Methods and systems for leash safety control of intelligent guide dogs and intelligent guide dogs themselves

By constructing a joint feasible command domain and an emergency recovery command domain, the coordination problem of obstacle avoidance and user balance protection for intelligent guide dogs in complex environments is solved, improving safety and stability and reducing the risks caused by traction impact and elastic energy release.

CN123113995APending Publication Date: 2026-11-17CHENGDU TONGQIAN TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611609782.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing intelligent guide dog leash devices are inadequate in terms of obstacle avoidance, leash component safety, and user balance protection. They are difficult to coordinate and control in complex environments, which may lead to user imbalance or leash danger. Furthermore, they cannot independently ensure safety when sensors malfunction.

Method used

A safety controller independent of the dominant blind controller is adopted. By constructing the body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal dynamics domain, and recoverable balance domain, a joint feasible command domain is formed. Combined with the safety arbitrator to coordinate traction and body movement, an emergency recovery command domain is constructed to ensure the reliability and effectiveness of safety control.

Benefits of technology

It improves the ability of intelligent guide dogs to avoid obstacles and protect the balance of users in complex environments, reduces the risk of imbalance caused by traction impact and release of elastic stored energy, and enhances safety and stability in emergency situations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113995A_ABST
    Figure CN123113995A_ABST
Patent Text Reader

Abstract

This application relates to the field of assistive walking tools for visually impaired individuals, specifically disclosing a traction safety control method, system, and intelligent guide dog. The method includes a safety controller that determines the status of the guide dog, obstacles, traction components, and the user based on safety sensor data. It constructs a braking occupancy domain, an obstacle prediction occupancy domain, a traction spatiotemporal dynamics domain, and a recoverable balance domain. Based on obstacle avoidance distance, space occupancy, elastic energy storage, traction impulse in the direction of imbalance, and balance state, it constructs a joint feasible command domain. Candidate joint control commands are released or modified. When the joint feasible command domain is empty, an emergency recovery command domain is constructed. By adjusting the length of the traction component and the direction of traction, the actual traction force vector is brought into the set of emergency permissible traction force vectors, limiting the traction impulse in the direction of imbalance and decelerating to a stop. When recovery is not possible or the exit conditions are met, emergency braking is executed and an energy absorption component is activated, improving the safety and reliability of the guide dog traction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of assistive walking tools for visually impaired people, and in particular to a method, system and intelligent guide dog for traction safety control. Background Technology

[0002] In the field of assistive walking tools for the visually impaired, to help them walk safely, guide canes, electronic guide devices, wearable guidance devices, or intelligent guide dogs with active mobility are commonly used. Intelligent guide dogs can guide users based on environmental information and transmit guidance information such as direction, speed, and stopping to the user through a traction component connecting the intelligent guide dog and the user. They are active assistive devices used to assist visually impaired individuals in walking and protect their movement.

[0003] Unlike ordinary mobile robots, intelligent guide dogs are designed for the user, and their movement output is directly applied to the human body through a traction component. The acceleration, deceleration, turning, and posture changes of the intelligent guide dog can all alter the magnitude and direction of the traction force, thus affecting the user's center of mass movement, trunk posture, gait transitions, and support stability. Especially for visually impaired users, as well as those with reduced lower limb motor skills, insufficient balance recovery, or unstable gait, even if the intelligent guide dog itself does not collide with an obstacle, inappropriate traction force or sudden braking can still cause the user to deviate from the support area, lose balance, or even fall. Therefore, the safety of such devices depends not only on the obstacle avoidance capabilities of the intelligent guide dog itself but also on whether the traction force remains within a tolerable and recoverable range for the user.

[0004] Existing assistive devices for the visually impaired primarily guide them by detecting obstacles ahead and providing sound, vibration, or traction cues. Some guide devices with active walking capabilities can control the user's movement according to a planned path and control the traction component through fixed length, constant tension, or traction force limiting. However, these control methods typically handle environmental obstacle avoidance, user movement, and human traction protection separately.

