Blind guiding robot speed adjusting method based on gait recognition and ground state

Abstract

The invention discloses a blind guiding robot speed adjusting method based on gait recognition and ground state, and the method comprises the steps: firstly, collecting the gait data of a user and the ground state data in front of a robot in real time through a sensor; secondly, on the basis of the ground state data, a ground sliding risk index is obtained through quantitative calculation by fusing visual texture analysis, terrain gradient calculation and specific wet and slippery substance identification results; meanwhile, on the basis of user gait data, a gait stability risk index is obtained through quantitative calculation by analyzing stride frequency abnormity and a trunk swing degree; and finally, the two risk indexes are substituted into a self-adaptive speed control function, and the advancing speed of the blind guiding robot is dynamically adjusted. According to the method, the user gait risk and the ground environment risk are cooperatively controlled, intelligent self-adaptive adjustment of the speed of the robot is achieved, active speed reduction can be achieved before potential dangers occur, and the safety and comfort of the guiding process are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of service robots, artificial intelligence, human behavior modeling and intelligent assistive devices, in particular to a blind guiding robot speed adjustment method based on gait recognition and ground state, which is used to improve the safety and comfort of the robot guiding blind people during walking. BACKGROUND

[0002] Current blind guiding robots and other intelligent assistive devices have basic path planning and obstacle avoidance capabilities, and can guide users to move from point A to point B. However, these systems often simplify the guiding process as a purely physical traction task in practical applications, ignoring the complexity of the two dynamic factors of human and environment.

[0003] Specifically, the existing technology mainly has the following defects: Insufficient perception of environmental risks: although the blind guiding robot can avoid macro obstacles such as walls and pedestrians, it cannot dynamically evaluate and respond to micro ground state risks such as wetness, icing, unevenness or oil stains. This makes users still face a high risk of slipping and falling when passing through these dangerous road surfaces.

[0004] Lack of perception of user state: existing technology usually assumes that the user can always keep up with the pace of the blind guiding robot. However, in fact, the physical state of blind users is dynamically changing, for example, due to nervousness, fatigue or fear, resulting in unstable gait, abnormal step frequency or body shaking. In this case, if the blind guiding robot maintains the original speed, it is easy to be out of sync with the user's pace, resulting in a strong pulling sensation, and even directly causing the user to lose balance and fall.

[0005] Poor human-machine collaboration: due to the lack of comprehensive perception of the environment and user state, the speed adjustment strategy of the blind guiding robot is usually static or based on simple obstacle avoidance logic, and cannot realize true human-machine collaboration. This leads to a rigid and mechanical guiding experience, lacking in safety and comfort.

[0006] Therefore, there is an urgent need for a method that can dynamically adjust the speed of the robot dog based on real-time and comprehensive evaluation of the stability of the blind user's own gait and the safety of the ground in front, in order to achieve a more safe, smooth and personalized human-machine collaborative navigation experience. SUMMARY

[0007] This invention primarily addresses the technical problem of insufficient safety and comfort in existing guide robots during guidance due to neglect of user gait stability and ground environment risks. It provides a speed adjustment method for guide robots based on gait recognition and ground condition. This method uses multimodal sensors to collect user gait characteristics and ground condition information in real time, constructs and quantifies gait stability risk and ground slippage risk, and dynamically and adaptively adjusts the guide robot's speed based on these two core risk indicators through a unified speed control function.

