A blind guiding robot walking control method and system based on a time sequence convolution network

Abstract

This invention relates to the field of guide robot technology, and more particularly to a walking control method and system for guide robots based on temporal convolutional networks. The method includes the following steps: during the process of the guide robot guiding the object to walk, continuously collecting the walking speed and turning state of the guide robot during the walking process, and constructing temporal data of walking speed and turning state; using a preset control period as the time interval, using a temporal convolutional network to extract the change relationship of the walking state temporal data in the time dimension, and determining the walking rhythm characteristics; this invention extracts walking rhythm characteristics and cooperative walking stability characteristics through a temporal convolutional network to achieve continuous perception of walking state changes, forward prediction of guidance behavior patterns, and smooth control of walking speed and turning adjustments during guided walking.

Description

Technical Field

[0001] This invention relates to the field of guide robot technology, and in particular to a walking control method and system for guide robots based on temporal convolutional networks. Background Technology

[0002] Guide robots are increasingly being applied to assist visually impaired individuals in their daily travel, with their core function being to provide safe, stable, and continuous guidance in complex environments. Existing guide robots largely rely on preset path planning, rule-driven control, or closed-loop control based on real-time sensor feedback, generating walking control commands based on current speed deviations or turning errors. However, these methods typically use the instantaneous state within a single control cycle as the basis for decision-making, lacking a holistic model of the walking state over time and failing to reflect the continuous evolution of walking rhythms during human-robot collaborative walking. In practical applications, when the gait of the guided individual changes, or during turning or deceleration, the guide robot is prone to abrupt changes in control commands, inconsistent speed adjustments, or frequent turning corrections, thus affecting walking stability and the comfort of the guided individual. Summary of the Invention

[0003] Therefore, it is necessary to provide a walking control method and system for guide robots based on temporal convolutional networks to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a walking control method for a guide robot based on temporal convolutional networks includes the following steps:

[0005] Step S1: During the process of the guide robot guiding the object to walk, continuously collect the walking speed and turning status of the guide robot during the walking process, and construct the walking status time series data of walking speed and turning status;

[0006] Step S2: Using a preset control cycle as the time interval, use a temporal convolutional network to extract the temporal relationship of the walking state time series data in the time dimension and determine the walking rhythm features;

[0007] Step S3: Analyze the continuity of walking state changes within adjacent control cycles based on walking rhythm characteristics, generate stability characteristics of the guide robot's collaborative walking stability, and predict the guided walking behavior pattern within the next control cycle based on the stability characteristics.

[0008] Step S4: Generate walking control commands based on the guided walking behavior pattern, and constrain the walking speed and turning state of the guide robot before outputting the walking control commands to achieve smooth control of the guide robot during the guided walking process.

[0009] The present invention also includes a walking control system for a guide robot based on a temporal convolutional network, used to execute the walking control method for a guide robot based on a temporal convolutional network as described above. The walking control system for the guide robot based on a temporal convolutional network includes:

[0010] The guide robot data acquisition module is used to continuously acquire the walking speed and turning status of the guide robot during the walking process of guiding the object, and to construct the walking status time series data of walking speed and turning status.

[0011] The walking rhythm feature determination module is used to extract the temporal relationship of walking state temporal data in the time dimension and determine the walking rhythm features by using a temporal convolutional network with a preset control cycle as the time interval.

[0012] The guided walking behavior pattern prediction module is used to analyze the continuity of walking state changes in adjacent control cycles based on walking rhythm characteristics, generate stability characteristics of the collaborative walking stability of the guide robot, and predict the guided walking behavior pattern in the next control cycle based on the stability characteristics.

[0013] The walking control command generation module is used to generate walking control commands based on the guided walking behavior pattern, and to constrain the walking speed and turning state of the guide robot before outputting the walking control commands, so as to achieve smooth control of the guide robot during the guided walking process.

[0014] This invention continuously collects walking speed and turning status during the guiding process of a blind person using a guide robot, constructing temporal data of the walking state. It then extracts walking rhythm features based on a temporal convolutional network, further analyzing the continuity of walking state changes within adjacent control cycles to form stability features characterizing the stability of coordinated walking. Based on this, it predicts the guided walking behavior pattern in the next control cycle and constrains the amplitude of speed and turning control changes before outputting walking control commands. This ensures that walking control commands change smoothly between consecutive control cycles, reducing the impact of sudden speed adjustments or turning corrections on the guided person and improving the stability, coordination, and safety of the guide robot during guided walking. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of a walking control method for a guide robot based on a temporal convolutional network.

[0016] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0017] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0021] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] To achieve the above objectives, please refer to Figures 1 to 3 A walking control method for a guide robot based on temporal convolutional networks includes the following steps:

[0023] All specific values ​​involved in this embodiment are exemplary parameters used to clearly illustrate the technical operation process and are not the only limitation of the present invention.

[0024] Step S1: During the process of the guide robot guiding the object to walk, continuously collect the walking speed and turning status of the guide robot during the walking process, and construct the walking status time series data of walking speed and turning status;

[0025] Step S2: Using a preset control cycle as the time interval, use a temporal convolutional network to extract the temporal relationship of the walking state time series data in the time dimension and determine the walking rhythm features;

[0026] Step S3: Analyze the continuity of walking state changes within adjacent control cycles based on walking rhythm characteristics, generate stability characteristics of the guide robot's collaborative walking stability, and predict the guided walking behavior pattern in the next control cycle based on the stability characteristics.

