safety status quantity for this operation. This status quantity is generated as a fixed Boolean matrix and then sent to the local execution node in the workshop. The execution node only calls the fixed matrix to perform logical matching and does not perform online polling updates, ensuring that the algorithm's response time is much less than the mechanical response delay of the physical motor rotor.

[0037] Based on the generated security state variables, the activation feature values ​​within the security state variables are extracted. Here, a hardware logic gate mapping method is used for the calculation.

[0038] A logical AND operation is performed on the safety status quantity and the basic priority identifier. Specifically, an intervention mask byte is superimposed on the basic priority identifier data packet of the current batch. The binary width of this intervention mask byte is set to 8 bits, and its selection method is jointly determined based on the length of the general fault diagnostic code of the relevant body control module and the highest forwarding word length of the bus gateway to ensure that the intervention has physical practical significance. After padding the 8-bit intervention mask byte with zeros to extend it to 16 bits, a bitwise logical AND operation is performed with the basic priority identifier to convert it into an absolute truncated signature code on the instruction transmission path.

[0039] When performing joint decision operations, the microcontroller input requires a complete timing feature vector. Therefore, the system assigns an activation feature value of all 1s to the current high-priority parameter (such as the anti-pinch collision signal), while strictly clearing the other two low-priority parameters (such as voice interaction and APP control) to a logic low level. These three parameters are then combined and input into the arbitration decision tree to obtain the truncated result of a single polling output. This new truncated result is assigned to the highest intervention mask word. Since the hardware time for the logic operation is only one machine cycle, the mask word calculated in this way approximates the instantaneous hardware response rate at that point in time.

[0040] Using the same traversal operation method, the mask word mapping operation is performed separately through the other control source attributes. The intervention mask word for the door opening state is set to hexadecimal FF (determined based on the closing impedance characteristics of the door position micro switch); the intervention mask word for the gear shift state is set to hexadecimal AA (determined based on the message encoding rules of the transmission control unit TCU). Thus, the corresponding intervention command and cutoff command are obtained respectively.

[0041] Subsequently, the basic priority identifier, security status quantity, and highest intervention mask word are merged. The values ​​of these three identifiers are then summarized and encapsulated to obtain the permission set.

[0042] By substituting the control source attributes into the permission mapping vector library to perform queries, and by extracting security state variables and basic priority identifiers to perform logical AND operations to generate the highest intervention mask word, the logical deadlock caused by concurrent preemption of multi-source instructions is alleviated, and the real system intervention priority is extracted.

[0043] The concurrent instruction set is converted into nodes, and the permission set and vehicle status signals are converted into constraints to build an arbitration decision tree. The interaction array is substituted into the arbitration decision tree to extract node status attributes. The priority filtering algorithm is substituted into the node status attributes to perform classification operations, generating retained feature items and interference feature items. The transmission paths pointing to interference feature items are cut off. The retained feature items are converted into activation values. The safety constraint distribution function of vehicle status signals under activation value conditions is calculated. The parking expectation parameters within the safety constraint distribution function are extracted. The parking expectation parameters and activation values ​​are fused to generate the net functional effect parameters. The net functional effect parameters and retained feature items are summarized to generate an effective instruction set.

[0044] Specifically, the process of generating an effective instruction set is as follows: After acquiring concurrent interaction commands, the observed data may contain unwanted interference caused by accidental human intervention. To obtain clean control signals, a priority filtering algorithm based on arbitration logic needs to be applied. Considering the distinction between high-privilege preemption and low-privilege queuing in the underlying hardware architecture, the preemption operation requires forcibly cutting off all input paths pointing to specific interference variables at the hardware bus level.

[0045] Based on this principle, the system converts the concurrent instruction set into nodes of a directed acyclic graph. When determining the direction of directed edges between nodes, the system strictly follows the physical timing chain of signal queuing and the safety control hierarchy within the vehicle microcontroller to assign the direction of the hierarchical arrows. Specifically, a four-level preemption tree is constructed from high to low: vehicle collision and system failure emergency braking are the highest-level nodes, pointing unidirectionally to local physical button manual operation nodes; physical button nodes point unidirectionally to voice interaction control nodes; and voice control nodes point unidirectionally to mobile terminal APP remote control nodes. Through the above hierarchical division, the pre-requisite physical requirements for permission arbitration are met. Subsequently, the permission set and vehicle status signals are converted into corresponding constraints, which are used as feature weights of the corresponding directed edges to build an arbitration decision tree.