[0005] In terms of human balance protection, existing traction assistive devices typically determine the safety of the traction state based on a preset upper limit of traction force. However, the impact of traction force on human balance is not only related to the amplitude of the traction force, but also to the traction direction, the rate of change of the traction force, the duration of action, the user's center of gravity position and velocity, trunk tilt angle, supporting foot position, and gait phase. The same traction force applied to different users, or applied to the same user during the two-legged and one-legged supporting phases, may produce different balance effects. Using only a uniform traction force threshold makes it difficult to determine whether a user can restore balance by adjusting their supporting foot or changing their gait, and it is also difficult to distinguish between traction forces that help the user restore balance and traction forces that may lead to further imbalance.

[0006] Regarding the safety of the traction assembly, to reduce traction impact and improve user comfort, elastic components, damping components, or length adjustment mechanisms can be incorporated into the traction assembly. When the relative distance between the intelligent guide dog and the user changes, the elastic component may accumulate recoverable elastic energy. When the intelligent guide dog suddenly decelerates, turns, or the traction assembly changes from a tensioned to a relaxed state, this elastic energy may be rapidly released, creating a rebound traction. Even if the instantaneous traction force does not exceed the upper limit of amplitude, under continuous action or rapid changes, it may still generate a large traction impulse in the direction of imbalance, thereby affecting the user's balance recovery.

[0007] Furthermore, the leash assembly, when relaxed, may experience gravity-induced drooping and lateral swaying. When the guide dog navigates around obstacles or changes the leash length, it may also sweep across surrounding fixed obstacles, moving obstacles, or even the human body. Existing assistive walking devices often rely on the guide dog's outline or nominal trajectory for obstacle avoidance, typically failing to include the actual sweeping area created by the leash assembly in its tensioned, relaxed, drooping, swaying, and rebounding states within the obstacle avoidance range. Therefore, it is difficult to comprehensively assess the pulling risk to the user after the leash assembly collides with an obstacle.

[0008] During the braking process of an intelligent guide dog, the dog undergoes a certain period of perception, calculation, execution, and mechanical braking from receiving the braking command to coming to a complete stop. The actual braking distance is also affected by factors such as travel speed, ground adhesion coefficient, road slope, foot contact state, and the dog's posture. If the decision to stop is based solely on the current obstacle distance or the nominal braking distance, it may not accurately reflect the space occupied by the intelligent guide dog under different braking conditions until it comes to a complete stop, nor can it simultaneously assess the impact of the braking action on the user's balance and the energy release process of the traction components.

[0009] When conventional obstacle avoidance conditions cannot be met, existing intelligent guide dogs typically respond by emergency braking, immediately tightening the traction assembly, or releasing the traction assembly. However, direct emergency braking can cause abrupt changes in traction force and body deceleration, and a concentrated release of the elastic energy stored in the traction assembly; immediately tightening the traction assembly can generate a large traction impulse on the user; and directly releasing the traction assembly can interrupt guidance or cause significant swaying of the traction assembly. These methods struggle to coordinate the prevention of collisions with maintaining the user's balance, and lack a control mechanism for sequentially adjusting traction energy, phased deceleration, support posture adjustment, and stable parking based on the remaining time available.

[0010] On the other hand, as a human assistive device that directly guides the user, intelligent guide dogs place high demands on the reliability of their sensors and controllers for safety control. If environmental detection, the dog's own state detection, traction force detection, or user state detection are interrupted, delayed, or experience data anomalies, erroneous movement and traction commands may directly affect the user. In existing devices, navigation control and safety control typically share controllers, sensors, or data channels. When the main controller or shared sensing link fails, it may be unable to independently reject dangerous commands, and there is a lack of alternative estimations and degraded parking measures for different sensor failure states.

[0011] Therefore, there is a need to provide a traction safety control method and system for intelligent guide dogs, with the main objectives of assisting the user's walking and protecting their balance. This system comprehensively evaluates the braking and obstacle avoidance capabilities of the intelligent guide dog itself, the sweeping and elastic energy storage state of the traction components, and the impact of traction on the user's balance. When conventional safety conditions cannot be met or relevant sensors malfunction, the system coordinates and controls the traction energy, body deceleration, foot support, and parking process to improve the safety of visually impaired people using intelligent guide dogs. Summary of the Invention

[0012] This application overcomes the shortcomings of the prior art and provides a method, system and intelligent guide dog for safe traction control, which can improve the safety of traction.