[0027] Step S4: Generate walking control commands based on the guided walking behavior pattern, and constrain the walking speed and turning state of the guide robot before outputting the walking control commands to achieve smooth control of the guide robot during the guided walking process.

[0028] In this embodiment, the guide robot includes a walking speed acquisition module, a turning state acquisition module, an interactive force acquisition module, a control processing unit, and a motion execution unit. The walking speed acquisition module is a wheel speed encoder installed at the axle ends of the left and right drive wheels; the turning state acquisition module is an inertial measurement unit (IMU) used to output yaw rate and turning angle change; the interactive force acquisition module is located at the traction structure between the guide robot and the guided object, used to acquire force changes in the traction direction. During the guided object's walking process, the linear velocities of the left and right wheels are continuously acquired at a sampling frequency of 50Hz, and the walking speed value is calculated based on the speed difference between the left and right wheels; simultaneously, the yaw rate output by the IMU is acquired and integrated to obtain the turning angle change. The walking speed data and turning angle change data are synchronously processed according to timestamps to form a time-series walking state data arranged in chronological order.

[0029] The control processing unit performs windowed processing on the walking status timing data with a control cycle of 200ms. Within each control cycle, the walking speed sequence and steering angle change sequence are statistically processed. The walking speed change is defined as the difference between the maximum and minimum walking speeds within the control cycle, and the steering angle change is defined as the cumulative change value of the steering angle change sequence within the control cycle. The walking speed change and steering angle change are combined into a two-dimensional parameter vector in a preset order, which serves as the initial parameters of the walking rhythm representing the walking speed adjustment amplitude and steering adjustment amplitude within the current control cycle.

[0030] The force change signal of the traction structure within a control cycle is synchronously acquired, and the difference in traction force between adjacent sampling points is calculated to form a force change sequence. The force change sequence is statistically analyzed within the control cycle to obtain the average change amplitude of traction force within that cycle, which serves as an interaction state parameter characterizing the interaction intensity change between the guided object and the guide robot. The interaction state parameter is aligned with the initial parameters of the walking rhythm according to timestamps. The interaction state parameter is introduced as an adjustment factor to proportionally scale and correct the changes in walking speed and turning angle in the initial parameters of the walking rhythm, resulting in corrected initial parameters of the walking rhythm that reflect the influence of human-machine interaction.

[0031] The initial parameters of the corrected walking rhythm obtained within multiple consecutive control cycles are input sequentially into a temporal convolutional network. This network consists of three one-dimensional convolutional layers with kernel sizes of 3, 5, and 7, used to extract rhythmic variation features at different time scales. During feature extraction, the network performs hierarchical modeling of the trends in walking speed and turning, thereby obtaining consistent features of walking rhythm changes within consecutive control cycles, which are then used as walking rhythm features.

[0032] Based on walking rhythm characteristics, the rhythm changes within adjacent control cycles are compared and analyzed. The amplitude and direction of change of rhythm characteristics in the time dimension are calculated to assess the maintenance of the rhythm within continuous control cycles. Based on the assessment results, stability characteristics reflecting the rhythm stability of the guide robot during cooperative walking are generated. Combined with the walking state in the current control cycle, the walking speed adjustment trend and turning adjustment trend in the next control cycle are determined, resulting in the corresponding guided walking behavior pattern.

[0033] Based on the guided walking behavior pattern, walking speed adjustment commands and steering adjustment commands corresponding to the current control cycle are generated. Before outputting the walking control commands, the walking speed adjustment commands and steering adjustment commands generated in adjacent control cycles are compared. The change amplitude of the walking speed adjustment command is defined as the absolute value of the difference between the target walking speed commands in two adjacent control cycles, and the change amplitude of the steering adjustment command is defined as the absolute value of the difference between the target steering angle commands in two adjacent control cycles. The change amplitudes of the walking speed adjustment command and the steering adjustment command are compared with preset change amplitude constraint thresholds. The walking speed change amplitude constraint threshold is used to limit the maximum adjustment of the walking speed command in a unit control cycle, and the steering change amplitude constraint threshold is used to limit the maximum adjustment of the steering angle command in a unit control cycle. When the change amplitude of the walking speed adjustment command or the steering adjustment command exceeds the corresponding change amplitude constraint threshold, the portion exceeding the threshold is limited to ensure that the processed walking control command meets the preset change amplitude constraint conditions, thereby generating a walking control command that meets the continuity requirement. The walking control command after change amplitude constraint is output to the motion execution unit to drive the guide robot to complete continuous and smooth adjustments of walking speed and steering state in adjacent control cycles.

[0034] Preferably, step S2 includes:

[0035] Step S21: According to the preset control cycle, the walking status time series data is windowed, and the walking speed change and turning change in the walking status time series data are extracted in each control cycle.

[0036] Step S22: Within the control cycle, determine the initial parameters of the walking rhythm based on the changes in walking speed and steering.

[0037] Step S23: Synchronously collect the interaction state parameters of the guide robot, and align the interaction state parameters with the initial parameters of the walking rhythm in time, and correct the initial parameters of the walking rhythm;

[0038] Step S24: Input the corrected initial parameters of the walking rhythm into the temporal convolutional network, extract the consistency features of the walking rhythm changes within the continuous control cycle, and form the walking rhythm features.

[0039] In this embodiment, the control cycle of the guide robot is set to 200ms. The control processing unit performs sliding window processing on the constructed walking state timing data at time intervals of this control cycle, with each window corresponding to a complete control cycle. Within each control cycle, the walking speed sequence and turning state sequence are differentially calculated to obtain the changes in walking speed and turning direction within that control cycle.