[0046] The interactive array is substituted into the arbitration decision tree to extract node state attributes. A priority filtering algorithm is used to classify the node state attributes, generating retained and interfering feature items. Based on the preemption criteria at the control system's lower level, all out-of-level propagation paths pointing to interfering feature items are identified. To cut off input paths at the code level, the program generates an intervention mask matrix of the same dimension as the node matrix, forcibly setting the matrix elements corresponding to out-of-level propagation paths to logic 0, while retaining the rest as logic 1. Subsequently, a Hadamard product is performed with the original node matrix to complete the hard truncation operation of the topology connection.

[0047] Simultaneously, retained features are converted into active values. Interference features are set as forced fixed constant values. The specific selection method for fixed constant values ​​is as follows: extract the level distribution sequence in the sleep state from the historical work log, and use the Gaussian kernel density function (with the bandwidth dynamic parameter set to 0.15, determined based on the ripple tolerance upper limit estimation of the vehicle power supply network) to plot a level probability curve. The zero-point voltage value corresponding to the highest peak of the absolute probability density on the curve is extracted as the fixed constant value.

[0048] Under the strict conditions of cutting off the transmission path and setting fixed constant values, the variance of the current bus load is detected. If the standard deviation is greater than 0.05%, Gaussian white noise with a mean of 0 and a variance of 0.01 is added to the activated value to simulate weak electromagnetic interference in the field. Finally, under the truncated topology condition, the safety constraint distribution function of the vehicle status signal under the activated value condition is calculated. The specific calculation process is as follows: the vehicle status signals of the most recent 100 normal completion cycles are extracted from the underlying relational database as a reference feature library. The gear position and vehicle speed value of each record in the feature library are extracted, and the restricted state is forcibly replaced with the fixed constant value obtained above. The feature library is input into the model after truncating the skip path one by one and batch forward propagation is performed. The 100 output state scalars constitute a discrete distribution set. The arithmetic mean of these 100 predicted values ​​is calculated to obtain the parking expectation parameter of the safety constraint distribution function.

[0049] Simultaneously, the historical baseline expected value is calculated when all control nodes are at fixed constant values. The parking expected parameter is then subtracted from the historical baseline expected value to obtain the single action drive quantity purely caused by high-privilege input. Subsequently, the single action drive quantity and the activation value are fused, and their dimensions are inversely mapped to the underlying drive dimension, formally defined as the functional net effect parameter. The aforementioned standard deviation threshold is set based on the statistical upper limit of the static background noise of the field wiring harness resistance.

[0050] Following the same logic, all input channel data within the concurrent instruction set are processed sequentially in turn. During each rotation, low-privilege nodes are set to fixed constants and their transmission paths are cut off. Through independent fusion calculations, the independent net functional effect parameters for each channel are obtained. The net functional effect parameters and retained feature terms are then aggregated to generate an effective instruction set for final device adjustment.

[0051] By using a priority filtering algorithm to cut off the transmission path pointing to interference features, setting the interference features to fixed constant values, and then calculating the parking expectation parameters of the safety constraint distribution function, the influence of crosstalk between signals is eliminated, and pure net functional effect parameters are obtained, thereby improving the accuracy of downstream action execution.

[0052] Further, the effective instruction set is converted into preset travel parameters to generate a target position set; the current pose parameters are obtained, and a difference operation is performed between the current pose parameters and the target position set to generate a pose deviation set; non-zero values ​​in the pose deviation set are extracted to generate non-zero deviation quantities, and the non-zero deviation quantities are converted into target driving sequences; corresponding to the trajectory localization module; the specific implementation process includes: The effective instruction set is converted into preset travel parameters; decomposition operations are performed on the preset travel parameters to generate target rotation parameters, target elevation parameters, and target translation parameters, which are then merged into a target position set; the current pose parameters are obtained; decomposition operations are performed on the current pose parameters to generate current rotation parameters, current elevation parameters, and current translation parameters; subtraction operations are performed on the target rotation parameters and the current rotation parameters to generate rotational motion deviation; subtraction operations are performed on the target elevation parameters and the current elevation parameters to generate elevation motion deviation; subtraction operations are performed on the target translation parameters and the current translation parameters to generate translational motion deviation; and the rotational motion deviation, elevation motion deviation, and translational motion deviation are combined to generate a pose deviation set.

[0053] Specifically, the process of generating the pose deviation set is as follows: The data contained within the valid instruction set is extracted and converted into preset travel parameters, and the current pose parameters fed back from the underlying hardware module are obtained. Decomposition operations are performed on the preset travel parameters to generate target rotation parameters, target elevation parameters, and target translation parameters, which are then merged into a target position set. Simultaneously, decomposition operations are performed on the current pose parameters to generate current rotation parameters, current elevation parameters, and current translation parameters. The specific calculation method for each action deviation is to perform a direct numerical subtraction operation between the target parameter parsed from each execution dimension and its corresponding current parameter to restore the spatial trajectory dimension of the data. Specifically, the calculation process is as follows: subtract the current rotation parameter from the target rotation parameter; the difference is defined as the rotational action deviation; subtract the current elevation parameter from the target elevation parameter; and subtract the current translation parameter from the target translation parameter.

