A blind guiding robot gait stability control method fusing IMU and foot end force perception

Abstract

The application relates to the technical field of blind guiding robot gait stability control, and particularly discloses a blind guiding robot gait stability control method fusing IMU and foot end force sensing. The application collects three-dimensional point cloud data through a head depth camera, analyzes terrain features and generates terrain safety confidence, and then adjusts the physical rigidity of a dielectric elastomer ankle joint. In the single-foot supporting stage, foot bottom pressure and trunk inertia data are synchronously collected, real-time foot bottom interaction snapshots are generated, and when instability risk is detected, the ankle joint rigidity is immediately locked to resist abnormal rotation. Then, the abnormal direction of the foot bottom pressure is identified, the mass center compensation adjustment amount is calculated to stabilize the body posture. Finally, the matching degree of the gait strategy and the road condition is evaluated according to the amplitude and frequency of the mass center compensation, the step length and the leg lifting height are dynamically adjusted, and the optimized gait parameters are generated. The application aims to improve the walking stability and safety of the robot in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of gait stabilization control technology for guide robots, and more specifically, to a gait stabilization control method for guide robots that integrates IMU and foot force sensing. Background Technology

[0002] Currently, legged guide robots or assistive robots face core challenges in achieving stable walking in unstructured environments. The complexity and unpredictability of the environment, such as sudden obstacles, slippery areas, or uneven terrain, can all cause robot instability. These dynamic disturbances require the robot control system to have millisecond-level rapid response capabilities and adaptive body coordination strategies to maintain user safety.

[0003] Patent CN116774689A discloses a control method and system for a guide robot based on point-to-point connection, including: wirelessly connecting a user's mobile terminal with the guide robot to ensure that the guide robot provides dedicated guide services to the user throughout the user's movement, and returning corresponding guide service messages to the mobile terminal in real time; it also utilizes the guide robot to perform visual recognition of the area near the user, and provides real-time and accurate reminders of the presence of obstacles in the area, thereby improving the accuracy, reliability, and real-time performance of the guide service.

[0004] The shortcomings of existing technologies lie in their limitation: the solutions remain at a superficial level of macroscopic environmental perception and human-machine collaboration, failing to fully consider the core challenge of the guide robot's dynamic gait stability in complex terrain. Specifically, treating the robot as a stable sensor platform, its core logic is to passively identify large obstacles in the external environment and alert the user, and to maintain a queue relationship with the user through simple distance measurement. Its control philosophy is essentially information service rather than physical stability. It lacks both a forward-looking strategy for predictive analysis of terrain through depth vision and proactive pre-setting of joint physical stiffness, and a refined perception of the microscopic physical interaction between the foot and the ground. Therefore, when the robot encounters sudden and unpredictable road disturbances, it cannot provide any effective physical self-rescue mechanism, nor does it possess a unified closed-loop framework that combines instantaneous stable recovery actions with long-term adaptive adjustment of gait strategies. This results in the system frequently being in an emergency intervention state, affecting the smoothness of walking. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a gait stabilization control method for guide robots that integrates IMU and foot force sensing is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: This invention provides a gait stability control method for a guide robot that integrates IMU and foot force perception, including: S1, activating the robot's head depth camera to collect three-dimensional point cloud data of the direction of travel, analyzing terrain features and evaluating perception reliability based on ambient lighting conditions, and generating a terrain safety confidence level that characterizes the reliability of visual prediction and the complexity of the terrain.

[0007] S2. Call the stiffness mapping rule, select the voltage value corresponding to the terrain safety confidence level, and apply it to the dielectric elastomer ankle joint of the robot's foot, so that the physical stiffness of the dielectric elastomer ankle joint is the ankle joint basic stiffness setting value that matches the terrain prediction.

[0008] S3. During the robot's single-leg support phase, with the ankle joint's basic stiffness set as the initial state, data from the foot flexible pressure sensor array and the torso inertial measurement unit are collected simultaneously to generate a real-time foot interaction data package describing the current foot-ground physical interaction and the robot's dynamic trend.

[0009] S4. Compare the parameters in the real-time foot interaction data packet with the preset normal interaction model. When the parameters are detected to exceed the normal fluctuation range, immediately generate an instantaneous stiffness locking command to deal with the risk of sudden instability.

[0010] S5. The instantaneous stiffness locking command is converted into a preset locking voltage signal and applied to the dielectric elastomer ankle joint to increase the physical torsional stiffness of the dielectric elastomer ankle joint to the maximum value to lock the dielectric elastomer ankle joint, resist abnormal rotation, and form a physically locked stable ankle joint state.

[0011] S6. Identify the abnormal direction of plantar pressure that causes a stable ankle joint state that leads to physical locking, reverse the calculation to counteract the shift of the body's center of mass caused by this abnormal plantar pressure, and calculate the corresponding joint angle adjustment value to generate a center of mass compensation adjustment amount for coordinating and stabilizing the body posture.

[0012] S7. Drive the hip and knee joints to perform center of gravity compensation adjustments to stabilize the body's center of gravity, and evaluate the matching degree between the gait strategy and road conditions based on the magnitude and frequency of center of gravity compensation. Adjust the default stride length and leg lift height for the next step, generate optimized gait parameters, and input them into the next gait control cycle.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention achieves rapid switching of joint physical properties by introducing a dielectric elastomer ankle joint and its active stiffness adjustment mechanism. When the system detects abnormal foot interaction, it can increase the ankle joint stiffness to the maximum value within milliseconds to physically lock the joint. This technical feature executed by a specific high-voltage drive circuit enables the robot to resist abnormal joint rotation caused by ground slippage or impact with almost no delay, providing a critical time window for compensation adjustment of the upper body, thereby helping to improve the response speed to sudden disturbances.

[0014] (2) This invention constructs a real-time plantar interactive data package that integrates depth visual prediction, plantar pressure distribution, and trunk inertial data, and compares thresholds using a preset normal interaction model, thereby achieving multimodal and refined perception of instability risk. This technical feature enables the system to distinguish different types of disturbances and accurately locate the direction and intensity of the disturbances. Based on this accurate perception, the inverse dynamics calculation can generate a centroid compensation adjustment amount that highly matches the disturbance pattern, driving the hip and knee joints to perform coordinated fine-tuning, thereby enhancing the countermeasure effect against disturbances in specific directions and improving the accuracy of posture recovery.