[0040] Within a control cycle, the changes in walking speed and steering are combined according to a unified time reference. Based on the relationship between their magnitude and direction of change within the same control cycle, rhythm parameters reflecting the characteristics of the current cycle's walking rhythm are constructed and used as initial parameters for the walking rhythm. For example, when the change in walking speed increases slightly while the change in steering remains stable, the corresponding initial parameters for the walking rhythm are characterized as a smooth acceleration rhythm; when the change in walking speed remains essentially constant while the change in steering changes continuously, the corresponding initial parameters for the walking rhythm are characterized as a steering adjustment rhythm.

[0041] While generating the initial parameters of the walking rhythm, a force sensing unit installed at the traction structure of the guide robot synchronously collects information on the force changes experienced by the robot during guided walking. The force change information is continuously sampled, and the force difference between adjacent sampling points is calculated to obtain interaction state parameters reflecting the human-machine interaction state. These interaction state parameters are aligned with the initial parameters of the walking rhythm within the corresponding control cycle according to timestamps. The degree of disturbance of external force changes to the rhythm parameters is analyzed through the mapping relationship, and the initial parameters of the walking rhythm are corrected accordingly to obtain the corrected initial parameters of the walking rhythm.

[0042] Subsequently, the initial parameters of the corrected walking rhythm obtained within multiple consecutive control cycles are input into the temporal convolutional network in the order of the control cycles. The temporal convolutional network uses the consecutive control cycles as the time dimension, performs convolution operations on the change relationship between the initial parameters of the walking rhythm in adjacent cycles, extracts the consistency feature of the rhythm change in the time dimension, and determines this consistency feature as the walking rhythm feature.

[0043] In another embodiment, the control cycle of the guide robot is set to 300ms. Within each control cycle, the control processing unit segments the walking state timing data and statistically analyzes the range of changes in walking speed and turning state within that control cycle, thereby obtaining the corresponding changes in walking speed and turning state. Within the control cycle, the changes in walking speed and turning state are jointly modeled. Based on their relative proportions within the same control cycle, rhythm parameters characterizing the walking rhythm state are generated, and these rhythm parameters are used as initial parameters for the walking rhythm. When the changes in walking speed and turning state maintain a relatively consistent proportion within consecutive cycles, the initial parameters for the walking rhythm exhibit a stable distribution; when a significant shift occurs in their proportions, the initial parameters for the walking rhythm are adjusted accordingly.

[0044] The interactive state acquisition module continuously collects force change information between the guide robot and the guided object, and organizes this information into a force change sequence arranged in chronological order. Based on this force change sequence, the force change characteristics within the control cycle are calculated and used as interactive state parameters. These interactive state parameters are time-aligned with the initial parameters of the walking rhythm within the corresponding control cycle. The initial parameters of the walking rhythm are then corrected according to the degree of influence of the force change characteristics on the rhythm parameters, resulting in corrected initial parameters of the walking rhythm that reflect the influence of human-robot interaction. The corrected initial parameters of the walking rhythm are input into a temporal convolutional network in the order of the control cycle. Through convolutional modeling of the rhythm parameter changes within continuous control cycles, the stability and continuity characteristics of the rhythm changes in the time dimension are extracted, forming walking rhythm features that characterize the walking state of the guide robot.

[0045] Preferably, within the control cycle, the initial parameters of the walking rhythm are determined based on the changes in walking speed and steering, including:

[0046] Within the control cycle, the changes in walking speed and steering are synchronized over time to form speed change sequences and steering change sequences.

[0047] By statistically analyzing the change amplitude of the velocity change sequence, the velocity change parameters within the control period are obtained.

[0048] By statistically analyzing the magnitude of changes in the steering change sequence, the steering change parameters within the control cycle are obtained.

[0049] Based on the time correspondence between speed change parameters and steering change parameters, a coordinated change relationship between speed change parameters and steering change parameters within the control cycle is constructed, and the coordinated change relationship within the control cycle is used as the initial parameter of the walking rhythm.

[0050] In this embodiment, the control cycle of the guide robot is set to 200ms. Within a single control cycle, the control processing unit performs time synchronization processing on the changes in walking speed and turning speed acquired within that control cycle. Walking speed changes at the same time reference are arranged in sampling order to form a speed change sequence, and turning speed changes at corresponding times are arranged in the same time order to form a turning change sequence. After obtaining the speed change sequence, the amplitude of change at each sampling point in the speed change sequence is statistically processed. The maximum, minimum, and average amplitude of walking speed change within the control cycle are calculated, and these three amplitudes together constitute a set of speed change parameters characterizing the degree of walking speed adjustment within the control cycle.

[0051] Accordingly, the variation amplitudes of each sampling point in the steering change sequence are statistically processed to calculate the maximum, minimum, and average variation amplitudes of the steering angle within the control cycle. These maximum, minimum, and average variation amplitudes together constitute a steering change parameter set characterizing the degree of steering adjustment within the control cycle. Based on this, and considering the time correspondence between the speed change parameter set and the steering change parameter set within the same control cycle, a joint analysis is performed to construct a coordinated change relationship reflecting the synchronous changes in walking speed and steering adjustment. This coordinated change relationship is then determined as the initial parameters of the walking rhythm.

[0052] In another embodiment, the control cycle of the guide robot is set to 300ms. Within each control cycle, the control processing unit performs unified timestamp calibration on the changes in walking speed and turning, eliminates abnormal sampling points that do not meet the time synchronization conditions, and constructs the calibrated changes in walking speed into a speed change sequence in chronological order, and constructs the corresponding changes in turning into a turning change sequence.