[0054] In this embodiment, the historical fluctuation variance of the corresponding sensor for each parameter is extracted to calculate the historical fluctuation confidence coefficient. This historical fluctuation confidence coefficient is then converted into confidence weight coefficients and assigned to each action deviation. Weighted scaling is performed, and the data is then combined to generate a pose deviation set with a confidence gradient. After obtaining the initial action deviations for rotation, elevation, and translation, the underlying relational database is retrieved to extract the raw sampling data streams from the physical sensors corresponding to these three actions in the most recent 50 polling cycles. The historical fluctuation variance and probability distribution characteristics of these three parameter sequences are calculated separately. Histogram statistics are used to obtain the probability distribution sequence of the sequence. The probability value of each interval is multiplied by the natural logarithm of the corresponding probability value. The products of all intervals are then summed, and the negative value of the sum is defined as the historical fluctuation confidence coefficient value of the corresponding parameter. A larger historical fluctuation confidence coefficient indicates that the data collected by the corresponding sensor is more divergent and has higher uncertainty (containing more physical measurement noise). The sum of the historical fluctuation confidence coefficients of the three parameters is calculated. The ratios obtained by dividing each historical fluctuation confidence coefficient by the sum are then subtracted from the value 1 to obtain the complements of the three parameters. Subsequently, each complement is divided by the sum of the three complements, and a strict normalization process is performed to construct the confidence weight coefficients for each parameter. Simultaneously, the initial motion deviations are multiplied by their corresponding confidence weight coefficients for penalty scaling, generating weighted motion deviations for ranking comparison: rotation-weighted motion deviation, rise-fall-weighted motion deviation, and translation-weighted motion deviation. The rotation-weighted motion deviation, rise-fall-weighted motion deviation, and translation-weighted motion deviation generated by the confidence weight coefficient penalty scaling are combined to generate the final pose deviation set, which is directly used for subsequent PID physical inverse calculations.

[0055] By subtracting the corresponding current parameter from each target parameter, the deviation of each action is obtained. The preset command stroke is deeply integrated with real-time physical feedback, which quantifies the real spatial difference of the final posture of the barrier-free seat by different dimension drive motors, and provides a unified benchmark for the establishment of equipment control commands.

[0056] Obtain the seat positioning reference and allowable fluctuation threshold; extract the rotational motion deviation, lifting motion deviation, and translational motion deviation from the pose deviation set; perform absolute difference calculations on the rotational motion deviation, lifting motion deviation, and translational motion deviation relative to the seat positioning reference to generate an absolute difference set; perform a comparison operation on the absolute difference set and the allowable fluctuation threshold, and extract values ​​greater than the allowable fluctuation threshold to generate non-zero deviations; perform mapping operations on the non-zero deviations to generate a motion execution weight set; extract the peak values ​​in the motion execution weight set and convert them into primary weight ratios; substitute the primary weight ratios into the non-zero deviations to perform a reverse search operation to match the target execution field; convert the target execution field into a target driving sequence.

[0057] Specifically, the generation process of the target-driven sequence is as follows: Before activating the control mechanism, it is necessary to obtain the seat positioning reference and allowable fluctuation threshold. Regarding the selection method of the seat positioning reference, it is set to the spatial origin coordinate value of 0, and the parameter selection is strictly determined according to the absolute zero point of the three-dimensional coordinate system in the accessible seat design drawings; regarding the selection method of the allowable fluctuation threshold, independent basic thresholds with corresponding dimensions are set for rotation, height, and translation (for example, the rotation tolerance is set to 2 degrees, and the height / translation tolerance is set to 5 millimeters), and each is compared.

[0058] Subsequently, a FIFO sliding data window of length 20 is established to continuously collect the measured feedback coordinates from the physical sensors over the most recent 20 polling cycles. The sample standard deviation of the coordinate sequence within the window is calculated and compared with the baseline standard deviation of the device under the historical best controlled state in the underlying database. If the current sample standard deviation is less than or equal to the baseline standard deviation, it indicates that the mechanical transmission is stable, and the static baseline threshold is directly taken as the allowable fluctuation threshold. If the sample standard deviation is greater than the baseline standard deviation, it indicates that the transmission chain exhibits a variance inflation trend, and the drift ratio is calculated (specifically, the drift ratio is obtained by calculating the ratio of the sample standard deviation to the baseline standard deviation). The dynamic penalty coefficient is set to be equal to 1.5 minus 0.5 multiplied by the drift ratio, and the penalty coefficient is forcibly limited to a closed interval between 0.6 and 1.0 through a threshold function.