[0015] (3) This invention evaluates the matching degree between gait strategy and road conditions by analyzing the amplitude and frequency of compensation events after performing centroid compensation, and dynamically adjusts the default stride length and leg lift height for the next step accordingly. This closed-loop feedback technology enables the robot to automatically generate more conservative gait parameters after experiencing unstable events, reducing the probability of instability recurring in subsequent gait cycles. It achieves adaptive optimization from single-event response to long-term walking strategy, which is beneficial to improving the overall adaptability and stability of walking in continuous unknown or complex terrain. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 This invention provides a gait stabilization control method for a guide robot that integrates IMU and foot force perception, including: S1, activating the robot's head depth camera to collect three-dimensional point cloud data of the direction of travel, analyzing terrain features and evaluating perception reliability based on ambient lighting conditions, and generating a terrain safety confidence level that characterizes the reliability of visual prediction and the complexity of the terrain.

[0020] In a specific embodiment of the present invention, the generation of terrain safety confidence, which characterizes the reliability of visual prediction and the complexity of terrain, includes: drawing grids representing landing areas in three-dimensional point cloud data, calculating the variance of height values ​​of points within each unit grid, and using the average of the variances of height values ​​of all unit grids as the flatness quantification value.

[0021] The reference plane is fitted based on 3D point cloud data, and the number of points with a height higher than the reference plane within a unit grid area is used as the obstacle density value.

[0022] Plane fitting is performed on the 3D point cloud data, the angle between the normal vector of the fitted plane and the direction of gravity is calculated to obtain the slope angle of the current area, and the rate of change of the slope angle in the direction of travel is tracked to obtain the slope change trend value.

[0023] It should be noted that the implementation process of step S1 is as follows. First, the depth camera installed on the robot's head is activated. This camera actively emits invisible light towards the area in front of the robot's direction of travel and receives the reflected light, thereby collecting the three-dimensional spatial coordinates of each point on the surface of the object in front, forming three-dimensional point cloud data. This data is continuously input into the processing unit in the form of a time series. Next, the processing unit analyzes the incoming three-dimensional point cloud data in real time. For flatness analysis, the algorithm draws a two-dimensional grid representing the main landing area in front in the point cloud data, calculates the variance of the height values ​​of all points in each cell of the grid, and this variance represents the degree of undulation of the ground within the cell. The average of the variances of all cell grids is the overall flatness metric of the area. For obstacle density analysis, the algorithm first fits a reference plane based on the point cloud, and then counts the number of points per unit area whose height is significantly higher than the reference plane. This number is the obstacle density value. For slope change trend analysis, the algorithm performs plane fitting on the point cloud, calculates the normal vector of the fitted plane, and calculates the slope angle of the current area based on the angle between the normal vector and the direction of gravity. At the same time, it tracks the rate of change of the slope angle in the robot's forward direction to assess whether the terrain is continuously uphill, downhill, or has a sudden change.

[0024] It should be further explained that, in order to achieve efficient fitting of the reference plane and accurate calculation of the subsequent slope angle, the algorithm specifically employs a robust method called random sample consistency for plane fitting. The specific implementation steps of this method can be simply understood as follows: First, three points are randomly selected from the collected 3D point cloud data. These three points are sufficient to mathematically define a unique candidate plane. Next, the algorithm iterates through all other points in the point cloud, automatically calculating their distances to this candidate plane and counting the total number of points whose distances are less than a preset small threshold (e.g., 1 cm). These points are called inliers. This process is iterated hundreds of times, each time randomly selecting three points to generate a new candidate plane and counting its inliers. Finally, the candidate plane with the most inliers is identified as the best-fitting plane representing the main trend of the current terrain. It can be used as a reference plane to distinguish obstacles and also for slope analysis. After determining the best-fitting plane, its plane equation is... The vector formed by the coefficients in This is the normal vector of the plane. The slope angle of the current region is the intersection of this normal vector and the gravity direction vector representing the absolute vertical direction (usually in the robot coordinate system). The angle between the two vectors. This angle can be calculated by taking the dot product of the two vectors, dividing by the product of their respective magnitudes, and finally taking the inverse cosine value. This is used to accurately determine and quantify the degree of inclination of the ground.

[0025] Compare the flatness quantization value, the obstacle density value, and the slope change trend value with the preset grade thresholds respectively, and combine the perception reliability evaluated based on the environmental light conditions to output three discrete grades of high, medium, and low as the terrain safety confidence level.

[0026] It should be noted that the preset grade thresholds of the flatness quantization value, the obstacle density value, and the slope change trend value are determined by collecting more than 1000 sets of sample data in standard experimental sites (such as flat epoxy floors, simulated pebble roads, slopes, etc.), and after statistically analyzing the distribution of each index value, selecting the 25th percentile and the 75th percentile as the lower and upper limits of the "good" grade. For example, the threshold of the flatness quantization value (height variance, unit: m²) is set as: poor > 1.0×10 -4 , good is between 1.0×10 -5 and 1.0×10 -4 , excellent < 1.0×10 -5 . If all indicators are "excellent" and the perception reliability is high, the front terrain is classified as the "high" safety level; if any indicator is "good" or the perception reliability is average, it is classified as the "medium" safety level; if any indicator is "poor" or the perception reliability is low, it is classified as the "low" safety level. The perception reliability based on the light conditions is specifically incorporated into the decision-making through the following formula: Final terrain safety confidence = Visual terrain confidence × Light weight, where the visual terrain confidence is initially determined by the comparison results of the above indicators, high = 1.0, medium = 0.7, low = 0.4. The light weight is mapped according to the illuminance value L (unit: lux): when L > 500, the weight is 1.0; when 100 < L ≤ 500, the weight is 0.8; when L ≤ 100, the weight is 0.5. If the weighted final value ≥ 0.8, "high" is output; if ≥ 0.5 and < 0.8, "medium" is output; if < 0.5, "low" is output.

[0027] In a specific embodiment of the present invention, the specific analysis method of the perception reliability evaluated based on the environmental light conditions is as follows: Use the visible light sensor自带 by the depth camera to obtain the illuminance value of the current environment in real time, and compare the illuminance value of the current environment with the illuminance value intervals corresponding to each perception reliability stored in the database. If the illuminance value of the current environment is within the illuminance value interval corresponding to a certain perception reliability, then use this perception reliability as the perception reliability of the current environment, where the perception reliability includes high, average, and low.