[0053] For the speed change sequence, the amplitude of change between adjacent sampling points within the sequence is calculated, and the amplitudes are accumulated to obtain speed change parameters characterizing the intensity of speed adjustment within the control cycle. For the steering change sequence, the cumulative result of the steering change amplitude is calculated using the same statistical method to form steering change parameters characterizing the intensity of steering adjustment. Subsequently, the speed change parameters and steering change parameters are combined according to the time correspondence within the control cycle, and the degree of matching between their changes within the same cycle is analyzed to construct a coordinated change relationship reflecting the synergistic characteristics of speed and steering changes. When the speed change parameters and steering change parameters show a synchronous increasing or decreasing trend at multiple consecutive sampling points, the coordinated change relationship is characterized as a consistent rhythm; when there is a significant deviation in their change trends, the coordinated change relationship is adjusted accordingly. The constructed coordinated change relationship is used as the initial parameter for the walking rhythm within the control cycle.

[0054] Preferably, based on the time correspondence between speed change parameters and steering change parameters, a coordinated change relationship between speed change parameters and steering change parameters within the control cycle is constructed, and this coordinated change relationship is used as the initial parameter for the walking rhythm, including:

[0055] Time series alignment of speed variation parameters and steering variation parameters;

[0056] The speed change parameters after time series alignment are combined with the steering change parameters to form a combined sequence of the coordinated changes of speed change parameters and steering change parameters;

[0057] The overall magnitude and trend of the velocity and steering parameters in the statistical combination sequence are analyzed, and statistical processing results are generated based on the overall magnitude and trend.

[0058] Based on the statistical processing results, a cooperative relationship between speed change parameters and steering change parameters is constructed, and this cooperative relationship is used as the initial parameter for walking rhythm.

[0059] In one embodiment, based on the time correspondence between speed change parameters and steering change parameters within the control cycle, sequence alignment processing under a unified time reference is performed on the speed change parameters and steering change parameters, so that they form a one-to-one corresponding data pair at the same sampling time. After completing the time alignment, the speed change parameters and steering change parameters at the corresponding time are combined to construct a combined sequence reflecting the synchronous characteristics of speed adjustment and steering adjustment during walking. Further, the overall change amplitude and trend of the combined sequence within the control cycle are statistically analyzed, for example, statistically analyzing the synchronous enhancement segment, synchronous weakening segment, or relatively stable segment of speed change and steering change. Based on the above statistical analysis results, a coordinated change relationship is constructed to characterize the coordinated adjustment characteristics of speed change parameters and steering change parameters, and the coordinated change relationship is used as the initial parameter of walking rhythm for rhythm adjustment in subsequent walking control processes.

[0060] In another embodiment, based on time alignment of the speed change parameters and steering change parameters, the amplitude and trend of their changes within the control cycle are combined to generate a coordinated change relationship that reflects the coordinated adjustment characteristics of speed and steering, and the coordinated change relationship is used as the initial parameter of the walking rhythm.

[0061] Preferably, the interaction state parameters of the guide robot are collected synchronously, and the force change information and the initial parameters of the walking rhythm are time-aligned. The correction of the initial parameters of the walking rhythm includes:

[0062] Continuously collect information on the force changes of the guide robot during its walking process to form time-series data of force changes;

[0063] Extract the force difference between adjacent sampling points in the time series data of force changes to determine the interaction state parameters of the guide robot;

[0064] The interactive state parameters of the guide robot are mapped to the initial parameters of the walking rhythm, and the disturbance effect of external force changes on the walking rhythm is analyzed as the result of the disturbance effect analysis.

[0065] Based on the analysis results of the disturbance effect, the initial parameters of the walking rhythm are proportionally corrected to obtain the initial parameters of the walking rhythm.

[0066] In one embodiment, the interaction state parameters of the guide robot during its walking process are collected synchronously, and the force change information corresponding to the interaction state parameters is aligned with the initial parameters of the walking rhythm according to a unified time reference. During the walking process, the force change information generated between the guide robot and the guided object or the external environment is continuously collected to form a force change time series data arranged in chronological order. Based on the force change time series data, the force difference between adjacent sampling times is calculated to characterize the interaction state of the guide robot during its walking process, such as changes in traction force, resistance force, or directional guidance force. The force difference is determined as the interaction state parameter of the guide robot.

[0067] After obtaining the interaction state parameters, the interaction state parameters are analyzed in correspondence with the initial parameters of the walking rhythm within the same control cycle. The initial parameters of the walking rhythm are composed of the changes in walking speed and turning angle within the control cycle. By analyzing the changes in the interaction state parameters in adjacent control cycles, the continuity of the external force changes in the time dimension is determined. The change amplitude of the interaction state parameters is compared with a preset disturbance judgment threshold. When the change amplitude of the interaction state parameters exceeds the disturbance judgment threshold, it is determined that the walking rhythm in the current control cycle is affected by external force disturbance.

[0068] When it is determined that the walking rhythm is affected by external force disturbance, the changes in walking speed and turning angle in the initial parameters of the walking rhythm are proportionally scaled and corrected according to the ratio of the interaction state parameters in the current control cycle to the disturbance judgment threshold, so as to reduce the change amplitude of the initial parameters of the walking rhythm under the action of external force disturbance, and the corrected initial parameters of the walking rhythm are obtained. The corrected initial parameters of the walking rhythm are then used for walking control in subsequent control cycles of the guide robot.