[0059] Finally, the static base threshold is multiplied by the penalty coefficient to obtain the tightened value, which is defined as the allowable fluctuation threshold for the current batch. The physical basis for forcibly truncating the lower limit to 0.6 is that when the allowable fluctuation threshold is tightened to 60% of the original base threshold, the fluctuation band width is close to the inherent execution tolerance band of the motor itself. A value less than 0.6 will cause normal gear clearance tolerance to be misjudged as abnormal, triggering the system to continuously send high-frequency micro-compensation commands to the control chip, thereby causing equipment resonance and actuator wear.

[0060] Extract the rotational, lifting, and translational motion deviations from the pose deviation set. Calculate the absolute value of the difference between each of these deviations and a pre-set seat positioning reference value. When the calculated absolute difference exceeds the allowable fluctuation threshold set above, it indicates that the current transmission node is at risk of deviating from the target trajectory. Extract values ​​with absolute differences greater than the allowable fluctuation threshold to generate non-zero deviations, thereby triggering subsequent targeted positioning calculations. Extract the specific values ​​of the weighted motion deviations (including rotation, lifting, and translation parameters) generated above, and extract the preset maximum rotational, lifting, and translational strokes from the equipment configuration set. Divide the absolute values ​​of the weighted motion deviations for rotation, lifting, and translation by their corresponding maximum stroke values ​​to obtain three dimensionless percentage deviation rates. Then, perform a summation operation on these three percentage deviation rates; the sum is the absolute value of the total weighted deviation.

[0061] Next, the corresponding weight ratios are calculated. The specific calculation process is as follows: The absolute values ​​of the rotation-weighted action deviation, the lifting-weighted action deviation, and the translational action deviation are calculated as a percentage of the absolute value of the total weighted deviation. This generates the corresponding weight ratio representing the percentage of responsibility each action bears for the overall spatial deviation. Subsequently, these corresponding weight ratios are merged and mapped one-to-one with a dictionary containing the names of specific drive motors to generate an action execution weight set. Using a numerical sorting algorithm, the numerical values ​​of all elements in the action execution weight set are compared, and the item with the highest numerical proportion is selected. The item with the largest percentage of responsibility is extracted separately and converted into the primary weight ratio. Substituting the primary weight ratio into the underlying data structure, a reverse search operation is performed on the non-zero deviations that retain the original true data, matching the physical execution motor field corresponding to the maximum trajectory offset, and establishing it as the target execution field.

[0062] Subsequently, the target execution field is converted into machine code that conforms to the underlying communication protocol and output as the target drive sequence.

[0063] By introducing a permissible fluctuation threshold as a precondition, and calculating the corresponding weight ratios of each item when the threshold is triggered, the primary weight ratio is selected to transform the target driving sequence, which effectively avoids over-adjustment within the range of small normal fluctuations and dynamically and uniquely locks the core abnormal link.

[0064] Furthermore, the numerical values ​​bound to the target drive sequence within the non-zero deviation range are converted into error input values. These error input values ​​are then substituted into the proportional-integral-derivative (PID) algorithm to calculate the duty cycle adjustment value, generating a motor control command containing the duty cycle adjustment value. This motor control command is then transmitted to the actuator for physical adjustment; this corresponds to the inverse calculation closed-loop module. The specific implementation process includes: Extract the numerical values ​​bound to the target drive sequence within the non-zero deviation range and convert them into error input values; obtain the preset gain constant; perform proportional-integral-differential operations on the error input values ​​and the preset gain constant to generate the reference adjustment base; extract the channel attribute features contained in the target drive sequence; obtain the current monitoring safety current limit value associated with the channel attribute features; substitute the reference adjustment base into the current monitoring safety current limit value to perform amplitude truncation operations to generate the duty cycle adjustment value.

[0065] Specifically, the process for generating the duty cycle adjustment value is as follows: Upon receiving the target execution field confirmed by the preceding node, the system extracts the inter-bound action execution deviations within the non-zero deviation range based on the string index of the target execution field (retaining the original positive and negative signs), and converts these values ​​into the pose error input value that needs to be compensated. Simultaneously, it extracts the mechanical transmission proportional coefficient value mapped to the target execution field within the device configuration set. The system then judges the absolute value of the pose error input value, checking whether it is less than or equal to 15% of the maximum physical travel reference value (i.e., the system's preset proportional-calculus local linear effective interval). Only when this condition is met can the proportional-calculus algorithm be directly used to calculate the pulse width adjustment base; if it exceeds the 15% interval, it indicates that a large deviation has caused the motor linearization assumption to fail. In this case, the system intercepts the reverse calculation process, no longer issuing automatic physical adjustment commands, but triggering a global high-level alarm, requiring manual intervention to investigate major equipment hardware faults.