[0013] Firstly, this application provides a leash safety control method for intelligent guide dogs, applied to intelligent guide dogs, comprising: Acquire safety sensor data from the intelligent guide dog; Based on the acquired safety sensor data, the body state, braking capacity, environmental obstacle state, traction component state, user balance state, and direction of imbalance are determined, and the corresponding body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal mechanics domain, and recoverable balance domain are constructed. Based on the body braking occupancy domain, the obstacle prediction occupancy domain, the recoverable balance domain, the traction spatiotemporal dynamics domain, and the preset conditions, the set of body motion control quantities and traction control quantities that satisfy the preset conditions is determined as the joint feasible command domain. Based on the candidate body movement commands generated by the dominant blind controller of the intelligent guide dog, the state of the traction component, and the balance state of the user, candidate traction control quantities are generated, and the candidate body movement commands and candidate traction control quantities are combined to form candidate joint control commands. When the candidate joint control instruction is located within the joint feasible instruction domain, the intelligent guide dog is controlled to execute the candidate joint control instruction. When the candidate joint control instruction is located outside the joint feasible instruction domain and the joint feasible instruction domain is not empty, the intelligent guide dog is controlled to execute the modified joint control instruction selected from the joint feasible instruction domain. When the joint feasible command domain is empty, an emergency recovery command domain is constructed based on the preset emergency obstacle avoidance distance limit, the preset emergency balance margin limit, the space occupancy constraint in the traction spatiotemporal mechanics domain, the energy storage absorption threshold, the traction impulse limit in the imbalance direction, and the braking capacity constraint. When the emergency recovery command field is not empty and the preset recovery start conditions are met, an emergency recovery joint control command is determined and executed from the emergency recovery command field. This command controls the intelligent guide dog to adjust the length of the traction component and the direction of traction, so that the actual traction force vector enters the set of emergency permissible traction force vectors determined according to the user's balance state and the user's individual traction safety limit. It also limits the traction impulse in the direction of imbalance and controls the intelligent guide dog to decelerate and stop. During the execution of the emergency recovery joint control command, the energy absorption component is controlled according to the comparison result between the recoverable elastic energy storage and the energy absorption threshold. When the emergency recovery command field is empty, the preset recovery start condition is not met, or the preset exit condition is met during the execution of the emergency recovery joint control command, the intelligent guide dog is controlled to perform emergency braking and the energy absorption component is activated.

[0014] Secondly, this application also provides an intelligent guide dog leash safety control system, including a main body, an adjustable leash assembly connecting the main body and the user, an environmental sensing assembly, a dominant blind controller, an independent safety sensor, a safety controller, a safety arbitrator, and an actuator. The environmental perception component is connected to the dominant blind controller, which is used to generate candidate body motion commands based on environmental perception data and navigation information. The independent safety sensor includes an environmental obstacle detection unit, a body status detection unit, a traction component status detection unit, and a user status detection unit, and is connected to the safety controller through a data channel independent of the dominant blind controller; The safety controller is used to determine the body state and braking capability, environmental obstacle state, traction component state and user balance state based on the safety sensor data output by the independent safety sensors, determine the direction of imbalance, and construct the body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal dynamics domain and recoverable balance domain; the safety controller is also used to determine the set of body motion control quantities and traction control quantities that satisfy the preset conditions as a joint feasible command domain based on the body braking occupancy domain, the obstacle prediction occupancy domain, the recoverable balance domain, the traction spatiotemporal dynamics domain and preset conditions. The safety controller is further configured to: output a safety verification signal to the safety arbitrator when the joint feasible instruction domain is not empty and the candidate joint control instruction is located within the joint feasible instruction domain; output a modified joint control instruction determined by the joint feasible instruction domain when the joint feasible instruction domain is not empty but the candidate joint control instruction is located outside the joint feasible instruction domain; construct an emergency recovery instruction domain when the joint feasible instruction domain is empty; output an emergency recovery joint control instruction when the emergency recovery instruction domain is not empty and a preset recovery start condition is met; and output an emergency braking instruction and an energy absorption instruction when the emergency recovery instruction domain is empty, the preset recovery start condition is not met, or a preset exit condition is met during the execution of the emergency recovery joint control instruction. The actuator includes a body motion actuator, a traction actuator, and an energy absorption actuator, and receives control commands only through the safety arbitrator; The safety arbitrator is used to: preferentially transmit the corresponding control command to the actuator when it receives a modified joint control command, an emergency recovery joint control command, or an emergency braking command output by the safety controller; transmit the corresponding candidate joint control command to the actuator when it only receives the safety verification signal; and transmit the default emergency braking command and default energy absorption command pre-stored in the safety arbitrator to the actuator when it neither receives the control command output by the safety controller nor the safety verification signal within a preset arbitration period.