[0069] In another embodiment, based on the force change information collected during the walking process of the guide robot, the corresponding interaction state parameters are extracted, and the interaction state parameters are time-aligned and mapped with the initial parameters of the walking rhythm. The initial parameters of the walking rhythm are proportionally corrected according to the degree of influence of external force disturbance on the walking rhythm.

[0070] Preferably, continuously collecting force change information of the guide robot during its walking process to form force change time series data includes:

[0071] Real-time detection of the traction force on the guide robot in the guiding direction and the lateral force when turning;

[0072] The traction force and lateral force are sampled synchronously at equal intervals at a preset frequency to generate traction force samples and lateral force samples.

[0073] The traction force sample and lateral force sample obtained at each sampling time are combined to construct a guidance state vector;

[0074] The guiding state vectors are arranged into a force vector sequence according to the sampling order, forming a time-series data of force change information.

[0075] In one embodiment, during the walking process of the guide robot, the traction force in the guidance direction and the lateral force generated when performing turning actions are monitored synchronously. The traction force is used to reflect the restraint of the guided object on the walking speed or forward direction, and the lateral force is used to reflect the change of lateral force during turning guidance or path correction. According to a preset sampling frequency, the traction force and lateral force are sampled synchronously at equal time intervals to obtain discrete force samples arranged in chronological order. At each sampling moment, the corresponding traction force sample and lateral force sample are combined to construct a guidance state vector representing the current walking guidance state. According to the sampling time sequence, the guidance state vectors corresponding to each sampling moment are arranged sequentially to form force change time series data reflecting the force evolution characteristics during walking, which is used for subsequent interactive state analysis.

[0076] In another embodiment, during the walking process of the guide robot, the traction force and lateral force are sampled synchronously and combined in chronological order to form a guide state vector sequence, so as to form corresponding force change time series data.

[0077] Preferably, the modified initial parameters of the walking rhythm are input into a temporal convolutional network to extract the consistency features of the walking rhythm changes within continuous control cycles, forming walking rhythm features including:

[0078] The corrected initial parameters of the walking rhythm are input into the temporal convolutional network in the order of the control cycle. The changes of the initial parameters of the walking rhythm in adjacent cycles are evaluated in a coordinated manner with the continuous control cycle as the time dimension.

[0079] During the network feature extraction process, a temporal convolutional network is used to extract the local fluctuations and overall trends of rhythm changes in the initial parameters of walking rhythm in a hierarchical manner, distinguish the rhythm change patterns caused by speed adjustment and steering adjustment, and identify rhythm change patterns that meet the requirements of guide robot walking control.

[0080] Based on the identified rhythmic change patterns, the maintenance of walking rhythm changes in the time dimension within a continuous control cycle is determined, forming a consistency feature of walking rhythm changes, and the consistency feature is used as the walking rhythm feature.

[0081] Following the temporal sequence of the control cycle, the corrected initial parameters of the walking rhythm are constructed into a parameter sequence with the continuous control cycle as the time dimension, and input into a temporal convolutional network. This network models the correlation between changes in the initial parameters of the walking rhythm within adjacent control cycles to evaluate the coordinated changes of the walking rhythm within the continuous control cycle. During the feature extraction process of the temporal convolutional network, the change information in the initial parameters of the walking rhythm is processed by multi-level convolution to extract local rhythmic fluctuation features reflecting short-term speed or steering adjustments, as well as long-term change features reflecting the overall rhythmic evolution trend during guided walking. This distinguishes the impact patterns of different control behaviors on the walking rhythm. Based on the combined results of local rhythmic fluctuation features and overall rhythmic trend features, a rhythmic change pattern with controlled amplitude and continuous direction of change within the continuous control cycle is identified. The degree of maintenance of the walking rhythm in the time dimension is analyzed accordingly, forming a feature result characterizing the consistency of walking rhythm changes, which serves as the walking rhythm feature.

[0082] In another embodiment, the modified initial parameters of the walking rhythm are input into a temporal convolutional network in the order of the control cycle. The local fluctuations and overall trends of the rhythm changes are jointly extracted, and the consistency features of the walking rhythm changes are formed based on the correlation of changes within the continuous control cycle.

[0083] Preferably, step S3 includes:

[0084] Step S31: Compare and analyze the changes in walking rhythm characteristics within adjacent control cycles, assess the degree of maintenance of walking rhythm between consecutive control cycles, and generate continuity analysis results;

[0085] Step S32: Based on the continuity analysis results, determine the rhythmic stability level of the guide robot during the cooperative walking process, and generate the stability characteristics of the cooperative walking stable state;

[0086] Step S33: After generating stability features, combine the walking status in the current control cycle to determine the speed adjustment trend and turning adjustment trend of the guide robot in the next control cycle, and predict the corresponding guide walking behavior pattern.