[0066] The specific calculation formula for the pulse width adjustment base is as follows: Divide the extracted pose error input value by the corresponding mechanical transmission proportional coefficient value. The quotient value obtained represents the additional adjustment base amount that the actual physical machine tool needs to supplement and reduce in order to offset the spatial trajectory deviation (e.g., how many degrees of rotation need to be added or removed, and how many millimeters of translation need to be added or removed).

[0067] Subsequently, a branch discrimination structure is established for different motor mechanisms, and the corresponding pulse width mapping constant modulus is retrieved: If the mechanism is identified as a rotary actuator, the torque constant is extracted from the moment of inertia mapping table. The mapping table is an empirical quick reference table pre-derived based on the mechanical curve of a specific seat frame within the 0 to 90 degree range, using discrete integral fitting in five-degree increments, and records the rate of change of gear transmission at specific angle points.

[0068] If the mechanism is identified as a lifting actuator, a dynamic gravity compensation constant is extracted based on Newton's second law. The system extracts the preset seat load mass value, multiplies it by the gravitational acceleration constant (9.8 m / s²) to calculate the gravity load, multiplies the calculated gravity load by the lead parameter of the lifting shaft screw, and divides the product by (2 multiplied by pi, the reduction ratio, and the transmission efficiency) to calculate the current required holding torque, which is used as the basic lifting compensation constant. Subsequently, combined with the preset worm gear reduction ratio and transmission efficiency of the lifting shaft, the current required holding torque is derived. The torque-duty cycle mapping coefficient built into the device is extracted; the pulse width adjustment base is multiplied by the required holding torque, and then multiplied by the torque-duty cycle mapping coefficient to obtain the basic duty cycle adjustment value.

[0069] If it is identified as a translation actuator, since the translation friction partial derivative in the equipment configuration set has eliminated the physical disturbance value during the generation stage, it already has real physical dimensions. The pulse width modulation compensation value with duty cycle dimensions can be obtained by directly dividing the pose error input value by the translation friction partial derivative.

[0070] For lifting actuators, the inverse calculation of pulse width modulation compensation values ​​needs to be segmented in conjunction with the mechanism's gravity-driven work state: The system extracts the current measurement value of the current motor. If the current measurement value is within a stable and uniform speed window (i.e., the range of 0.5 amps to 2 amps), the previously obtained adjustment base is divided by the matched corresponding constant value, and the result is the final duty cycle adjustment value. If the current is detected to be in the nonlinear stall edge period (i.e., greater than 2 amps), considering that the driver will trigger an overload and overcurrent protection mechanism and actively limit the effective current output, resulting in limited actual effective output torque of the motor, the adjustment base needs to be multiplied by the nonlinear damping coefficient and then divided by the corresponding constant for inverse calculation to prevent excessive current output in the low torque stage, which could lead to burn-out defects. The nonlinear damping coefficient is equal to the difference between the maximum allowable current of the motor and the current current. The difference is used as the dividend, and the difference between the maximum allowable current and the current current is used as the divisor to perform division operations to obtain the quotient. For the lifting actuator, the duty cycle adjustment value calculated here specifically refers to the kinetic energy output parameter of the motor during the gravity resistance phase. By fine-tuning the output parameters in specific high-load segments, the wear and loss of mechanical components inside the seat can be effectively controlled.

[0071] The spatial pose deviation is used as the dividend and the partial derivative is used as the divisor to perform a division operation to generate the adjustment base. The pulse width mapping constant modulus corresponding to the branch discrimination structure matching is established, and then the duty cycle adjustment value is calculated.

[0072] Extract the duty cycle adjustment value and the target drive sequence; obtain the channel address sequence number from the preset hardware configuration file; construct a data packet structure containing control header code, register address code, channel address sequence number, and execution parameter tail string; write the duty cycle adjustment value into the execution parameter tail string; convert the target drive sequence into internal register address code and write it into the register address code; combine the control header code, register address code, channel address sequence number, and execution parameter tail string to generate a motor control command; and deliver the motor control command to the actuator for physical adjustment.

[0073] Specifically, the process for generating motor control commands is as follows: After obtaining the pulse width modulation compensation value with duty cycle dimensions, the value must be translated into a vehicle bus communication data packet structure that the underlying actuator can parse. The Controller Area Network (CAN) bus communication protocol specification is adopted. The calculated duty cycle compensation value and the confirmed target execution field are extracted. Simultaneously, by calling the underlying instruction buffer queue, the timing check code carried by the subsequent queued action about to enter the instruction execution cycle is obtained. Subsequently, a communication data packet structure strictly adhering to the CAN bus extended frame specification is constructed. The data packet structure contains four logical blocks from front to back: frame start flag, arbitration identifier code, data length code, and execution parameter tail string. The arbitration identifier code occupies 29 bits, including a priority identifier (used to match the underlying hardware response authority, occupying 3 bits), a protocol type identifier (which must be constantly filled with 0 for the execution layer, occupying 8 bits), and the physical node segment addressing the specific motor drive node (occupying 18 bits). In addition, following the arbitration identifier code is the remote transmission request flag (occupying 1 binary bit). For example, the value used to issue a one-way write command is 0, while the value used to request a read status value is 1.