[0015] Thirdly, this application also provides an intelligent guide dog, including the intelligent guide dog traction safety control system described in the above embodiments; or, the intelligent guide dog performs the intelligent guide dog traction safety control method described in the above embodiments.

[0016] The beneficial effects of the intelligent guide dog leash safety control method and system provided in this application include: (1) This application, through a safety controller independent of the dominant blind controller, transforms the body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal dynamics domain, and user recoverable balance domain into safety constraints and takes their intersection to form a joint feasible command domain that simultaneously limits the body motion control quantity and traction control quantity. This unifies the intelligent guide dog obstacle avoidance, traction component safety, and user balance protection into the same control framework. For candidate commands that do not meet the joint safety constraints, hierarchical correction can be performed within the feasible domain, and the safety arbitrator can prevent unverified commands from directly acting on the actuator, thus avoiding user imbalance or traction danger caused by each control link individually meeting local conditions but being jointly executed.

[0017] (2) This application predicts the complete pose sequence of the intelligent guide dog from switching to a braking strategy limited by braking capacity until it comes to a stop at different braking switching times in the prediction time domain. It also constructs the body braking occupancy domain by combining the ground adhesion coefficient, road slope, foot contact state, total perception-execution delay, and state error boundary. Compared with obstacle avoidance judgment based solely on the current distance, nominal trajectory, or a single braking distance, this method can more accurately characterize the braking occupancy range of the intelligent guide dog under actual working conditions, and improve the reliability of braking obstacle avoidance judgment in complex road surface and dynamic obstacle environments.

[0018] (3) This application constructs a recoverable balance domain based on the user's center of mass, trunk posture, supporting foot position and gait phase, and uses the position of the capture point relative to the target support area to determine the balance margin, the direction of imbalance and the recovery direction, so that the traction safety judgment is expanded from a simple limitation of traction force amplitude to a comprehensive evaluation of traction direction, traction impulse, gait recovery ability and allowable adjustment steps; at the same time, the recoverable balance domain is adjusted in combination with the traction safety limit obtained by individual user calibration, which can implement adaptive balance protection for different users and different gait phases, and reduce the risk of users falling due to improper traction direction or traction time during assisted walking.

[0019] (4) This application constructs a traction spatiotemporal mechanical domain that simultaneously includes the predicted occupancy domain of the traction component, recoverable elastic energy storage, and directional traction impulse. Under tension, it considers the dynamic sweep range of the traction connection segment; under relaxation, it further considers the uncertain regions formed by traction length margin, gravity sag, and lateral sway. Based on the energy storage margin, impulse margin, and traction force change rate margin, it dynamically expands the sweep range, tightens traction adjustment, and limits the body acceleration. Therefore, it can simultaneously suppress the safety risks caused by the traction component colliding with obstacles, lateral swinging, elastic rebound, and continuous traction, improving the stability of the traction guidance process and the safety of human use.

[0020] (5) This application further constructs an emergency recovery command domain when the conventional joint feasible command domain is empty, and compares the predicted remaining collision time with the safe handling time determined by the braking time, traction energy release time, and total perception-execution delay; when recovery conditions are met, traction energy shaping, first-stage deceleration, second-stage limited-jump deceleration, foot support adjustment, and stable parking are executed sequentially; when recovery conditions deteriorate, emergency braking, energy absorption, and mechanical limited-range release are switched in a timely manner. This can coordinate the contradiction between collision avoidance by intelligent guide dogs and balance protection for users, reduce the traction impact and concentrated release of elastic stored energy caused by direct emergency braking, and increase the possibility of users maintaining or restoring balance in emergency situations. Attached Figure Description

[0021] Figure 1This is a structural schematic diagram of the intelligent guide dog provided in this application; Figure 2 This is a flowchart of the data acquisition and candidate instruction generation process for the intelligent guide dog traction safety control method provided in this application; Figure 3 This is a flowchart illustrating the standard feasibility assessment process for the traction safety control method for intelligent guide dogs provided in this application; Figure 4 This is an emergency recovery initiation and execution flowchart of the intelligent guide dog traction safety control method provided in this application; Figure 5 This is a flowchart of the exit judgment and safe parking process of the intelligent guide dog traction safety control method provided in this application. Detailed Implementation

[0022] The present application will be further described below with reference to the accompanying drawings. The following embodiments are used to illustrate the technical solutions of the present application and are not intended to limit the scope of protection of the present application. Where there is no conflict, the technical features of the various embodiments can be combined with each other.