[0008] The present invention addresses the aforementioned technical problems primarily through the following technical solution: a speed adjustment method for a guide robot based on gait recognition and ground conditions, comprising the following steps: S1: Data Acquisition: Real-time acquisition of first and second types of data through sensors deployed on the guide robot or the user. Sensors on the guide robot include LiDAR, depth camera, etc., and sensors on the user include rope tension sensor, IMU (Inertial Unit), foot pressure pad, etc. The first type of data is used to characterize the user's gait stability, such as parameters such as blind person's gait frequency, stride length, and gait rhythm. The second type of data is used to characterize the ground condition in front of the guide robot, such as information on ground slope, texture changes, and slippery areas. S2: Calculate the ground slip risk index R slip Based on the second type of data, calculate the ground physical characteristic risk S. friction Topographic slope risk S slope Risk of identifying specific substances S stain Then, the three risks mentioned above are weighted and fused to obtain the ground slip risk index R. slip ; S3: Calculate the gait stability risk index Rgait: Based on the first type of data, calculate the gait frequency risk α(f step ) and trunk sway risk function β(a torso Then, the step frequency risk α(f) is calculated. step ) and trunk sway risk function β(a torso The fusion yields gait stability risk indicators; S4: Adaptive speed regulation: based on ground slip risk index R slip and gait stability risk index R gait The travel speed v of the guide robot is dynamically adjusted according to the following speed control function. t v t =v base ×(1-γ1×R slip -γ2×R gait ); where v baseThe user can customize the base speed, and γ1 and γ2 are adjustment coefficients. When there is a risk of slipping or unstable gait, the robot automatically slows down and prepares to stop, enhancing the blind person's sense of control and safety.

[0009] The user can select a set of basic γ values through the pre-set "cautious mode", "normal mode", and "sports mode". These basic values are obtained through experimental calibration, ensuring safety and comfort in general scenarios.

[0010] After the user selects the basic mode, the system does not use this set of γ values unconditionally, but dynamically adjusts the two weights in real time according to the external environmental information perceived by the system. This makes the guide robot more intelligent and able to automatically switch decision preferences in different scenarios.

[0011] Examples are as follows: Environmental information acquisition: The system uses laser radar or depth camera to analyze the openness of the space around the robot. For example, by calculating the obstacle density or the width of the passable path within a certain distance (such as 5 meters) in front of the robot.

[0012] Open environment: such as squares, wide sidewalks. Low obstacle density.

[0013] Narrow / complex environment: such as indoor corridors, crowded streets. High obstacle density.

[0014] Dynamic adjustment logic: When the system determines that it is currently in an open environment, it means that the user has more reaction space and avoidance space. At this time, the system can appropriately reduce the value of γ (for example, reduce by 10%-20% based on the user's selected mode). This makes the robot's speed adjustment more gentle in the presence of slight ground or gait risks, improving the smoothness of travel.

[0015] When the system determines that it is currently in a narrow / complex environment, any slight gait instability or wet ground may lead to a collision, with higher risk. At this time, the system will automatically increase the value of γ (for example, increase by 10%-20% based on the user's selected mode). This makes the robot's response to risk more sensitive and rapid, prioritizing absolute safety in complex environments.

[0016] As a preferred embodiment, the step S2 is specifically: S21: Calculate the ground physical property risk S friction : Collect the front ground texture image through the depth camera and input it into the pre-trained convolutional neural network model to obtain the texture wetness probability P texture ; At the same time, the current location is obtained through the GPS, and the historical risk map of the built-in history is queried to obtain the historical risk score H riskTake the maximum of the two as S. friction S friction =max(P texture H risk ); The risk value for ground physical characteristics ranges from [0,1], and this factor combines visual texture and historical data.

[0017] Visual texture analysis: Depth cameras deployed on guide robots capture texture images of the ground in front of them. These images are then fed into a pre-trained convolutional neural network (CNN) model. By learning from a large number of image samples of both "dry" and "slippery" surfaces (such as polished marble or wet tiles), this model can analyze the visual texture of the current ground and output a slipperiness probability P. texture The range is between [0,1]. texture The higher the value, the closer the surface texture is to a slippery material.

[0018] Historical Risk Area Query: The robot obtains its current location coordinates via GPS. The system queries a built-in "Historical Slip Risk Map." This map records known areas where slippery conditions or slips have historically occurred frequently (such as specific subway station exits or market entrances that are frequently watered). If the current location is within a risk area, a historical risk score H is obtained based on the risk level of that area. risk (For example, 0.8 for high-risk areas and 0.5 for medium-risk areas); if not in any known risk area, H risk =0.

[0019] Final S friction Take the larger of the two to ensure a response to either source of risk. S22: Calculate the terrain slope risk S slope The ground slope angle θ in front of the guide robot is obtained by measuring its attitude using an inertial measurement unit (IMU). slope And according to formula S slope =min(1.0,(θ slope / θ max ) 2 ) Calculate the terrain slope risk, where θ max This is the preset upper limit of the safe slope; The range of values ​​for terrain slope risk is [0,1]. This term is used to quantify the risk of slipping due to ground inclination.