[0028] A depth camera is a sensor that acquires the three-dimensional coordinates of points on the surface of an object within its field of view. It calculates distance by emitting light pulses and measuring the reflection time. Three-dimensional point cloud data refers to a series of point cloud frames acquired sequentially over time. Each frame consists of a large number of data points containing X, Y, and Z coordinates, forming a three-dimensional spatial description of the environment ahead. The acquisition frequency is set to 10Hz to balance ensuring real-time terrain information with avoiding data processing overload. Flatness refers to the smoothness and evenness of the terrain surface ahead, quantified by calculating the statistical variance of point cloud height values. A smaller variance indicates greater flatness, and the threshold is set based on statistical analysis of 200 sets of measured data from different road surfaces. Obstacle density refers to the number of protruding objects per unit area that may pose a tripping hazard. It is calculated by counting the number of discrete points higher than the fitted plane, and the setting is based on robot passability tests in simulated obstacle scenarios in the laboratory. Slope change trend refers to the terrain tilt angle and its variation in the direction of travel, obtained through plane fitting and normal vector analysis, used to predict the robot's center of gravity shift trend. Lighting conditions refer to the brightness of the light in the robot's environment, quantified by illuminance values ​​measured by sensors, and measured in lux.

[0029] S2. Call the stiffness mapping rule, select the voltage value corresponding to the terrain safety confidence level, and apply it to the dielectric elastomer ankle joint of the robot's foot, so that the physical stiffness of the dielectric elastomer ankle joint is the ankle joint basic stiffness setting value that matches the terrain prediction.

[0030] In a specific embodiment of the present invention, the physical stiffness of the dielectric elastomer ankle joint is set to a basic ankle joint stiffness setting value that matches the terrain prediction. This includes: the stiffness mapping rule defines the output voltage range corresponding to high, medium and low terrain safety confidence levels through a lookup table.

[0031] Based on the currently received level of terrain safety confidence, a preset voltage value is selected from the corresponding output voltage range.

[0032] A preset voltage value is applied to the electrodes of the dielectric elastomer ankle joint through a high-voltage drive circuit. By utilizing the characteristic of the dielectric elastomer material to change the macroscopic elastic modulus under the action of an electric field, a basic stiffness setting value for the ankle joint that matches the terrain prediction is generated.

[0033] It should be noted that, firstly, the control system receives terrain safety confidence levels generated in step S1, characterized by three levels: "high," "medium," and "low." Next, the system invokes a stiffness mapping rule pre-stored in memory. This stiffness mapping rule is a lookup table that explicitly defines the output voltage range corresponding to each terrain safety confidence level. For example, the mapping rule is: "high" safety level corresponds to an output voltage range of 0 to 1 kV; "medium" safety level corresponds to an output voltage range of 1 to 2 kV; and "low" safety level corresponds to an output voltage range of 2 to 3 kV. Then, based on the currently received specific terrain safety confidence level, the system selects a preset voltage value from the corresponding output voltage range. The selection strategy can be to select the median value within the corresponding output voltage range; for example, when the terrain safety confidence level is "medium," 1.5 kV is selected. This selected preset voltage value is directly defined as a set of control commands. Finally, the system applies this voltage control command to the dielectric elastomer ankle joint of the robot's foot through a high-voltage drive circuit. When a dielectric elastomer material is subjected to an external electric field, the polarization state of its molecular chains changes, leading to an increase in the material's macroscopic elastic modulus, i.e., its physical stiffness. The higher the applied voltage, the stronger the generated electric field, and the torsional stiffness of the ankle joint increases linearly or non-linearly. Through this process, the physical stiffness of the dielectric elastomer ankle joint is adjusted from a default state to a specific value that matches the macroscopic terrain prediction, thus generating the basic ankle joint stiffness setting.

[0034] It should also be noted that the stiffness mapping rule defines the output voltage range corresponding to the high, medium, and low terrain safety confidence levels through a lookup table. This lookup table is established through the following experimental calibration process: (1) Material property calibration: On a material testing machine, the dielectric elastomer film used for the ankle joint is tested, and the fitting relationship between its voltage (V, unit kV) and equivalent Young's modulus (E, unit MPa) is obtained as follows: (2) Joint stiffness conversion: Based on the crank-slider structure design parameters of the ankle joint (crank radius) =0.05m), convert the material modulus to joint torsional stiffness K(V) (unit N·m / rad), the relationship is: ,in It is a geometric transformation factor that integrates the mechanical structural dimensions of the ankle joint. Its physical meaning is to convert the Young's modulus of the dielectric elastomer material into the torsional stiffness of the entire joint (unit: N·m / rad). In a specific embodiment of this invention, through experimental calibration of the designed ankle joint structure, the value of the geometric transformation factor was measured to be 120 m³ / rad. Therefore... (3) Walking stability test: Walking tests were conducted on a bipedal robot prototype weighing 25kg on three calibrated terrains (flat epoxy flooring, a test platform covered with pebbles with an average particle size of 2cm, and a slope with an inclination angle of 15° covered with 3cm thick sponge). For each terrain, the voltage V was adjusted in the range of 0-3kV with a step size of 0.2kV, and 10 walking tests were conducted at each voltage point. The standard deviation of ankle joint angular displacement was recorded for each test. The sum of the maximum lateral and vertical sway amplitudes of the robot's torso center of mass. Define stability metrics The weight =0.6, =0.4. (4) Determine the optimal range: Analyze the curve of S(V) changing with V under each terrain. The principle for determining the "optimal voltage range" is: the S(V) value in the range is not lower than 95% of its maximum value, and the lower voltage value in the range is taken to save energy. The final calibration results are shown in Table 1 below - find the example:

[0035] Table 1 - Lookup Table

[0036]

[0037] S3. During the robot's single-leg support phase, with the ankle joint's basic stiffness set as the initial state, data from the foot flexible pressure sensor array and the torso inertial measurement unit are collected simultaneously to generate a real-time foot interaction data package describing the current foot-ground physical interaction and the robot's dynamic trend.

[0038] In a specific embodiment of the present invention, a real-time foot interaction data packet describing the current foot-ground physical interaction and the dynamic trend of the robot is generated, including: when it is determined that the robot is in the single-foot support stage, the pressure matrix data output by the foot flexible pressure sensor array and the attitude angular velocity data output by the torso inertial measurement unit are collected simultaneously.

[0039] The coordinates of the pressure center on the sole of the foot and its movement speed over time are calculated based on the pressure matrix data, and local pressure peaks in the pressure matrix that are greater than their neighborhood values ​​are identified.

[0040] Establish data alignment timestamps, and package and combine the pressure center coordinates, movement speed, local pressure peak and attitude angular velocity data at the same time to generate real-time plantar interaction data packets.