[0023] This embodiment provides a method for safe control of the traction of intelligent guide dogs, which is applied to the safety control of intelligent guide dogs.

[0024] Please refer to Figure 1 In this embodiment, the intelligent guide dog can be a quadrupedal walking robot, a wheeled-legged robot with an adjustable posture support mechanism, or even a wheeled guide robot.

[0025] In this embodiment, a quadrupedal walking robot-type intelligent guide dog is used as an example. It includes a main body, a foot support mechanism, a body posture adjustment mechanism, an adjustable traction component, an environmental perception component, a dominant blind controller, independent safety sensors, a safety controller, a safety arbitrator, and actuators. The foot support mechanism is connected to the main body and driven by the main body's motion actuators to enable the intelligent guide dog to walk, turn, decelerate, and stop. The body posture adjustment mechanism is used to adjust the main body's posture and center of gravity height. The adjustable traction component is positioned between the main body and the user. The environmental perception component is connected to the dominant blind controller to provide environmental perception data. Independent safety sensors are respectively located on the main body, the adjustable traction component, and the user's side, and are connected to the safety controller via independent data channels. The dominant blind controller and the safety controller are respectively connected to the safety arbitrator, which is connected to each actuator to transmit safety-verified or modified control commands to the actuators.

[0026] In one specific structure, the adjustable traction assembly includes a traction length adjustment mechanism, a flexible traction element, a series elastic component, a damping component, an energy absorption component, a direction adjustment anchor point, a traction force sensor, and a traction length encoder. The traction length adjustment mechanism is located on the intelligent guide dog body and includes a servo motor, a reducer, a winding wheel, a leash guide mechanism, and a controlled locking mechanism. The servo motor is connected to the winding wheel via the reducer. One end of the flexible traction element is wound around the winding wheel, and the other end is connected to the user-side connection point via the leash guide mechanism and the direction adjustment anchor point. The traction length encoder is located on the winding wheel and is used to detect the traction length and its rate of change.

[0027] The series elastic component uses a helical spring connected in series in the traction force transmission path of the flexible traction component. The traction force sensor is connected in series between the series elastic component and the user-side connection point. The damping component uses a viscous damper, and the energy absorption component uses a controllable friction energy absorption mechanism. Both are connected between the two relatively moving ends of the series elastic component to suppress normal traction vibration and absorb the elastic energy released by the series elastic component in emergency situations. The direction adjustment anchor point includes a linear moving slide fixed to the main body and a guide wheel set on the slide. The anchor point drive unit is connected to the linear moving slide, and the position of the leash and the direction of traction on the intelligent guide dog side are adjusted by changing the position of the guide wheel.

[0028] The independent safety sensors include an environmental obstacle detection unit, a body state detection unit, a traction component state detection unit, and a user state detection unit. The environmental obstacle detection unit includes a safety ranging sensor mounted on the body and facing the environment, used to acquire obstacle distance, point cloud data, or target detection data. The body state detection unit includes a safety inertial sensor fixed to the body and state detection units located at the joints and feet, used to acquire body acceleration, angular velocity, attitude, joint state, foot contact state, and foot support force.

[0029] The traction component status detection unit includes a traction force sensor positioned in the traction force transmission path, a traction length encoder positioned on the winding wheel, an anchor point position detection unit positioned at the direction adjustment anchor point, and a deformation detection unit positioned on the series elastic component, used to obtain traction force, traction length, anchor point position, and elastic deformation. The user status detection unit includes wearable inertial sensors positioned on the user's torso, waist, lower limbs, or feet, and a vision sensor facing the user, used to obtain the user's posture, movement state, key body point positions, and human-machine relative position for each body part. All sensor data is timestamped and transmitted to the safety controller via a data channel independent of the master blind controller.