[0020] The robot's current attitude is measured using its built-in IMU (Inertial Measurement Unit), thus obtaining the slope angle θ of the ground in front of it. slope (Unit: degrees). In physics, the steeper the slope, the greater the downward component of gravity along the slope, making it easier to slip and fall. Therefore, a safe upper limit for the slope θ is set.max (eg 15 degrees), considering that beyond this slope the risk is maximum. The formula is: S slope = min(1.0, (θ slope / θ max ) 2 ), this square relationship makes the risk grow slowly at small slopes, and rapidly to 1 when approaching or exceeding θ max . S23: Calculate the specific substance identification risk S stain : Analyze the ground image by a visual identification model, detect whether there is water stain, ice or oil stain, and directly take the confidence score of this model as S stain ; The specific substance identification risk is in the range [0,1], which is used to directly identify specific slippery substances such as water stains, ice and snow, etc.

[0021] Use a visual identification algorithm (such as YOLOv5 model) to analyze the ground image captured by the camera. This model is specially trained to identify specific targets such as "water stain", "ice", "oil stain", etc. When the model detects these targets on the robot's forward path, it outputs a confidence score Confidence (range [0,1]). S stain Directly take this confidence score. For example, the model detects a patch of water in front with 95% confidence, then S stain = 0.95. If no specific slippery substance is detected, S stain = 0.

[0022] S24: Weighted fusion of the above three risks to obtain the ground slip risk indicator R slip , the calculation formula is: R slip = w1 x S friction + w2 x S slope + w3 x S stain , where w1, w2 and w3 are preset weight coefficients and the sum is 1, these weights reflect the importance of different risk factors.

[0023] For example, the weight coefficients can be set as follows: w1 = 0.3 (ground physical properties are the basic risk), w2 = 0.2 (slope is an important auxiliary risk), w3 = 0.5 (directly seeing water / ice / snow is the most direct and dangerous signal, so the weight is the highest). Through this weighted formula, the final R slip value normalized to the [0,1] interval can be calculated. The higher the value, the greater the risk of the ground in front of the robot.

[0024] As a preferred, the step S3 is specifically: S31: Calculate the step frequency risk function α(fstep ):Collect periodic acceleration data of the user through the inertial measurement unit and calculate the current step frequency f step ; Set the normal step frequency interval [f min , f max ], when f min ≤f step ≤f max , α(f step )=0; when f step is not in the interval, the value of α(f step ) increases with the deviation; S32: Calculate the trunk swing risk function β(a torso ), which is calculated as follows: collect the acceleration data of the user's trunk in the lateral direction through the inertial measurement unit, and calculate the variance a torso in a time window, according to the formula β(a torso )=min(1,(a torso / a thresh ) 2 ) calculation, wherein a thresh is a preset trunk lateral acceleration variance threshold; S33: Fusion of the above two risks, get the gait stability risk index R gait , the calculation formula is: R gait =α(f step )+β(a torso ).

[0025] As preferred, in step S31, when f step <f min , α(f step )=C1×(f min -f step ); when f step >f max , α(f step )=C2×(f step -f max ), wherein C1 and C2 are preset proportional coefficients.

[0026] α(f step ) represents the step frequency variance, and β(a torso ) is the trunk acceleration fluctuation amplitude (normalized processing), and the overall value range is [0, 1].

[0027] In the present application, the gait risk index R gait is composed of the step frequency risk function α(f step ) and the trunk swing risk function β(a torso ). The specific calculation process is as follows: Parameter definition and acquisition: fstep User cadence (unit: Hertz Hz) is calculated in real time by detecting the periodic acceleration changes of steps through an IMU (inertial measurement unit) installed on the guide cane or on the user.

[0028] a torso The variance of the user's lateral trunk acceleration characterizes the degree of body sway. It is obtained by collecting lateral (left-right) acceleration data of the user's torso (e.g., waist) using an IMU and calculating its variance within a short time window (e.g., 2 seconds). A larger variance indicates more severe body sway and a more unstable gait.