[0041] It should be noted that the implementation process of step S3 is as follows. The system receives the ankle joint basic stiffness setting value generated in step S2, which serves as the current initial physical state of the robot's foot joint. When the robot enters the single-foot support phase, i.e., when only one foot is in contact with the ground supporting the body weight, the system initiates the data acquisition process. First, starting from the physical state corresponding to the ankle joint basic stiffness setting value, the system continuously acquires pressure distribution data from the flexible pressure sensor array installed on the robot's foot plate. This data is updated in real time in matrix form, where each element corresponds to the pressure value at a specific location on the sole of the foot. Simultaneously, the system synchronously records the attitude angular velocity data from the inertial measurement unit installed on the robot's torso, including angular velocity components around the front-rear axis, left-right axis, and vertical axis. Next, the system processes the pressure distribution matrix acquired in each frame. The coordinates of the pressure center point in the foot coordinate system are calculated using the following formula: , ,in, and These indicate that the center of pressure is within the plane of the foot. shaft and The coordinates along the axis, in meters. This indicates the first in the flexible plantar pressure sensor array. The pressure value measured by each sensing unit is in Newtons. and Indicates the first The system displays the position coordinates of each sensing unit in a preset foot plane coordinate system, in meters. The summation symbol Σ indicates the summation calculation for all sensing units in active working condition. Simultaneously, the system calculates the velocity of the pressure center point over time by comparing the pressure center coordinates of two or more consecutive frames, in m / s. Furthermore, the system identifies local pressure peaks in the current pressure distribution map, i.e., it finds all points in the pressure matrix whose pressure values ​​are greater than those of other units in their neighborhood and records their positions and pressure values. Finally, the system establishes a data alignment timestamp, packaging and combining the pressure center coordinates, pressure center velocity, identified local pressure peak information, and torso posture angular velocity data collected at the same sampling time into a structured data packet. This data packet generates a real-time foot-ground interaction data packet describing the current foot-ground physical interaction details and the overall dynamic trend of the robot.

[0042] The single-leg support phase is a specific time window in the gait cycle where only one foot fully bears the body weight while the other foot is swinging. Its start and end are determined by a foot contact switch or a total pressure threshold. A flexible foot pressure sensor array is a sensor composed of multiple discrete flexible pressure sensing units arranged in a specific topology. It measures the pressure at various points within the foot-ground contact area, outputting a continuously changing two-dimensional pressure distribution map over time. Pressure distribution data refers to the two-dimensional matrix data output by the aforementioned sensor array, characterizing the pressure at various points on the foot at a given moment. Attitude angular velocity refers to the instantaneous angular velocity of the robot's torso rotating around its own coordinate system axes, measured by an inertial measurement unit, typically including roll, pitch, and yaw components. The pressure center point is the point of application of the resultant ground reaction force within the foot contact area, its coordinates calculated using the aforementioned formula. Coordinates refer to the position of the pressure center point in a two-dimensional planar reference system with a fixed point on the foot as the origin, used to quantify the precise location of the pressure center. Movement speed refers to the change in the position of the pressure center point per unit time, calculated by differentiating the coordinates of adjacent moments and dividing by the sampling time interval. Local pressure peak refers to the point in the current plantar pressure distribution map where the pressure value is significantly higher than the pressure value of the surrounding adjacent areas, usually corresponding to the high-stress area generated when stepping on a small hard protrusion or contact edge.

[0043] S4. Compare the parameters in the real-time foot interaction data packet with the preset normal interaction model. When the parameters are detected to exceed the normal fluctuation range, immediately generate an instantaneous stiffness locking command to deal with the risk of sudden instability.

[0044] In a specific embodiment of the present invention, generating an instantaneous stiffness locking command for responding to sudden instability risks includes: pre-defining a normal interaction model, which sets threshold values ​​for the normal fluctuation range of pressure center movement speed, local pressure peak value, and posture angular velocity during stable walking for different ankle joint basic stiffness settings.

[0045] It should be noted that the normal interaction model and its threshold were established through the following offline learning process: The robot was controlled to perform a total of 1080 (3x3x120) stable gait cycles on a flat experimental field at three speeds of 0.3 m / s, 0.5 m / s, and 0.7 m / s, with small, medium, and large strides respectively. Simultaneously, plantar pressure and IMU (Inertial Measurement Unit) data were recorded when the ankle joint's basic stiffness was set to 1.7 kV (corresponding to a "medium" confidence level). Statistical analysis was performed on all sample data of the lateral movement velocity parameter of the pressure center. The 2.5 percentile (P2.5) and 97.5 percentile (P97.5) were calculated, yielding the normal fluctuation range threshold for this parameter at this specific stiffness as [P2.5, P97.5] = [-0.118, 0.123] m / s. For simplification, the system used ±0.12 m / s as the threshold boundary for this parameter. Similarly, threshold values ​​for parameters such as local pressure peaks and attitude roll angular velocity under corresponding conditions can be obtained. All threshold values ​​are categorized according to stiffness preset values ​​and stored in the normal interaction model in the form of a lookup table.

[0046] When the movement speed of the pressure center toward the edge of the foot exceeds its corresponding normal fluctuation range threshold, or the local pressure peak exceeds the expected normal fluctuation range threshold of the current gait phase, or the attitude angular velocity exceeds its corresponding normal fluctuation range threshold, it is determined that there is a risk of sudden instability, and an instantaneous stiffness locking command is immediately generated.

[0047] It should be noted that the system extracts each parameter from the current real-time foot-to-ground interaction data packet and compares it in real-time with the threshold values ​​corresponding to the current ankle joint basic stiffness settings in the normal interaction model. The comparison logic determines whether the parameter values ​​in the snapshot fall within the corresponding normal fluctuation range defined by the model. Finally, the system executes anomaly detection logic: when any parameter in the snapshot, such as the pressure center movement velocity, local pressure peak, or posture angular velocity, is detected to exceed the corresponding fluctuation range threshold defined by the normal interaction model, the system immediately determines that the current foot-ground interaction is abnormal and there is a risk of sudden instability. In response, the system generates a control signal with the highest priority, namely, an instantaneous stiffness lock command.

[0048] The normal interaction model is a database storing the allowable normal value ranges of various key interaction parameters under different preset joint stiffness conditions. Different base stiffnesses refer to a series of discrete joint physical stiffness values ​​pre-set in step S2 based on different terrain predictions. The normal fluctuation range is the upper and lower limits of each interaction parameter's value under a specific base stiffness, statistically determined under stable walking conditions; its setting is based on statistical analysis of large-scale measured data. The threshold is the boundary value of the normal fluctuation range, used as a threshold to judge whether a parameter is abnormal in real-time comparisons. Any parameter exceeding the normal fluctuation range means that at least one of the three monitored parameters—pressure center movement speed, local pressure peak, and attitude angular velocity—is no longer within its preset normal fluctuation range in real time. Rapid sliding of the pressure center towards the foot edge is a specific manifestation of abnormal pressure center movement speed, meaning the movement speed of the pressure center point towards the foot edge exceeds the speed threshold set for that direction. Unexpected impact pressure peaks are a specific manifestation of abnormal local pressure peaks, where the pressure value at a certain point on the sole of the foot rises sharply within a short period of time, exceeding the pressure peak threshold expected for the current gait phase and stiffness. Sudden instability risk refers to the possibility that the robot may lose balance due to events such as sudden slippage on the ground, stepping on an unforeseen obstacle, or experiencing a lateral impact.