[0087] In one embodiment, within adjacent control cycles, the walking rhythm characteristics corresponding to consecutive control cycles are compared and analyzed cycle by cycle. By comparing the amplitude, direction, and continuity of change of the walking rhythm characteristics in the time dimension, the degree of maintenance of the walking rhythm between consecutive control cycles is evaluated, generating a continuity analysis result reflecting the continuity of the walking rhythm evolution. For example, in the process of a guide robot guiding a guided object to walk straight at a stable speed, if the walking rhythm characteristics show small amplitude changes and consistent directions of change within adjacent control cycles, it is determined that the degree of maintenance of the walking rhythm in that stage is high. However, when encountering turning or obstacle avoidance scenarios, if the walking rhythm characteristics show staged changes within a controllable range within adjacent control cycles, these changes are still included in the continuity analysis result. This study aims to reflect the natural rhythm adjustments during human-robot collaborative walking. Based on the continuity analysis results, it comprehensively analyzes the maintenance of walking rhythm characteristics across multiple adjacent control cycles to determine the rhythmic stability level of the guide robot during collaborative walking. Using this rhythmic stability level as a basis, it constructs stability features characterizing the stable state of collaborative walking, which are used to depict the guide robot's control capability over the walking rhythm during the current guidance phase. After generating the stability features, it combines the walking state information of the guide robot within the current control cycle to perform correlation analysis on the stability features, determine the influence trend of walking rhythm changes on subsequent control, and then determine the speed adjustment trend and turning adjustment trend of the guide robot in the next control cycle, predicting the corresponding guided walking behavior pattern.

[0088] In another embodiment, a continuous comparative analysis of the walking rhythm characteristics within adjacent control cycles is performed to generate a continuous analysis result reflecting the degree of maintenance of the walking rhythm; based on the continuous analysis result, a stability feature of the coordinated walking stable state is constructed, and the speed adjustment trend and steering adjustment trend within the next control cycle are determined in combination with the current walking state to predict and guide the walking behavior pattern.

[0089] Preferably, based on the continuity analysis results, the rhythmic stability level of the guide robot during cooperative walking is determined, and the stability characteristics of the cooperative walking stable state are generated, including:

[0090] Based on the continuity analysis results, the degree of continuity of the changes in walking rhythm characteristics in the time dimension is compared and analyzed within adjacent control cycles to generate comparative analysis results;

[0091] Based on the comparative analysis results, the maintenance characteristics of walking rhythm within a continuous control cycle are evaluated, and walking rhythm change parameters are formed based on the distribution characteristics and amplitude characteristics of rhythm changes in the time dimension.

[0092] A unified characterization of the stability of walking rhythm variation parameters is established to construct stability characteristics of the walking rhythm stability under the collaborative walking state of guide robots.

[0093] In one embodiment, based on the continuity analysis results, a cycle-by-cycle comparative analysis is performed on the continuity of changes in walking rhythm features over time within adjacent control cycles. By comparing the consistency of rhythm change amplitude, direction of change, and interval of change within adjacent control cycles, comparative analysis results are generated to characterize the consistency of rhythm evolution. For example, when a guide robot guides an object to walk along a straight path, if the amplitude of changes in walking rhythm features is small and the trend of change is consistent across multiple adjacent control cycles, the comparative analysis results reflect continuous and smooth rhythm changes. However, in scenarios involving turning or navigating obstacles, if rhythm changes are concentrated in a few control cycles and quickly recover to stability after the changes, the comparative analysis results show that the rhythm changes are continuous and smooth. The analysis results reflect that the rhythm adjustment has phased rather than random fluctuations. Based on the comparative analysis results, the maintenance characteristics of the walking rhythm within the continuous control cycle are evaluated. Combining the distribution characteristics of rhythm changes in the time dimension and the magnitude of the corresponding changes, walking rhythm change parameters are extracted to quantify the rhythm fluctuation characteristics, reflecting the frequency and intensity of rhythm adjustment. The stability of the walking rhythm change parameters within the continuous control cycle is uniformly characterized. By comprehensively analyzing the concentration, dispersion, and persistence of the change parameters, stability characteristics of the walking rhythm stability under the collaborative walking state of the guide robot are constructed to characterize the overall collaborative stability state during the current guided walking process.

[0094] Of particular importance, step S4 includes:

[0095] Based on the guided walking behavior pattern, walking control commands corresponding to the current control cycle are generated. The walking control commands include walking speed adjustment commands and steering adjustment commands.

[0096] Before outputting the walking control command, the control change range of the walking speed adjustment command and the steering adjustment command are calculated respectively, and the control change amount exceeding the constraint range is limited according to the preset change range constraint conditions.

[0097] The walking control command, constrained by the change amplitude, is output to the motion execution unit of the guide robot to achieve continuous control of walking speed and steering adjustment during guided walking.

[0098] In one embodiment, based on the predicted guided walking behavior pattern and combined with the walking state parameters of the guide robot in the current control cycle, walking control commands corresponding to the current control cycle are generated. These walking control commands correspond to walking speed adjustment commands and steering adjustment commands, respectively, to describe the speed and direction adjustment needs of the guide robot in the next control cycle. After generating the walking control commands but before outputting them, the control change magnitude of the walking speed adjustment commands and steering adjustment commands is calculated. By comparing the difference between the corresponding control commands in the current control cycle and the previous control cycle, the speed control change and steering control change are obtained, which characterize the dramatic change in control adjustment between adjacent control cycles. The intensity of the change is assessed. Based on pre-set variation range constraints, the changes in speed control and steering control are constrained and judged separately. When any control change exceeds the corresponding constraint range, the excess portion is restricted to keep the constrained control change within the allowable range. For example, when the stability characteristics of the walking rhythm are detected during coordinated walking, indicating that the rhythm changes are continuous and stable, the variation range constraints allow for small-amplitude continuous adjustments to ensure the smoothness of guided walking. However, in scenarios where the rhythm changes are concentrated or there are significant short-term fluctuations, the variation range constraints further tighten the adjustment range of speed or steering to avoid disrupting the gait of the guided object due to sudden changes in control commands.