[0029] Step frequency risk function α(f step This is used to quantify the risk posed by abnormal gait frequencies (too fast or too slow). A normal, stable gait has an ideal gait frequency range. A normal gait frequency range is defined as [f...]. min ,f max For example, [1.5Hz, 2.0Hz]. When the user's step frequency f step Within this range, the risk is considered low, α(f) step When f approaches 0. step Beyond this range, the risk increases with the magnitude of the deviation.

[0030] C1 and C2 are preset scaling factors used to normalize the calculation results to the [0,1] interval. For example, a step frequency deviation corresponding to the maximum risk can be set to determine the values ​​of C1 and C2.

[0031] The trunk sway risk function quantifies the risk associated with body swaying. The more intense the swaying, the higher the risk. A threshold 'a' is set for the variance of the trunk lateral acceleration. thresh Exceeding this threshold indicates significant bodily instability. When a torso When a is small, risk β increases more slowly; when a torso Approaching or exceeding the threshold at hresh At this point, the risk increases rapidly and is eventually limited to a maximum of 1. This aligns with reality, where slight shaking is normal, while violent shaking is extremely dangerous.

[0032] Preferably, a speed adjustment method for a guide robot based on gait recognition and ground conditions also includes an emergency stop judgment step: setting a preset ground safety threshold T. slip and gait safety threshold T gait (For example, T) slip =0.8,T gait =0.85); when the real-time calculated R slip >T slipWhen the ground is extremely slippery (e.g., large areas of water or ice are detected), the system determines there is a serious risk of slipping; when the real-time calculated R... gait >T gait If the above conditions are met, it indicates that the user's gait is extremely unstable (such as staggering or about to fall), and the system determines that there is a serious gait risk. If any of the above conditions are met, the system will immediately execute the "automatic deceleration and prepare to stop" command and trigger the highest priority alarm (such as a strong voice prompt "Danger! Please stand firm!" and high-frequency vibration).

[0033] Preferably, a method for adjusting the speed of a guide robot based on gait recognition and ground conditions also includes a multimodal feedback step: when the calculated walking speed v t When the speed drops below the preset low-speed threshold, the voice prompt module and vibration module are automatically activated. The system provides feedback on potential risks to the user through voice prompts such as "Slippery ahead, please be careful" and vibration rhythms related to the level of risk.

[0034] The substantial effects of this invention are: It achieves joint control of blind person's state perception and ground state perception: it quantifies the user's own dynamic gait risk and the external environment's dynamic ground risk and integrates them into a unified speed control model, realizing a leap from "passive obstacle avoidance" to "active risk prediction and collaborative control".

[0035] Improved walking safety and significantly reduced the risk of falls: By accurately identifying and quickly responding to various risk factors such as slipperiness, slope, and gait instability, it can proactively slow down or even stop before danger occurs, greatly reducing the probability of users slipping or being pulled down.

[0036] It provides a natural human-machine collaborative guidance experience, adaptable to different groups of people: the robot's speed is no longer fixed, but closely coupled with changes in the user's state and environment, making the guidance process smoother and more natural, just like an experienced human guide, and able to adapt to users with different walking habits and physical conditions.

[0037] It features high compatibility and low cost, and can be deployed on common robot dog platforms on the market: The sensors (such as IMU and depth camera) on which this invention relies are all commercially available mature components, and the algorithm model can also run on general computing platforms, making it easy to integrate and deploy on existing guide robot products. Attached Figure Description

[0038] Figure 1 This is a flowchart of a speed adjustment method for a guide robot based on gait recognition and ground conditions according to the present invention. Detailed Implementation

[0039] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0040] Example: The system of the present invention can be deployed on a guide robot (such as Unitree Go1), and mainly includes: Sensor modules include depth cameras (such as RealSense D435) and lidar for ground condition awareness, as well as inertial measurement units (IMUs) (such as BOSCH IMUs, which can be worn on the user's waist or integrated into a guide rope) and rope tension sensors for user gait awareness.

[0041] Risk assessment module: This is usually a processor built into the robot, which runs the core algorithm of this invention, is responsible for receiving sensor data and performing risk calculations.