[0049] S5. The instantaneous stiffness locking command is converted into a preset locking voltage signal and applied to the dielectric elastomer ankle joint to increase the physical torsional stiffness of the dielectric elastomer ankle joint to the maximum value to lock the dielectric elastomer ankle joint, resist abnormal rotation, and form a physically locked stable ankle joint state.

[0050] In a specific embodiment of the present invention, forming a physically locked stable ankle joint state includes: receiving an instantaneous stiffness locking command and parsing a preset locking voltage signal in the instantaneous stiffness locking command.

[0051] Within milliseconds, the high-voltage drive circuit increases the voltage applied to the dielectric elastomer ankle joint to a preset locking voltage signal. By utilizing the material polarization and stretching induced by the electric field, the physical torsional stiffness of the ankle joint increases nonlinearly and rapidly to the maximum allowable value.

[0052] By utilizing the resisting torque generated at the maximum value, a physically locked and stable ankle joint state is formed.

[0053] It should be noted that the implementation process of step S5 is as follows. The system receives the instantaneous stiffness locking command generated in step S4. This instantaneous stiffness locking command, as an emergency trigger signal, is input to the high-voltage drive circuit specifically controlling the dielectric elastomer ankle joint. The high-voltage drive circuit has a built-in command decoding unit, which translates the instantaneous stiffness locking command into a specific, preset locking voltage signal output command. For example, this command is defined as applying a constant 3kV DC voltage to the electrodes of the dielectric elastomer ankle joint. Immediately after receiving the command, the high-voltage drive circuit rapidly increases its output voltage from the current value to the preset high voltage value within a very short time, such as 10 milliseconds. This high voltage is applied to the electrodes of the dielectric elastomer material constituting the ankle joint. Under the action of the high-intensity electric field, the molecular chains of the dielectric elastomer material are polarized and stretched, causing its macroscopic elastic modulus, i.e., physical torsional stiffness, to increase sharply and nonlinearly. This process increases the ankle joint's ability to resist rotation about its axis of rotation to the maximum value allowed by its material and structure within a millisecond timescale. When joint stiffness reaches its maximum value, it exhibits a physically locked state. This means that in the face of abnormal joint rotation caused by external impacts or foot slippage, the joint generates a powerful resisting torque through its immense rigidity, effectively suppressing abnormal joint angle changes. Through the aforementioned voltage application and sudden stiffness change operation, the robot's ankle joint rapidly and actively transforms from a pre-predicted, terrain-based state with a degree of compliance into a highly rigid support state with extremely high resistance to deformation. This state, the physically locked stable ankle joint state, provides an instantaneously stable support foundation for the robot's foot from a physical perspective.

[0054] The preset locking voltage signal is a DC or AC drive signal with an amplitude much higher than the conventional regulating voltage. Its specific voltage value is preset based on the characteristics of the selected dielectric elastomer material and its structural design to achieve the maximum stiffness the joint can generate, for example, 3kV. Millisecond time refers to the total duration from issuing the voltage control command to the ankle joint reaching its maximum stiffness. This time is determined by the response speed of the high-voltage drive circuit, the strain rate of the dielectric elastomer material, and the inertia of the mechanical structure, and is typically designed to be less than 50 milliseconds. Physical torsional stiffness refers to the ability of the dielectric elastomer ankle joint to resist torsional deformation about its functional axis. Its value is equal to the ratio of the applied torque to the resulting angle of rotation, and the unit is N·m / rad. The maximum value refers to the stiffness limit achievable by applying the maximum permissible safe voltage under the specific design and material parameters of the dielectric elastomer ankle joint. A locking joint refers to a state in which the joint stiffness is brought to or near its maximum value, thereby greatly restricting or even temporarily prohibiting free rotation of the joint. The ability to resist abnormal rotation caused by external impact or slippage refers to the ability of a highly rigid ankle joint to generate a sufficiently large internal resistance torque when the sole of the foot is subjected to lateral impact or tangential slippage, preventing the foot from tilting at a large angle relative to the lower leg that could lead to instability.

[0055] S6. Identify the abnormal direction of plantar pressure that causes a stable ankle joint state that leads to physical locking, reverse the calculation to counteract the shift of the body's center of mass caused by this abnormal plantar pressure, and calculate the corresponding joint angle adjustment value to generate a center of mass compensation adjustment amount for coordinating and stabilizing the body posture.

[0056] In a specific embodiment of the present invention, identifying the abnormal direction of plantar pressure that triggers a stable ankle joint state that forms a physical lock includes: analyzing the movement trajectory of the pressure center point and the distribution of local pressure peak positions just before triggering the stable ankle joint state that forms a physical lock.

[0057] If the trajectory of the pressure center point shows that the velocity component moving towards the outer side of the foot is greater than its corresponding normal fluctuation range threshold, then the abnormal direction of the foot pressure is determined to be outward. The value of the normal fluctuation range threshold is consistent with the boundary value of the normal fluctuation range threshold of the lateral movement velocity of the pressure center in the normal interaction model for the current ankle joint basic stiffness setting.

[0058] If the local pressure peak appears in the posterior region of the heel and its absolute amplitude is greater than the upper limit threshold of the local pressure peak set in the normal interaction model, then the abnormal direction of the plantar pressure is determined to be posterior.

[0059] In one specific embodiment of the present invention, the lateral velocity threshold for the pressure center used to identify abnormal plantar pressure direction is consistent with the boundary value of the normal fluctuation range threshold for the lateral movement velocity of the pressure center in the normal interaction model, i.e., the lateral velocity threshold for the pressure center is 0.12 m / s. The upper limit threshold for local pressure peak is the 97.5 percentile of the local pressure peak obtained through offline learning. Actual testing has verified that this value is approximately 1.3 times the robot's own weight.

[0060] In a specific embodiment of the present invention, the corresponding joint angle adjustment value is calculated to generate a center of mass compensation adjustment amount for coordinating and stabilizing body posture. This includes: based on a single rigid body dynamics model, querying a predefined mapping table to calculate the body center of mass offset required to generate the torque to counteract the disturbance. The body center of mass offset includes the target offset direction and its suggested offset velocity. The target offset direction is opposite to the abnormal direction of plantar pressure.