[0030] The braking capacity parameters, motor torque constant, transmission ratio, winding wheel radius, and sensor error parameters of the main motion actuator are obtained through factory calibration or system initialization calibration; the restoring force-deformation relationship of the series elastic component and the damping parameters of the damping component are obtained through pre-calibration; the individual user traction safety limit is obtained through calibration using individual user parameters. These parameters are stored in a parameter storage unit that the safety controller can read.

[0031] To facilitate unified calculations, a ground-fixed coordinate system, a body coordinate system, and a user coordinate system are established. The states of environmental obstacles, the body, the connection points of traction components, the user, and the traction force vector are all transformed to the ground-fixed coordinate system or a unified calculation coordinate system with a known transformation relationship to the ground-fixed coordinate system.

[0032] The safety controller operates according to a preset safety control cycle. The start time of the current control cycle is denoted as t0, and the predicted time domain length is denoted as T. p The prediction step size is denoted as Δt, and each discrete prediction time is represented as follows: Where, NΔt=T p The motion of the body and the motion of the traction components between adjacent discrete prediction times are represented by a continuous swept envelope, so that the discrete prediction results cover the continuous motion between adjacent prediction times.

[0033] The joint control quantity to be evaluated is expressed as: Among them, u B These are the motion control variables for the body, including one or more of the following: target velocity, acceleration, steering angular velocity, braking deceleration, rate of change of deceleration, gait parameters, and body attitude adjustment variables; u T The traction control quantities include one or more of the following: traction target length, traction length change rate, position or angle of direction adjustment anchor point, anchor point adjustment speed, and energy absorption control quantities.

[0034] When evaluating candidate joint control commands, the body motion control quantity u B The candidate body motion commands originate from the dominant blind controller, generated based on navigation information and environmental perception data. Target velocity, acceleration, steering angular velocity, gait parameters, and body attitude adjustment are determined by path tracking and motion control results. Braking deceleration and deceleration rate are determined based on the current body state and braking capacity. Traction control quantity u TThe safety controller generates the target length and its rate of change based on the candidate body motion command, the traction component status and the user's balance status. The target length and its rate of change are determined based on the relative speed between the human and the machine, the deviation between the target traction force and the actual traction force, and the energy storage margin. The position or angle of the direction adjustment anchor point and its adjustment speed are determined based on the deviation between the target traction direction and the current traction direction. The energy absorption control quantity is determined based on the comparison between recoverable elastic energy storage and the energy absorption threshold.

[0035] When constructing a joint feasible command domain or an emergency recovery command domain, the safety controller uses the aforementioned control variables as optimization variables and generates the joint control variables to be evaluated within the allowable range of the body motion actuator, traction actuator, and energy absorption actuator.

[0036] The joint control quantity to be evaluated can be a control vector within a control cycle or a control sequence covering the prediction time domain. When using a control sequence, the safety controller solves the problem using a rolling time domain approach, sending only the first control quantity corresponding to the current control cycle to the safety arbitrator, and re-acquiring safety sensor data and updating the prediction results in the next control cycle.

[0037] In this embodiment, the preset obstacle avoidance distance limit is denoted as d. run The preset emergency obstacle avoidance distance limit is denoted as d. emg The two satisfy: d run ≥d emg ≥0; the energy storage absorption threshold is denoted as E abs The energy storage conversion threshold is denoted as E. conv Both satisfy: 0 ≤ E conv <E abs .

[0038] The operational balance margin limit is a preset non-negative value used to determine the balance state that the user should achieve when restoring balance under normal operating conditions; the emergency balance margin limit is used to construct the emergency recovery command domain; the lower limit of the balance margin is determined based on the individual user parameters and is used to trigger the tightening of the emergency traction limit; the parking balance margin threshold is used to determine whether the user has reached the state that allows for safe parking.

[0039] In the optimization calculations using ratios for normalization described later, the energy storage absorption threshold, traction impulse limit, traction force change rate limit, obstacle avoidance distance limit, and attitude stability margin reference value used as the denominator are positive values. When the corresponding physical threshold is allowed to be zero, a positive normalization reference value corresponding to the physical quantity is set to avoid the division by zero problem in the normalization calculation.

[0040] Within each control cycle, the safety controller sequentially performs the following steps: acquiring safety sensor data, determining status, constructing four types of safety domains, constructing joint feasible command domains, verifying candidate joint control commands, and correcting control or handling emergencies.