[0042] Speed ​​control module: Based on the output of the risk assessment module, it sends speed commands to the robot's underlying drive system.

[0043] Feedback module: Includes a voice broadcast unit and a vibration unit, used to provide risk alerts to users.

[0044] like Figure 1 As shown, the specific implementation steps of the speed adjustment method of the present invention are as follows: Step 1: Data Collection Gait recognition data acquisition: Acceleration and angular velocity data of the user's torso in three axes are acquired at a frequency of 100Hz using an IMU worn on the user's waist. Simultaneously, signals related to the user's gait rhythm are acquired through an IMU mounted on a guide rope or by analyzing the periodic changes of a rope tension sensor. These constitute the "first type of data" described in this invention.

[0045] Ground state perception data acquisition: Using a depth camera at the robot's front end, RGB and depth images of the ground in front (e.g., within a range of 0.5 to 3 meters) are acquired at a frequency of 30fps. Simultaneously, the robot's own attitude (pitch angle, roll angle) is obtained using an IMU. These constitute the "second type of data" described in this invention.

[0046] Step 2: Risk estimation model construction and calculation This step is performed within the risk assessment module, and its core is calculating two key indicators: the ground slip risk index R. slip and gait stability risk index R gait .

[0047] 2.1 Ground sliding risk index R slip Calculation (R) slip ∈[0,1]) R slipIt is a weighted combination of three independent risk sub-items, with a larger value indicating a more slippery and dangerous ground.

[0048] R slip =w1×S friction +w2×S slope +w3×S stain ; Here, w1, w2, and w3 are preset weight coefficients, and w1 + w2 + w3 = 1. These weights are not arbitrarily set but are based on experimental calibration. For example, a test set containing various risk combinations (such as dry slopes, slippery flat ground, and slopes with water accumulation) can be constructed, and the weights can be adjusted using optimization algorithms (such as grid search or gradient descent) to ensure that the calculated comprehensive risk index R... slip The correlation between the risk of slipping and the actual risk observed in testing (e.g., the number of slips as subjectively rated by testers or measured by sensors) is maximized. In a preferred embodiment, w1=0.3 (base risk), w2=0.2 (auxiliary risk), and w3=0.5 (direct hazard signal, with the highest weight) are set.

[0049] The calculation method for each sub-item is explained in detail below: Sub-item 1: Ground physical characteristics risk S friction This study integrates visual texture analysis and historical data.

[0050] S friction =max(P texture H risk ); Visual texture slip probability P texture The process involves inputting RGB ground images captured by a depth camera into a pre-trained convolutional neural network (CNN) model. The training process is as follows: First, a training dataset of at least 10,000 images is collected, covering samples of various materials (tiles, marble, cement, asphalt, etc.) in dry and slippery (after rain, watering, condensation, etc.) conditions, manually labeled as "dry" or "slippery". Then, using a mature CNN architecture such as ResNet-18, supervised learning is performed with these labeled images, training the model to output a probability value between 0 and 1, i.e., P. texture P texture The higher the value, the closer the current ground texture is to the "slippery" features learned by the model (such as enhanced specular reflection and blurred texture details).

[0051] Historical risk score H riskThe robot obtains its current geographic coordinates via a GPS module. The system has a built-in "historical slip risk map," which can be pre-entered by administrators (e.g., specific subway exits, entrances to perpetually slippery farmers' markets) or constructed using crowdsourced data (allowing multiple robots to upload location information when high-risk conditions are detected, which is then aggregated in the cloud to generate a heatmap). The map's data structure can be a hash table with geographic coordinates as keys and risk levels (0-1) as values. If the current location falls within an area of ​​the risk map, its corresponding risk level is taken as H. risk Otherwise H risk =0.

[0052] Sub-item 2: Terrain slope risk S slope This item is used to quantify the risk of slipping due to ground tilt.

[0053] S slope =min(1.0,(θ slope / θ max ) 2 ); The robot measures its own pitch angle using its built-in IMU, thereby obtaining the slope angle θ of the ground in front of it. slope (Unit: degrees). θ max This is a preset upper limit for the safe slope, such as 15 degrees. Exceeding this slope is considered to have reached the maximum risk of 1. A quadratic relationship is used instead of a nonlinear relationship because, in physics, the component of gravity along a slope is sinusoidally related to the slope. At small angles, it is approximately linear, but the increase in the risk of slipping is nonlinear, especially when the risk rises sharply near the limit angle. A quadratic relationship can better simulate this effect.