[0099] The walking speed adjustment command and steering adjustment command, after being constrained by the change range, are output to the motion execution unit of the guide robot, so that the guide robot can complete the speed and steering adjustment in a restricted and gradual manner within a continuous control cycle, thereby realizing continuous control in the guided walking process.

[0100] Most importantly, before outputting the walking control command, the control change amplitude of the walking speed adjustment command and the steering adjustment command are calculated respectively, and the control change amount exceeding the constraint range is limited according to the preset change amplitude constraint conditions, including:

[0101] Before outputting the walking control command, the changes in the walking speed adjustment command and steering adjustment command in adjacent control cycles are analyzed to obtain the corresponding control change amplitude.

[0102] Based on preset variation range constraints, the range of control variation is determined, and the amount of control variation exceeding the constraint range is limited, so as to generate a walking control command that satisfies the variation range constraints.

[0103] In one embodiment, before outputting the walking control command, the walking speed adjustment command and steering adjustment command generated in the current control cycle are first aligned with the corresponding control commands executed in the previous control cycle, using the control cycle as the time reference. After the command alignment is completed, the difference between the walking speed adjustment commands and the difference between the steering adjustment commands in adjacent control cycles are calculated respectively. The difference is used as the control change amplitude of the corresponding control command to characterize the adjustment intensity of the control command in a continuous control cycle. Subsequently, the calculated control change amplitude is range-determined according to the preset change amplitude constraint conditions. The change amplitude constraint conditions are preset according to the collaborative walking control requirements of the guide robot and are used to limit the speed adjustment and steering within a unit control cycle. The allowable range of adjustment is defined. When any control change exceeds the corresponding constraint range, the excess portion is restricted to ensure that the processed control change falls within the preset constraint range, thereby generating a walking control command that satisfies the change range constraint conditions. For example, when the predicted guided walking behavior pattern corresponds to a straight-line or gentle-curve guided scenario, the change range constraint conditions limit the speed adjustment command and steering adjustment command to only allow small, continuous changes to avoid abrupt changes in the gait of the guided object. However, when the predicted guided walking behavior pattern corresponds to a scenario where steering adjustment requirements increase, the change range constraint conditions allow for relatively larger adjustments to the steering adjustment command within a defined range while keeping speed changes limited, thus balancing guidance responsiveness and cooperative walking stability.

[0104] The present invention also includes a walking control system for a guide robot based on a temporal convolutional network, used to execute the walking control method for a guide robot based on a temporal convolutional network as described above. The walking control system for the guide robot based on a temporal convolutional network includes:

[0105] The guide robot data acquisition module is used to continuously acquire the walking speed and turning status of the guide robot during the walking process of guiding the object, and to construct the walking status time series data of walking speed and turning status.

[0106] The walking rhythm feature determination module is used to extract the temporal relationship of walking state temporal data in the time dimension and determine the walking rhythm features by using a temporal convolutional network with a preset control cycle as the time interval.

[0107] The guided walking behavior pattern prediction module is used to analyze the continuity of walking state changes in adjacent control cycles based on walking rhythm characteristics, generate stability characteristics of the collaborative walking stability of the guide robot, and predict the guided walking behavior pattern in the next control cycle based on the stability characteristics.

[0108] The walking control command generation module is used to generate walking control commands based on the guided walking behavior pattern, and to constrain the walking speed and turning state of the guide robot before outputting the walking control commands, so as to achieve smooth control of the guide robot during the guided walking process.

[0109] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A walking control method for a guide robot based on temporal convolutional networks, characterized in that, Includes the following steps: Step S1: During the process of the guide robot guiding the object to walk, continuously collect the walking speed and turning status of the guide robot during the walking process, and construct the walking status time series data of walking speed and turning status; Step S2: Using a preset control cycle as the time interval, use a temporal convolutional network to extract the change relationship of the walking state time series data in the time dimension and determine the walking rhythm features; including: according to the preset control cycle, perform windowing processing on the walking state time series data, and extract the walking speed change and turning change in the walking state time series data in each control cycle. Within the control cycle, the initial parameters of the walking rhythm are determined based on the changes in walking speed and steering. The interaction status parameters of the guide robot are collected synchronously, and the interaction status parameters and the initial parameters of the walking rhythm are time-aligned to correct the initial parameters of the walking rhythm. The corrected initial parameters of the walking rhythm are input into a temporal convolutional network to extract consistent features of walking rhythm changes within continuous control cycles, forming walking rhythm features, which include: The corrected initial parameters of the walking rhythm are input into the temporal convolutional network in the order of the control cycle. The changes of the initial parameters of the walking rhythm in adjacent cycles are evaluated in a coordinated manner with the continuous control cycle as the time dimension. During the network feature extraction process, a temporal convolutional network is used to extract the local fluctuations and overall trends of rhythm changes in the initial parameters of walking rhythm in a hierarchical manner, distinguish the rhythm change patterns caused by speed adjustment and steering adjustment, and identify rhythm change patterns that meet the requirements of guide robot walking control. Based on the identified rhythmic change patterns, the maintenance of walking rhythm changes in the time dimension within a continuous control cycle is determined, forming a consistency feature of walking rhythm changes, and the consistency feature is used as the walking rhythm feature. Step S3: Analyze the continuity of walking state changes within adjacent control cycles based on walking rhythm characteristics, generate stability characteristics of the guide robot's collaborative walking stability, and predict the guided walking behavior pattern in the next control cycle based on the stability characteristics. Step S4: Generate walking control commands based on the guided walking behavior pattern, and constrain the walking speed and turning state of the guide robot before outputting the walking control commands to achieve smooth control of the guide robot during the guided walking process.