[0061] It should be noted that the predefined mapping table is constructed through the following steps: First, a single rigid body model of the robot is established in the dynamics simulation software. Then, during the single-leg support phase, a series of disturbances with different directions (outer, inner, front, and rear) and different intensities (quantified as impact force or sliding velocity) are applied to the sole of the foot. For each disturbance, the minimum centroid offset (in the opposite direction to the disturbance) required to restore the robot's zero-moment point to the center of the support polygon and the average velocity corresponding to that offset in the shortest time are calculated using an optimization algorithm. The simulation results form a set of raw data pairs: "disturbance characteristics (direction + intensity) → centroid compensation amount (offset direction + velocity)". Finally, a verification experiment is conducted on an actual robot platform to fine-tune and calibrate the simulation results, and the discrete data is normalized into an instruction mapping table that can be directly queried through the disturbance direction. For example, a record in the mapping table could be: "Abnormal direction: outer; Disturbance intensity: medium (lateral velocity of pressure center 0.2 m / s); Output: centroid target offset direction: inner; Suggested offset velocity: 0.05 m / s".

[0062] Using the robot's leg kinematics model, the target offset direction and suggested offset velocity of the body's center of mass are converted into hip joint angle adjustment values ​​and knee joint angle adjustment values ​​on the supporting leg side, which are used as center of mass compensation adjustment amounts.

[0063] It should be noted that, next, based on the identified abnormal plantar pressure direction, the system performs a reverse calculation. The calculation process is based on a simplified single-body dynamics model of the robot. Its core principle is: to counteract plantar disturbances from a certain direction, the robot's center of mass needs to be actively shifted to the opposite side of that direction to generate a restoring torque that counteracts the disturbance torque. Based on the distance of the pressure center shift or the amplitude of the local peak, the system queries a predefined mapping table obtained from offline dynamics simulation and experimental calibration to determine the target shift direction of the body's center of mass required to effectively counteract the disturbance, as well as the suggested shift velocity required to reach that position. The shift direction is always opposite to the identified abnormal pressure direction. The shift velocity is positively correlated with the quantified value of the current disturbance level and the robot's current movement speed. Finally, combining the robot's actual movement speed, the system uses the robot's leg kinematics model to convert the calculated center of mass shift direction and shift velocity into specific angle adjustment values ​​for the servo motors of the hip and knee joints on the supporting leg. This transformation is achieved through calculation, specifically by inversely calculating the angle changes of the hip and knee joints that can achieve the desired center of gravity shift based on the desired direction and velocity of the shift. These calculated joint angle adjustment values ​​are encapsulated into a structured control instruction package, which generates the center of gravity compensation adjustment for coordinated body posture stabilization.

[0064] It's also worth noting that the reverse engineering process can be understood as a simplified geometric problem. In the robot's control system, a mathematical model describing its body configuration is pre-established. This model specifies the spatial position of the body's center of mass when the hip and knee joints of the supporting legs are bent at different angles. Therefore, when the control system determines a desired direction and velocity of center of mass offset (i.e., the next target position the center of mass needs to move to) to maintain balance, the calculation it performs is not a forward deduction of "Where will the center of mass go if the joints move like this?", but rather a reverse query or solution using the model: "To move my body's center of mass precisely to that target position, how many degrees do my hip and knee joints need to bend?" Through this reverse reasoning, the system can directly obtain a set of target hip and knee joint angles corresponding to the target center of mass position. The difference between the current angle and the target angle is the final angle change that needs to be performed.

[0065] Here, "confirmation" refers to the process by which the system verifies, by reading status flags or monitoring specific signals, that the stable ankle joint state forming a physical lock has been activated and is functioning. The abnormal plantar pressure direction that triggers this state refers to the directional characteristics of the plantar force corresponding to the root cause when the instantaneous stiffness locking command is triggered in step S4, such as the sliding or impact direction towards the outside, inside, forward, or backward. "Inverse calculation" refers to the process of deriving the active compensation action of the robot body required to counteract the perceived external disturbance effect. "Body center of mass offset direction" refers to the orientation in which the robot's torso center of mass needs to move relative to the current supporting polygon to stabilize its posture. "Body center of mass offset velocity" refers to the suggested rate at which the robot's torso center of mass moves towards the target offset direction; this velocity value needs to counteract the disturbance as quickly as possible while ensuring stability. "Current robot motion velocity" refers to the velocity vector of the robot traveling in the horizontal plane, the value of which is directly given by the wheel encoder, inertial measurement unit fusion positioning, or gait generator. "Calculation" refers to the process of obtaining a definite value through mathematical operations or table lookup mapping. The specific angle adjustment values ​​for the hip and knee joints refer to the incremental or absolute change in the target joint angle relative to its currently planned angle, which is required to drive the servo motors of the corresponding joints. The unit is degrees. The center of gravity compensation adjustment amount is a data package containing the specific angle adjustment values ​​planned for one or more joints to perform center of gravity compensation, as well as the corresponding execution timing information.

[0066] S7. Drive the hip and knee joints to perform center of gravity compensation adjustments to stabilize the body's center of gravity, and evaluate the matching degree between the gait strategy and road conditions based on the magnitude and frequency of center of gravity compensation. Adjust the default stride length and leg lift height for the next step, generate optimized gait parameters, and input them into the next gait control cycle.

[0067] In a specific embodiment of the present invention, generating optimized gait parameters and inputting them into the next gait control cycle includes: inserting the centroid compensation adjustment amount as a high-priority instruction into the servo motor control queue of the hip and knee joints of the robot's supporting leg for execution.

[0068] It should be noted that the system receives the center of mass compensation adjustment amount generated in step S6, which includes specific angle adjustment values ​​for the hip and knee joints. First, the system marks the center of mass compensation adjustment amount as a high-priority instruction and inserts it at the front of the real-time control instruction queue for the robot's leg joint servo motors. The sorting logic of this instruction queue ensures that high-priority instructions are parsed and executed first. The servo motor controller parses the instruction, obtains the target angle adjustment value, and drives the corresponding hip and knee joint servo motors for rapid fine-tuning. The fine-tuning process is achieved through closed-loop position control; the motors adjust the joint angles from the current value to the target value within tens of milliseconds, thereby changing the leg configuration and driving the body's center of mass to move in the offset direction calculated in step S6, ultimately achieving the goal of stabilizing the body's center of mass.

[0069] After the centroid compensation is completed, the absolute value of the angle adjustment range of this centroid compensation and the compensation trigger frequency per unit time are calculated.

[0070] It should be noted that, following the confirmation that the center of gravity compensation action has been completed, the system analyzes the center of gravity compensation events triggered in this instance and in recent historical data. The analysis mainly includes calculating the magnitude of the current center of gravity compensation, i.e., the absolute value of the joint angle adjustment performed; and calculating the frequency of center of gravity compensation, i.e., the number of times center of gravity compensation is triggered per unit time.