[0054] Sub-item 3: Risk of identifying specific substances S stain This item is used to directly identify specific slippery substances.

[0055] S stain The confidence score is directly taken from a visual recognition model (such as YOLOv5). This model is specifically trained to detect specific target categories on the ground, such as "water stains," "ice surfaces," and "oil stains." The training data consists of a large number of labeled images containing these targets. When the model detects a puddle of water on the robot's path and outputs a confidence score of 0.95, then S... stain =0.95. If no target is detected, then S stain =0.

[0056] 2.2 Gait stability risk index R gait Calculation (R) gait ∈[0,1]) R gaitIt consists of two independent risk sub-items, with a larger value indicating a more unstable user gait.

[0057] R gait =α(f step )+β(a torso (In a preferred embodiment, a weighted sum w may also be used.) a ×α+w b ×β).

[0058] Sub-item 1: Step frequency risk function α(f) step ) This quantification is based on the risks associated with abnormal step frequency (too fast or too slow).

[0059] First, the real-time step frequency f needs to be calculated from the IMU data. step The specific method is as follows: Bandpass filtering (e.g., retaining frequency components from 0.5Hz to 3Hz) is applied to the vertical acceleration signal acquired by the user's waist IMU to remove body sway and high-frequency noise. Then, a peak detection algorithm is used to identify the acceleration peak generated by each step. step This is the number of peak values ​​detected per unit of time (e.g., the past 2 seconds) divided by the time.

[0060] Set a normal step frequency range [f] min ,f max For example, [1.5Hz, 2.0Hz]. This range can be personalized by guiding the user to walk normally on flat ground for 1 minute when the system is first used.

[0061] Example of risk calculation formula: If f min ≤f step ≤f max Then α(f step )=0.

[0062] If f step <f min Then α(f step )=C1×(f min -f step ).

[0063] If f step >f max Then α(f step )=C2×(f step -f max ).

[0064] C1 and C2 are preset scaling factors used to normalize the cadence deviation to the [0,1] interval. Their values ​​can be derived by statistically analyzing a large amount of walking data to determine a maximum acceptable cadence deviation.

[0065] Sub-item 2: Torso sway risk function β(a) torso ) This quantification is based on the risk caused by violent swaying of the body from side to side.

[0066] β(a torso )=min(1,(a torso / a thresh ) 2 ); a torso This is the variance of the user's torso's acceleration in the lateral (left-right) direction, calculated by measuring the variance of lateral acceleration data from the lumbar IMU over a short time window (e.g., 2 seconds). A larger variance indicates more severe body swaying. thresh It is a preset stability threshold, which is determined by collecting data from multiple test subjects during normal walking and simulated unstable (e.g., staggering) states. torso For the data, select a value that can effectively distinguish between these two states (such as the 95th percentile of the normal walking data distribution) as the threshold. Similarly, use a squared relationship to ensure the risk is within a certain range. torso When the value is low, growth is slow (allowing for normal physiological shaking), but when it approaches or exceeds the threshold, it rapidly increases to a maximum value of 1.

[0067] Step 3: Adaptive speed control function After calculating R slip and R gait Then, the speed control module updates the robot's target speed v in real time according to the following core formula. t : v t =v base ×(1-γ1×R slip -γ2×R gait ); v t The target speed that the robot should achieve at the current moment.

[0068] v base Base speed: This is a preset value that users can set according to their personal preferences, such as "slow (0.8m / s)", "medium (1.0m / s)" or "fast (1.2m / s)" via the mobile app.

[0069] γ1, γ2: Adjustment coefficients, representing the system's sensitivity to ground risk and gait risk, respectively. These two coefficients are not fixed.

[0070] In a preferred embodiment, the present invention includes a dynamic adjustment coefficient module: The system uses lidar or depth cameras to analyze the openness of the path ahead (for example, by calculating the point cloud density of obstacles or the width of the passable path in a cone-shaped area 5 meters ahead).