[0074] The absolute target duty cycle value, calculated in reverse and converted to hexadecimal encoding (if the variable has a direction indicator, it is forced to use the two's complement format of a 16-bit signed integer), is loaded into the execution parameter tail string area. Next, the built-in vehicle hardware address distribution mapping table is read, the target execution field is translated into physical drive node addressing code, and it is loaded into the physical node segment area. According to the vehicle local area network physical addressing standard, the read and write variable parameters of the actuator (such as motor running speed) are stored in a data area called the "control register". The absolute address space is distinguished by a specific node identifier and ranges from hexadecimal 10 to hexadecimal 90. For example, the node base address corresponding to the rotary actuator is set to hexadecimal 10, the lifting actuator to hexadecimal 20, and the translation actuator to hexadecimal 30. During addressing, the absolute address is directly retrieved by matching dictionary key-value pairs.

[0075] In addition, to adapt to the load attributes of different drive motors, mechanism matching needs to be performed before loading the execution parameter tail string: if the target execution field is a rotary actuator and a translational actuator, the system needs to read the original reference pulse duty cycle currently recorded in the corresponding control register through the status query instruction, and then perform algebraic addition operation between the original reference pulse duty cycle and the pulse width compensation value with positive and negative signs calculated this time to generate the target synthetic duty cycle.

[0076] Finally, the target composite duty cycle is loaded as the execution parameter written to the register; if the target execution field is a lifting actuator that does work against gravity, the maximum allowable torque value of the motor preset inside the device is extracted, and the maximum allowable torque value is used as the numerator and the sum of the original basic holding torque and the duty cycle compensation value is used as the denominator to calculate the required transient current output limit value; combined with the worm gear reduction ratio, the conversion constant from current to duty cycle is derived, and the transient current output limit value is multiplied by the conversion constant to convert it into the direct control parameter of the lifting motor drive chip. When loading the execution parameter tail string, the system uniformly performs underlying data type normalization: for incremental compensation variables in rotary and translational actuators that allow bidirectional fine-tuning, they are forcibly loaded in binary two's complement format of 16-bit signed integers; for absolute setting variables (such as unidirectional lifting thrust) in lifting actuators where the drive chip does not allow negative input values, the calculated floating-point duty cycle value is multiplied by the fixed resolution multiplier specified in the communication protocol and rounded down, and then forcibly converted to 16-bit unsigned integer data (for example, converting 50.5% to unsigned integer 5050). After completing the data type adaptation, it is loaded into the execution parameter tail string area.

[0077] After filling each block, the start frame flag, request flag, arbitration identifier code containing the physical node segment, timing check code record segment, and execution parameter tail string loaded with compensation duty cycle are combined to generate the final binary stream device control command. The combined device control command is delivered via the vehicle's physical bus to the pulse width modulation generator integrated within the driver chip bound to the target execution field. After the control command is parsed, the data in the corresponding control register inside the pulse width modulation generator is overwritten. This ensures that when subsequent queuing action parameters enter the driver chip, the preset motor speed reference is automatically changed according to the duty cycle compensation value, achieving unattended automated feedback control.

[0078] During the synchronous process of issuing motor control commands and executing physical adjustments, the system enters a continuous closed-loop feedback and fault self-checking phase. It acquires real-time pose signals and current characteristics from the actuator sensors to determine if they match the expected trajectory characteristics of the control command. If the expected trajectory is not reached within a preset timeout window, or if the packet loss and retransmission rate of the drive communication link exceeds the limit, the system determines that a hardware or communication fault has been detected. At this point, a standard fault code is immediately generated and reported to the main controller. Simultaneously, subsequent regular control commands are intercepted, and data pushes are triggered to the voice broadcast module and the mobile device APP, achieving dual alarm warnings and thus constructing a complete closed-loop control architecture from perception and execution to health monitoring.

[0079] The timing check code of subsequent queuing actions is obtained, and the instruction pulse width modulation generator changes the preset motor speed reference for subsequent actions according to the duty cycle compensation value, which effectively mitigates the trajectory drift phenomenon in long-cycle cross-action cooperation.