[0041] like Figure 2 As shown, safety sensor data acquisition and status determination: The safety controller acquires safety sensor data independent of the processing results of the dominant blind controller. This safety sensor data includes environmental obstacle detection data, vehicle status detection data, traction component status detection data, and user status detection data.

[0042] Each safety sensor outputs data with a timestamp. The safety controller aligns the data from each safety sensor to the start time of the current control cycle. For data sampled earlier than the start time of the current control cycle, a short-time extrapolation is performed based on the rate of change of the corresponding state variable, and the extrapolation error is included in the corresponding state uncertainty region. For multi-source data with different sampling frequencies, time alignment is achieved through interpolation or state filtering.

[0043] Environmental obstacle detection data can be obtained from one or more of the following: a safety ranging sensor, lidar, millimeter-wave radar, ultrasonic sensor, or depth vision sensor. The safety controller uses this data to determine the position, velocity, direction of motion, geometric dimensions, and perception error limits of the environmental obstacles.

[0044] The vehicle's status detection data may include data from safety inertial sensors, joint encoders, foot contact sensors, plantar force sensors, and vehicle odometers. The safety controller uses this data to determine the vehicle's position, attitude, speed, acceleration, pitch angle, roll angle, foot contact status, and current braking capability.

[0045] The braking capacity includes at least the maximum available braking deceleration, the range of deceleration rate of change, and the braking strategy permitted under the current foot contact state. The safety controller determines the current braking capacity based on the ground adhesion coefficient, road slope, number of foot contacts, foot support reaction force margin, body posture, upper limit of driving mechanism braking force, and total perception-execution delay.

[0046] When the ground adhesion coefficient decreases, the number of foot contacts decreases, the unfavorable downhill component along the current direction of travel increases, the foot support reaction margin decreases, or the body roll angle approaches the corresponding limit, the maximum available braking deceleration is reduced. For travel uphill, the auxiliary deceleration effect generated by gravity along the slope component can be included in the determination of the current braking capacity. The traction component status includes the position and velocity of the intelligent guide dog side connection point and the user side connection point, the traction length and its rate of change, the traction force vector and its rate of change, the tensioned or slack state, and the elastic and damping parameters.

[0047] The safety controller compares the traction length with the geometric distance between the two connection points and uses the traction force sensor output to determine the tension state. When the difference between the traction length and the distance between the connection points is less than the tension margin threshold and the traction force is not lower than the tension judgment threshold, the traction component is determined to be in a tensioned state. When the traction length has a length margin greater than the slack margin threshold or the traction force is lower than the slack judgment threshold, the traction component is determined to be in a slack state. The tension and slack judgment thresholds can be configured with hysteresis to avoid frequent state switching.

[0048] User status detection data is obtained from one or more of wearable inertial sensors, vision sensors, traction sensors, and connection point position detection devices. The safety controller uses this data to determine the user's center of gravity position and velocity, torso tilt angle and rate of change of tilt angle, supporting foot position, swing foot position, gait phase, relative position between the user and the intelligent guide dog, and the current traction vector.

[0049] User's balance state, recoverable balance range, and direction of imbalance: The safety controller determines the target support area based on the user's gait phase. When the user is in a single support phase, the contact area between the current supporting foot and the ground is determined as the current support area; in a double support phase, the support polygon formed by the contact areas of the two supporting feet is determined as the current support area; when the swinging foot is about to land, a predicted support area is determined based on the swinging foot's position, swing speed, reachable landing point, and the ground's contact area. The safety controller selects the area from the current support area and the predicted support area that matches the current gait phase as the target support area.

[0050] In one implementation, an equivalent inverted pendulum model is used to calculate the user capture point: , where p com v is the projection of the user's center of mass onto the supporting plane. com The velocity ω is the projected velocity of the center of mass on the supporting plane, determined by gravitational acceleration and the user's equivalent center of mass height. For slopes or situations where the center of mass height changes significantly, the projected center of mass and equivalent center of mass height can be corrected based on the slope normal, or the equivalent capture point can be determined using a center of mass momentum model or a human multi-rigid-body dynamics model.

[0051] Let the target support area be S. H The directed distance from the capture point to the boundary of the target support area is: Wherein, when the capture point is located inside the target support area, μ>0; when it is located on the boundary, μ=0; and when it is located outside the target support area, μ<0. μ represents the user's balance margin.