[0071] When in open environments such as squares or wide sidewalks, users have ample room to react, and the system can appropriately reduce the values ​​of γ1 and γ2 (for example, by reducing them by 10%-20% based on the user's selected mode) to make speed changes smoother and improve the smoothness of movement.

[0072] When in narrow / complex environments such as indoor corridors or crowded streets, any minor risk could lead to a collision. The system will automatically increase the values ​​of γ1 and γ2 (for example, by 10%-20% on top of the user-selected mode), making the robot more sensitive and rapid in responding to risks and prioritizing absolute safety.

[0073] It also includes special case handling and multimodal feedback. Automatic emergency stop judgment: The system presets two independent hard thresholds: ground safety threshold T. slip (e.g., 0.8) and gait safety threshold T gait (e.g., 0.85). These two thresholds were obtained through statistical analysis of a large amount of real-world walking data. For example, R values ​​were collected from 100 test subjects under safe and dangerous (e.g., impending fall) conditions. gait The data is used to select a value that can distinguish between the two states with 99% accuracy as T. gait .

[0074] When R is calculated in real time slip >T slip (e.g., if the camera detects a large area of ​​ice in front) or R gait >T gait If a user stumbles violently and is about to fall, the system determines that there is a serious risk. At this time, the speed control function is bypassed, and the system immediately executes the highest priority command of "automatic deceleration and preparation to stop," while triggering a loud alarm, such as a voice prompt "Danger! Please stand firm!" and high-frequency vibration of the traction rope handle.

[0075] Multimodal feedback mechanism: In non-emergency situations, when v t Due to increased risk, it may be reduced to below a certain threshold (e.g., below v). baseWhen the risk level reaches 70% (or below 70%), the system will initiate gentle feedback. For example, it may provide a voice prompt such as "Slippery road ahead, please slow down," while simultaneously increasing the vibration of the traction rope handle. This helps the user perceive the presence and level of risk through touch, enhancing communication and trust between the user and the system. The system can also connect to wearable devices (such as smart bracelets) to provide more personalized feedback.

[0076] The system of this invention is deployed on a Unitree Go1 robot platform, equipped with a RealSense D435 depth camera and a BOSCHIMU module. The user wears an independent IMU module on their waist.

[0077] Set the base speed v base The speed is 1.0 m / s. In a typical street environment, the dynamic adjustment coefficient module is set to γ1=0.4 and γ2=0.6, which means that the system is more sensitive to changes in the user's gait stability.

[0078] Test scenario: When the user walks normally, R slip Approximately 0.1 (dry road surface), R gait Approximately 0.05 (stable gait), at which point v t =1.0×(1-0.4×0.1-0.6×0.05)≈0.93m / s, the robot is smoothly guided. When it reaches a section of road that has just been traversed by a water truck, the vision system detects the slippery surface, R slip It jumps to 0.6. At this point, v t =1.0×(1-0.4×0.6-0.6×0.05)=0.73m / s, the robot automatically and gently slows down. The user feels slightly tense due to the change in ground, and their steps become smaller and faster. R gait If it rises to 0.3, the speed will further decrease to v. t =1.0×(1-0.4×0.6-0.6×0.3)=0.58m / s. The robot automatically adjusts throughout the process without user intervention, ensuring safe passage. Tests show that this control method reduces sudden stops and pulling events by 30% compared to fixed-speed guidance, significantly improving users' subjective sense of safety and comfort.

[0079] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0080] Although this document uses various terms extensively, the possibility of using other terms is not excluded. These terms are used merely for the convenience of describing and explaining the essence of the invention; interpreting them as any additional limitation would contradict the spirit of the invention.