[0080] This invention provides a multimodal interactive control system for accessible seating. By relying on an arbitration transmission network and a strong preemption isolation algorithm, it eliminates unauthorized interference items and clarifies the causal mapping between the multimodal command set and the vehicle's safety state. Combined with command contribution set calculation, it targets and locks onto the effective command set and inversely calculates the pose deviation into a pulse width compensation duty cycle value. This mechanism achieves in-depth communication across modal signals. By outputting device control commands supported by logical calculations, it systematically resolves multi-source concurrent interference, preventing the potential risk of unsteady oscillations in actuators caused by disordered responses, and ensuring the robust operation of the overall control chain.

[0081] Example 2 This embodiment describes the specific implementation process of a barrier-free seat multimodal interactive control system applied to the intelligent adjustment of a multipurpose vehicle cabin.

[0082] For the queuing action of the rear-seat welfare seats exiting the vehicle, a linkage drive sequence with strict spatial trajectory requirements was set. Based on the requirements for smooth mechanical movement, the target rotation parameter baseline was set to 90 degrees. When the operation reached the middle stage, the received real-time angle measurement value was too low and significantly exceeded the preset extremely narrow allowable fluctuation threshold. The state association perception module responded immediately, and through a timing alignment procedure, successfully traced and completely extracted the actual multimodal command set input at the voice node, button node, and application node of the current operating cycle.

[0083] Because the measured angle deviation exceeded the safety limit, the program determined that the current trajectory posed a significant risk of jamming and immediately triggered the permission decoupling calculation process. The arbitration decision tree performed logical extraction operations on the multimodal instruction set to derive the permission attribute values ​​for each channel attribute. Subsequently, the permission attribute values ​​were substituted into the arbitration transmission network, and a strong preemption isolation algorithm was used to cut off the input paths of invalid nodes. The vehicle safety status quantity was set as the highest intervention mask, thereby logically separating the actual intervention factors of the application layer, button layer, and voice layer on the seat posture.

[0084] During the pose deviation calculation phase, the program no longer simply compares surface deviations but multiplies the net effect of each action with its corresponding spatial partial derivative to quantify the specific spatial contribution of each independent execution axis to the overall trajectory deviation. By calculating and sorting the proportion of spatial contribution in the absolute value of the weighted total deviation, the program finds that one item accounts for more than half of the absolute responsibility weight. After extracting the peak weight ratio and performing a reverse search, the program clearly confirms that the core target execution field causing this trajectory hysteresis is the rotary actuator. Since the peak weight ratio extracted earlier determines that the spatial contribution of the rotary action exceeds 85% (the preset single-axis dimensionality reduction takeover threshold), the secondary axis is temporarily latched to maintain its current duty cycle because it is within a small tolerance band. Logically, the system has safely reduced the three-axis coupled system to a single-axis system dominated by the rotary actuator. Based on this dimensionality reduction isolation premise, the program extracts the bound action execution deviation and safely converts it into the additional pulse width adjustment base that must be added to recover the trajectory based on the gear transmission partial derivative of the rotary actuator, ensuring the unique solution attribute of this inverse mapping compensation.

[0085] After locking the rotary actuator, the program extracts the bound pose error input value and, based on the transmission partial derivative, converts it inversely into the pulse duty cycle size that must be added to increase the rotation angle. Next, it retrieves the torque mapping constant modulus of the current motor and directly converts the rotation angle requirement into a specific duty cycle compensation value. The compensated duty cycle is automatically appended to the original preset speed reference of the underlying pulse width modulation generator, generating a new target synthetic duty cycle. Finally, the program converts the duty cycle data into hexadecimal encoding, assembles it into a local area network bus extended frame instruction, and writes it to the driver chip. Subsequent queuing operations are executed according to the compensation parameters, successfully increasing the motor output torque and restoring the seat posture to the target reference of 90 degrees.

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multimodal interactive control system for accessible seating, characterized in that, include: The multimodal perception module acquires voice commands, APP commands, and physical button signals, and merges them into a concurrent command set; it also acquires vehicle status signals. The concurrent instruction set and vehicle status signals are time-aligned to generate an interactive array; The priority arbitration module parses the control source attributes contained in the interaction array to obtain the permission set; The concurrent instruction set is converted into nodes, and the permission set and vehicle status signals are converted into constraints to build an arbitration decision tree; the interaction array is substituted into the arbitration decision tree, and the priority filtering algorithm is substituted into the arbitration decision tree to perform filtering operations and generate an effective instruction set; The trajectory localization module converts the valid instruction set into preset travel parameters to generate a target position set; it acquires the current pose parameters, performs a difference operation on the current pose parameters and the target position set to generate a pose deviation set; it extracts the non-zero values ​​in the pose deviation set to generate a non-zero deviation amount, and converts the non-zero deviation amount into a target driving sequence. The inverse calculation closed-loop module converts the numerical values ​​bound to the target drive sequence within the non-zero deviation into error input values, substitutes the error input values ​​into the proportional-integral-derivative algorithm to inversely calculate the duty cycle adjustment value, generates a motor control command containing the duty cycle adjustment value, and transmits the motor control command to the actuator for physical adjustment.

2. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that the specific generation process of the interactive array includes: It receives voice commands, APP commands, and physical button signals, and combines them into a concurrent command set; it also receives vehicle status signals. Extract the channel identifier and receive timestamp carried by the concurrent instruction set; Extract the status message identifier and message timestamp carried by the vehicle status signal; Calculate the absolute difference between the received timestamp and the message timestamp. If the absolute difference falls within a preset synchronization time window, then merge the corresponding concurrent instruction set and vehicle status signal into an associated instruction group. The received timestamp and the message timestamp are merged to construct a trigger timing chain; the associated instruction group is reorganized and aligned along the trigger timing chain to generate an interactive array.

3. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that, The specific generation process of the permission set includes: extracting control source attributes contained in the interaction array; obtaining preset priority sorting constants to generate a permission mapping vector library; obtaining vehicle speed and gear linkage restriction items to generate a safety intervention vector library; substituting the control source attributes into the permission mapping vector library to perform a matching query operation to generate a basic priority identifier; performing a joint judgment operation on the basic priority identifier and the safety intervention vector library to generate a safety state quantity; extracting activation feature values ​​from the safety state quantity; performing a logical AND operation on the basic priority identifier and the activation feature values ​​to generate a highest intervention mask word; and merging the basic priority identifier, the safety state quantity, and the highest intervention mask word to generate a permission set.

4. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that the specific generation process of the effective instruction set includes: The concurrent instruction set is converted into nodes, and the permission set and vehicle status signals are converted into constraints to build an arbitration decision tree; the interaction array is substituted into the arbitration decision tree to extract node status attributes; the priority filtering algorithm is substituted into the node status attributes to perform classification operations and generate retained feature items and interference feature items. The transmission path of the interference feature is cut off; the retained feature is converted into an active value; the safety constraint distribution function of the vehicle state signal under the active value condition is calculated; the parking expectation parameter in the safety constraint distribution function is extracted; the parking expectation parameter and the active value are fused to generate the net functional effect parameter; the net functional effect parameter and the retained feature are summarized to generate the effective instruction set.

5. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that the specific generation process of the pose deviation set includes: Convert the valid instruction set into preset travel parameters; Perform decomposition operations on the preset travel parameters to generate target rotation parameters, target elevation parameters, and target translation parameters, and merge them into a target position set; obtain the current pose parameters; The target position set is vector subtracted from the current pose parameters to calculate the motion deviation in the rotation, elevation and translation dimensions, and the results are combined to form a pose deviation set.

6. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that the specific generation process of the target driving sequence includes: Obtain the seat positioning reference and allowable fluctuation threshold; extract the rotational motion deviation, lifting motion deviation, and translational motion deviation from the pose deviation set; The absolute difference calculation is performed on the rotational movement deviation, lifting movement deviation, and translational movement deviation relative to the seat positioning reference to generate an absolute difference set; the parameters in the absolute difference set are compared with the allowable fluctuation threshold, invalid jitter within the threshold range is filtered out, and the overshoot parameter is extracted to form a non-zero deviation. Subsequently, based on the proportion of non-zero deviation in the overall trajectory offset, an action execution weight set containing the responsibility weight of each drive axis is generated; the peak value in the action execution weight set is extracted and converted into the primary weight ratio; the primary weight ratio is substituted into the non-zero deviation to perform a reverse retrieval operation to match the target execution field; and the target execution field is converted into the target drive sequence.

7. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that, The specific process for generating the duty cycle adjustment value includes: extracting the numerical values ​​bound to the target drive sequence within the non-zero deviation range and converting them into error input values; obtaining a preset gain constant; performing proportional-integral-differential operations on the error input value and the preset gain constant to generate a reference adjustment base; extracting the channel attribute features contained in the target drive sequence; obtaining the current monitoring safety current limit value associated with the channel attribute features; and substituting the reference adjustment base into the current monitoring safety current limit value to perform a limiting and truncation operation to generate the duty cycle adjustment value.

8. The barrier-free seat multimodal interactive control system according to claim 1, characterized in that the specific generation process of the motor control command includes: Extract duty cycle adjustment value and target driving sequence; Retrieve the channel addressing sequence number from the preset hardware configuration file; Construct a data packet structure containing control header code, register address code, channel address sequence number, and execution parameter tail string; write the duty cycle adjustment value into the execution parameter tail string; convert the target drive sequence into internal register address code and write it into the register address code; combine the control header code, register address code, channel address sequence number, and execution parameter tail string to generate a motor control command; and deliver the motor control command to the actuator for physical adjustment.

Images