2. The walking control method for a guide robot based on a temporal convolutional network according to claim 1, characterized in that, Within the control cycle, the initial parameters of the walking rhythm are determined based on changes in walking speed and steering. Within the control cycle, the changes in walking speed and steering are synchronized over time to form speed change sequences and steering change sequences. By statistically analyzing the change amplitude of the velocity change sequence, the velocity change parameters within the control period are obtained. By statistically analyzing the magnitude of changes in the steering change sequence, the steering change parameters within the control cycle are obtained. Based on the time correspondence between speed change parameters and steering change parameters, a coordinated change relationship between speed change parameters and steering change parameters within the control cycle is constructed, and the coordinated change relationship within the control cycle is used as the initial parameter of the walking rhythm.

3. The walking control method for a guide robot based on a temporal convolutional network according to claim 2, characterized in that, Based on the time correspondence between speed change parameters and steering change parameters, a coordinated change relationship between speed change parameters and steering change parameters within the control cycle is constructed, and this coordinated change relationship is used as the initial parameters of the walking rhythm, including: Time series alignment of speed variation parameters and steering variation parameters; The speed change parameters after time series alignment are combined with the steering change parameters to form a combined sequence of the coordinated changes of speed change parameters and steering change parameters; The overall magnitude and trend of the velocity and steering parameters in the statistical combination sequence are analyzed, and statistical processing results are generated based on the overall magnitude and trend. Based on the statistical processing results, a cooperative relationship between speed change parameters and steering change parameters is constructed, and this cooperative relationship is used as the initial parameter for walking rhythm.

4. The walking control method for a guide robot based on a temporal convolutional network according to claim 1, characterized in that, The interaction state parameters of the guide robot are collected synchronously, and the force change information and the initial parameters of the walking rhythm are time-aligned. The initial parameters of the walking rhythm are corrected as follows: Continuously collect information on the force changes of the guide robot during its walking process to form time-series data of force changes; Extract the force difference between adjacent sampling points in the time series data of force change to determine the interaction state parameters of the guide robot; The interactive state parameters of the guide robot are mapped to the initial parameters of the walking rhythm. The impact of external force changes on the walking rhythm is analyzed, and the results of the disturbance effect analysis are used as the analysis results. Based on the analysis results of the disturbance effect, the initial parameters of the walking rhythm are proportionally corrected to obtain the initial parameters of the walking rhythm.

5. The walking control method for a guide robot based on a temporal convolutional network according to claim 4, characterized in that, Continuously collecting force change information of the guide robot during its walking process, forming time-series data of force changes, including: Real-time detection of the traction force on the guide robot in the guiding direction and the lateral force when turning; The traction force and lateral force are sampled synchronously at equal intervals at a preset frequency to generate traction force samples and lateral force samples. The traction force sample and lateral force sample obtained at each sampling time are combined to construct a guidance state vector; The guiding state vectors are arranged into a force vector sequence according to the sampling order, forming a time-series data of force change information.

6. The walking control method for a guide robot based on a temporal convolutional network according to claim 1, characterized in that, Step S3 includes: Within adjacent control cycles, the changes in walking rhythm characteristics are compared and analyzed to assess the degree of maintenance of walking rhythm between consecutive control cycles, and to generate continuity analysis results. Based on the continuity analysis results, the rhythmic stability level of the guide robot during the cooperative walking process is determined, and the stability characteristics of the cooperative walking stable state are generated. After generating stability features, the robot determines the speed and turning adjustment trends of its guided walking in the next control cycle by combining the walking status in the current control cycle, and predicts the corresponding guided walking behavior pattern.

7. The walking control method for a guide robot based on a temporal convolutional network according to claim 6, characterized in that, Based on the continuity analysis results, the rhythmic stability level of the guide robot during cooperative walking was determined, and the stability characteristics of the cooperative walking steady state were generated, including: Based on the continuity analysis results, the degree of continuity of the changes in walking rhythm characteristics in the time dimension is compared and analyzed within adjacent control cycles to generate comparative analysis results; Based on the comparative analysis results, the maintenance characteristics of walking rhythm within a continuous control cycle are evaluated, and walking rhythm change parameters are formed based on the distribution characteristics and amplitude characteristics of rhythm changes in the time dimension. A unified characterization of the stability of walking rhythm variation parameters is established to construct stability characteristics of the walking rhythm stability under the collaborative walking state of guide robots.

8. A walking control system for a guide robot based on a temporal convolutional network, characterized in that, A method for controlling the walking of a guide robot based on a temporal convolutional network as described in any one of claims 1 to 7, wherein the guide robot walking control system based on a temporal convolutional network comprises: The guide robot data acquisition module is used to continuously acquire the walking speed and turning status of the guide robot during the walking process of guiding the object, and to construct the walking status time series data of walking speed and turning status. The walking rhythm feature determination module is used to extract the temporal relationship of walking state temporal data in the time dimension and determine the walking rhythm features by using a temporal convolutional network with a preset control cycle as the time interval. The guided walking behavior pattern prediction module is used to analyze the continuity of walking state changes in adjacent control cycles based on walking rhythm characteristics, generate stability characteristics of the collaborative walking stability of the guide robot, and predict the guided walking behavior pattern in the next control cycle based on the stability characteristics. The walking control command generation module is used to generate walking control commands based on the guided walking behavior pattern, and to constrain the walking speed and turning state of the guide robot before outputting the walking control commands, so as to achieve smooth control of the guide robot during the guided walking process.

Images