[0071] If the absolute value of the angle adjustment of the center of mass compensation or the compensation trigger frequency per unit time is greater than the corresponding preset threshold, the current gait strategy is determined to be too aggressive. According to the proportion proportional to the magnitude and frequency of the center of mass compensation, the default stride value and default leg lift value of the next step are reduced to generate more conservative optimized gait parameters.

[0072] It should be noted that if the absolute value of the angle adjustment range for this centroid compensation is... Greater than its preset threshold Or the compensation trigger frequency per unit time Greater than its preset threshold If the current gait strategy is poorly matched to the actual road conditions, then the default stride length for the next step is adjusted. Leg lift height The specific formula is:

[0073] ,

[0074] ,

[0075] in, , , , To adjust the coefficients, the robot was controlled to move using different combinations of coefficients on a mixed test road containing flat, pebble, and sloping surfaces. The number of times stiffness locking was triggered within the total walking distance was used as the optimization objective function. A hill-climbing method was employed for optimization, resulting in a set of coefficients that minimized the objective function: =0.15, =0.1, =0.2, =0.15. New parameters generated [ , This refers to more conservative optimization of gait parameters.

[0076] It should also be noted that the threshold and Determining the robot's stability boundary. The maximum allowable joint angle adjustment for a single center of mass compensation was set. Dynamic analysis showed that when the combined adjustment of the hip and knee joints caused a lateral displacement of the center of mass exceeding 0.08m, the robot was at the tipping point. Inverse kinematics analysis determined that this displacement corresponded to an equivalent hip joint angle change of approximately 8 degrees. The maximum number of compensations tolerated per unit time is set based on the gait cycle and system response time. It is 2Hz.

[0077] In a specific embodiment of the present invention, the aforementioned preset thresholds are key boundary values ​​set based on stability margin analysis of the robot's dynamics model and statistical data from numerous physical or simulation experiments. The preset threshold for compensation amplitude specifically refers to the upper limit of the single-step joint angle change or center of mass displacement calculated to achieve center of mass shift. Its value is determined by the maximum allowable center of mass shift that ensures the robot's zero-torque point remains within the stable support polygon of the supporting foot. Typically, a certain safety margin (e.g., 80% of the theoretical maximum value) is reserved on top of this theoretical limit to cope with model errors and unknown external disturbances. The preset threshold for compensation frequency refers to the upper limit of the number of times center of mass compensation actions are initiated per unit time. Its value is determined by distinguishing between normal posture fine-tuning of the robot and high-frequency oscillations caused by continuous instability. This value (e.g., set to 5Hz) is derived through experimental statistics. When the frequency of compensation actions exceeds this value, it usually means that adjusting the upper body posture alone is insufficient to suppress the instability trend, and the system is about to enter an unrecoverable oscillation state. Therefore, the setting of these two thresholds together constitutes a quantitative basis for judging whether the current stabilization strategy is about to fail.

[0078] High-priority instructions refer to control commands that have the highest execution priority in the control system's instruction scheduling. They are set to ensure that compensation actions are executed immediately when stability is threatened, without being delayed by other routine tasks. A joint servo motor is a motor and its drive control system capable of precisely controlling the output shaft's rotation angle or position. Drive refers to the controller sending pulse or voltage signals to the servo motor to produce rotational motion. Rapid fine-tuning refers to the servo motor completing a relatively small angle adjustment within a short time after receiving the instruction, typically on a timescale of tens of milliseconds. Stabilizing the body's center of mass refers to adjusting the leg joints to change the position of the robot's torso, moving its center of mass to a more stable or optimal position within the supporting polygon. Completing center of mass compensation means that the servo motor has reached the target angle specified by the instruction and entered a position-holding state, and the system confirms through sensor feedback that the center of mass position has changed as expected. The degree of matching with actual road conditions is a qualitative or semi-quantitative evaluation result used to measure whether the parameter settings of the current gait strategy are suitable for the robot's actual walking ground environment. It is indirectly evaluated by analyzing the statistical characteristics of compensation events. Appropriate reduction refers to numerically reducing gait parameters based on evaluation results and according to preset logical rules. Default stride length refers to the distance the gait generator plans for the robot's single-step horizontal movement without adaptive adjustments. Default leg lift height refers to the maximum vertical height the swing foot lifts off the ground during the swing phase, planned by the gait generator without adaptive adjustments. More conservative gait parameters are characterized by reduced stride length and height, aiming to sacrifice some walking speed for higher stability and safety. Optimized gait parameters refer to the new set of gait parameter values ​​generated after this step's evaluation and adjustment, including updated stride length and leg lift height. The next gait control cycle refers to the next complete gait planning and execution cycle immediately following the current gait cycle.

[0079] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A gait stabilization control method for a guide robot integrating IMU and foot force sensing, characterized in that, Includes the following steps: S1. Activate the robot's head depth camera to collect 3D point cloud data in the direction of travel, analyze terrain features and evaluate perception reliability based on ambient lighting conditions, and generate terrain safety confidence scores that characterize the reliability of visual prediction and the complexity of terrain. S2. Call the stiffness mapping rule, select the voltage value corresponding to the terrain safety confidence, and apply it to the dielectric elastomer ankle joint of the robot's foot, so that the physical stiffness of the dielectric elastomer ankle joint is the ankle joint basic stiffness setting value that matches the terrain prediction. S3. During the robot's single-leg support phase, with the ankle joint's basic stiffness set as the initial state, data from the foot flexible pressure sensor array and the trunk inertial measurement unit are collected simultaneously to generate a real-time foot interaction data package describing the current foot-ground physical interaction and the robot's dynamic trend. S4. Compare the parameters in the real-time plantar interaction data packet with the preset normal interaction model. When the parameters are detected to exceed the normal fluctuation range, immediately generate an instantaneous stiffness locking command to deal with the risk of sudden instability. S5. Convert the instantaneous stiffness locking command into a preset locking voltage signal and apply it to the dielectric elastomer ankle joint to increase the physical torsional stiffness of the dielectric elastomer ankle joint to the maximum value in order to lock the dielectric elastomer ankle joint, resist abnormal rotation, and form a physically locked stable ankle joint state. S6. Identify the abnormal direction of plantar pressure that causes the stable ankle joint state to form physical locking, reversely calculate the body center of mass shift caused by the abnormal plantar pressure, calculate the corresponding joint angle adjustment value, and generate the center of mass compensation adjustment amount for coordinating and stabilizing body posture. S7. Drive the hip and knee joints to perform center of gravity compensation adjustments to stabilize the body's center of gravity, and evaluate the matching degree between the gait strategy and road conditions based on the magnitude and frequency of center of gravity compensation. Adjust the default stride length and leg lift height for the next step, generate optimized gait parameters, and input them into the next gait control cycle.

2. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 1, characterized in that, The generated terrain safety confidence score, which characterizes the reliability of visual predictions and the complexity of terrain, includes: In the 3D point cloud data, a grid representing the landing area is drawn, the variance of the height value of each point in the unit grid is calculated, and the average of the variance of the height value of all unit grids is used as the flatness quantification value. The reference plane is fitted based on the 3D point cloud data, and the number of points with a height higher than the reference plane within a unit grid area is used as the obstacle density value. Plane fitting is performed on the 3D point cloud data, the angle between the normal vector of the fitted plane and the direction of gravity is calculated, the slope angle of the current area is obtained, and the rate of change of the slope angle in the direction of travel is tracked to obtain the slope change trend value. The flatness measurement value, obstacle density value, and slope change trend value are compared with the preset level thresholds, and combined with the perception reliability based on the environmental lighting conditions, three discrete levels of high, medium and low are output as the terrain safety confidence level.

3. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 2, characterized in that, The specific analysis method for the perception reliability based on ambient lighting conditions is as follows: using the visible light sensor built into the depth camera, the illuminance value of the current environment is acquired in real time. The illuminance value of the current environment is compared with the illuminance value range corresponding to each perception reliability stored in the database. If the illuminance value of the current environment is within the illuminance value range corresponding to a certain perception reliability, then the perception reliability is taken as the perception reliability of the current environment. The perception reliability includes high, moderate and low.

4. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 1, characterized in that, The method of setting the physical stiffness of the dielectric elastomer ankle joint to a base stiffness setting that matches the terrain prediction includes: The stiffness mapping rule defines the output voltage range corresponding to high, medium, and low terrain safety confidence levels through a lookup table; Based on the currently received level of terrain safety confidence, a preset voltage value is selected from the corresponding output voltage range; A preset voltage value is applied to the electrodes of the dielectric elastomer ankle joint through a high-voltage drive circuit. By utilizing the characteristic of the dielectric elastomer material to change the macroscopic elastic modulus under the action of an electric field, a basic stiffness setting value for the ankle joint that matches the terrain prediction is generated.

5. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 1, characterized in that, The generation of real-time plantar interaction data packets describing the current foot-ground physical interaction and robot dynamic trends includes: When the robot is determined to be in the single-leg support stage, pressure matrix data output by the flexible pressure sensor array on the sole of the foot and attitude angular velocity data output by the trunk inertial measurement unit are collected simultaneously. The coordinates of the pressure center on the sole of the foot and its movement speed over time are calculated based on the pressure matrix data, and local pressure peaks in the pressure matrix that are greater than their neighborhood values ​​are identified. Establish data alignment timestamps, and package and combine the pressure center coordinates, movement speed, local pressure peak and attitude angular velocity data at the same time to generate real-time plantar interaction data packets.

6. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 5, characterized in that, The generation of instantaneous stiffness locking commands for responding to sudden instability risks includes: A predefined normal interaction model is used, which sets threshold values ​​for the normal fluctuation range of pressure center movement speed, local pressure peak and posture angular velocity during stable walking for different ankle joint basic stiffness settings. When the movement speed of the pressure center toward the edge of the foot exceeds its corresponding normal fluctuation range threshold, or the local pressure peak exceeds the expected normal fluctuation range threshold of the current gait phase, or the attitude angular velocity exceeds its corresponding normal fluctuation range threshold, it is determined that there is a risk of sudden instability, and an instantaneous stiffness locking command is immediately generated.

7. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 1, characterized in that, The stable ankle joint state that forms physical locking includes: Receive instantaneous stiffness locking command and parse the preset locking voltage signal in the instantaneous stiffness locking command; Within milliseconds, the high-voltage drive circuit increases the voltage applied to the dielectric elastomer ankle joint to the preset locking voltage signal. By using the electric field-induced material polarization and stretching, the physical torsional stiffness of the ankle joint increases nonlinearly and rapidly to the maximum allowable value. By utilizing the resisting torque generated at the maximum value, a physically locked and stable ankle joint state is formed.

8. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 5, characterized in that, The identification of abnormal plantar pressure directions that trigger a stable ankle joint state that results in physical locking includes: Analyze the movement trajectory of the pressure center point and the distribution of local pressure peak locations just before triggering the stable ankle joint state that forms physical locking. If the movement trajectory of the pressure center point shows that the movement speed component towards the outer side of the foot is greater than its corresponding normal fluctuation range threshold, then the abnormal direction of the foot pressure is determined to be outward. The value of the normal fluctuation range threshold is consistent with the boundary value of the normal fluctuation range threshold of the lateral movement speed of the pressure center in the normal interaction model for the current ankle joint basic stiffness setting. If the local pressure peak appears in the posterior region of the heel and its absolute amplitude is greater than the upper limit threshold of the local pressure peak set in the normal interaction model, then the abnormal direction of the plantar pressure is determined to be posterior.

9. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 8, characterized in that, The calculation of the corresponding joint angle adjustment values ​​generates a center of mass compensation adjustment amount for coordinating and stabilizing body posture, including: Based on the single rigid body dynamics model, a predefined mapping table is queried to calculate the body center of mass offset required to generate the torque to counteract the disturbance. The body center of mass offset includes the target offset direction and its suggested offset velocity. The target offset direction is opposite to the direction of abnormal plantar pressure. Using the robot's leg kinematics model, the target offset direction and suggested offset velocity of the body's center of mass are converted into hip joint angle adjustment values ​​and knee joint angle adjustment values ​​on the supporting leg side, which are used as center of mass compensation adjustment amounts.

10. The gait stabilization control method for a guide robot integrating IMU and foot force sensing according to claim 1, characterized in that, The process of generating optimized gait parameters and inputting them into the next gait control cycle includes: The center of mass compensation adjustment amount is inserted as a high-priority instruction into the servo motor control queue of the hip and knee joints of the robot's support leg for execution; After the centroid compensation is completed, the absolute value of the angle adjustment range of this centroid compensation and the compensation trigger frequency per unit time are calculated. If the absolute value of the angle adjustment of the center of mass compensation or the compensation trigger frequency per unit time is greater than the corresponding preset threshold, the current gait strategy is determined to be too aggressive. According to the proportion proportional to the magnitude and frequency of the center of mass compensation, the default stride value and default leg lift value of the next step are reduced to generate more conservative optimized gait parameters.

Images