Claims

1. A speed adjustment method for a guide robot based on gait recognition and ground conditions, characterized in that, Includes the following steps: S1: Data Acquisition: Real-time acquisition of first type of data and second type of data through sensors deployed on the guide robot or the user. The first type of data is used to characterize the user's gait stability, and the second type of data is used to characterize the ground state in front of the guide robot. S2: Calculate the ground slip risk index R slip Based on the second type of data, calculate the ground physical characteristic risk S. friction Topographic slope risk S slope Risk of identifying specific substances S stain Then, the three risks mentioned above are weighted and fused to obtain the ground slip risk index R. slip ; S3: Calculate the gait stability risk index Rgait: Based on the first type of data, calculate the gait frequency risk α(f step ) and trunk sway risk function β(a torso Then, the step frequency risk α(f) is calculated. step ) and trunk sway risk function β(a torso The fusion yields gait stability risk indicators; S4: Adaptive speed regulation: based on ground slip risk index R slip and gait stability risk index R gait The travel speed v of the guide robot is dynamically adjusted according to the following speed control function. t :v t =v base ×(1-γ1×R slip -γ2×R gait ); Among them, v base The base speed is user-defined, and γ1 and γ2 are adjustment coefficients.

2. The method for speed adjustment of a guide robot based on gait recognition and ground conditions according to claim 1, characterized in that, Step S2 specifically involves: S21: Calculate the risk of ground physical properties S friction The texture image of the ground in front is acquired by a depth camera and input into a pre-trained convolutional neural network model to obtain the probability P of the texture being slippery. texture Simultaneously, the system obtains the current location via GPS and queries the built-in historical slip risk map to obtain a historical risk score (H). risk Take the maximum of the two as S. friction S friction =max(P texture H risk ); S22: Calculate terrain slope risk S slope The slope angle θ of the ground in front of the guide robot is obtained by measuring its posture using an inertial measurement unit. slope And according to formula S slope =min(1.0,(θ slope / θ max ) 2 ) Calculate the terrain slope risk, where θ max This is the preset upper limit of the safe slope; S23: Calculate the risk of identifying a specific substance S stain The system analyzes ground images using a visual recognition model to detect the presence of water stains, ice, or oil stains, and directly uses the model's confidence score as the S-value. stain ; S24: The above three risks are weighted and fused to obtain the ground slip risk index R. slip Its calculation formula is: R slip =w1×S friction +w2×S slope +w3×S stain , where w1, w2 and w3 are preset weight coefficients and their sum is 1.

3. The method for speed adjustment of a guide robot based on gait recognition and ground conditions according to claim 1, characterized in that, Step S3 specifically involves: S31: Calculate the step frequency risk function α(f) step The user's periodic acceleration data is collected by the inertial measurement unit and the current step frequency f is calculated by the peak detection algorithm. step Set the normal step frequency range [f] min ,f max ], when f min ≤f step ≤f max When, α(f) step )=0; when f step When not in this interval, α(f) step The value of ) increases as the deviation increases; S32: Calculate the trunk sway risk function β(a) torso The calculation method is as follows: the user's torso acceleration data in the lateral direction is collected by the inertial measurement unit, and the variance 'a' within a time window is calculated. torso According to the formula β(a) torso )=min(1,(a torso / a thresh ) 2 ) calculate, where a thresh The preset threshold for the variance of trunk lateral acceleration; S33: Combining the above two risks, we obtain the gait stability risk index R. gait Its calculation formula is: R gait =α(f step )+β(a torso ).

4. The method for speed adjustment of a guide robot based on gait recognition and ground conditions according to claim 3, characterized in that, In step S31 shown, when f step <f min When, α(f) step )=C1×(f min -f step When f step >f max When, α(f) step )=C2×(f step -f max ), where C1 and C2 are preset proportional coefficients.

5. A speed adjustment method for a guide robot based on gait recognition and ground conditions according to any one of claims 1 to 4, characterized in that, It also includes emergency stop judgment steps: Preset ground safety threshold T slip and gait safety threshold T gait When R is calculated in real time slip >T slip Or R gait >T gait When the speed is reduced to a stop, the command to slow down to a stop is executed immediately, and a high-priority voice and vibration alarm is triggered.

6. The method for speed adjustment of a guide robot based on gait recognition and ground conditions according to claim 5, characterized in that, It also includes a multimodal feedback step: When the calculated travel speed v t When the speed drops below the preset low-speed threshold, the voice prompt module and vibration module are automatically activated. The system provides feedback on potential risks to the user through voice prompts such as "Slippery ahead, please be careful" and vibration rhythms related to the level of risk.

Images