[0052] The safety controller identifies the boundary point on the target support area boundary that is closest to the capture point, and determines the outward normal direction at that boundary point as the imbalance direction n. loss The opposite direction is determined as the recovery direction: n rec =−n loss When there are two or more equidistant nearest boundary points, the direction of the outer normal that maximizes the projected outward velocity of the capture point is selected as the direction causing imbalance; if no single direction can be selected, the directional envelope formed by each candidate outer normal is evaluated as the constraint for the direction causing imbalance.

[0053] To determine the recoverable equilibrium region, further define the target recovery region within the target support region that meets the operational equilibrium margin limit: .in: This is the operating balance margin limit, and it is a preset non-negative value.

[0054] The safety controller searches for traction control sequences and support foot adjustment sequences based on the user's state to be determined. The recoverable balance domain is a set of user states that meet the following conditions: there exists a traction control sequence that satisfies the user's individual traction safety limits, center of mass dynamic constraints, and the reachability range of the support foot landing point, ensuring that the user's torso tilt angle and tilt angle change rate do not exceed the corresponding limits, and that the capture point enters the target support area within the allowed number of adjustment steps with a balance margin not less than the operating balance margin limit.

[0055] Before the capture point enters the target support area, the balance margin is allowed to be negative. However, in this embodiment, the balance margin must not fall below the recoverable lower bound determined based on the individual user calibration results, and it must not continuously decrease within a preset duration. After the capture point enters the target support area, it should further move to the target recovery area S. run .

[0056] The individual traction safety limits for users include at least the permissible traction force vector range, the traction force change rate limit, the traction impulse limit in the direction of imbalance, the traction impulse limit in the direction of recovery, the torso tilt angle and its change rate limit, the capture point recovery time, and the permissible number of adjustment steps.

[0057] The set of emergency permissible traction force vectors can be represented as: Among them, F rec,min and F rec,max These are the lower and upper limits of the traction force component in the recovery direction, respectively, F loss,max The upper limit of the traction force component in the direction of imbalance. J is the limit for the rate of change of emergency traction force. loss,emg The limit for traction impulse in the direction of emergency imbalance.

[0058] When the user's balance margin is less than the lower limit of the balance margin, the upper limit of the traction force component in the direction of imbalance, the limit of the rate of change of traction force, and the limit of the traction impulse in the direction of imbalance shall be adopted as emergency limits that are less than the corresponding preset operating limits.

[0059] Body braking occupancy region and obstacle prediction occupancy region: The safety controller constructs a braking occupancy domain based on the current system state, braking capability, and the joint control commands to be evaluated. This braking occupancy domain is composed of the available braking spacetime tubes within the prediction time domain.

[0060] Determine the set of braking switching times within the prediction time domain: ; For each braking switching moment, first predict the pose of the intelligent guide dog from the execution of the joint control command to be evaluated to the braking switching moment, and then predict the pose sequence from the switch to the braking strategy limited by the current braking capability until parking.

[0061] Let the ontological geometric model be G. B The m-th braking branch at the predicted time t k The pose transformation is T(q) k The region of uncertainty is ΔBk,m, and the corresponding spatial section is: .

[0062] Here, "⊕" represents the Minkowski summation.

[0063] The state uncertainty region expands along the direction of travel based on speed error, braking capability error, and total perception-execution delay; along the lateral slip direction based on lateral speed error, ground adhesion error, and foot slip error; and along the spatial offset direction caused by pitch and roll errors based on attitude error limits and body geometry.

[0064] Predicting time t k The cross section occupied by the main braking system is the union of the cross sections of each braking branch: The cross-sections and time parameters corresponding to each predicted time point together constitute a brakeable spacetime tube.

[0065] To cover the continuous situation between discrete braking switching moments, the safety controller continuously sweeps and expands the corresponding spatial occupancy section based on the maximum linear displacement, maximum rotation angle, and maximum attitude change that the body may achieve between adjacent braking switching moments. Alternatively, an interval dynamics model can be used to directly obtain the pose reachable envelope between adjacent braking switching moments.

[0066] The joint control command to be evaluated can be a candidate joint control command, or a joint control command to be optimized generated during the hierarchical optimization process, or an emergency joint control command to be evaluated in the emergency recovery command domain.

[0067] For the j-th environmental obstacle, the safety controller determines its position at prediction time t based on its location, velocity, acceleration range, geometric model, and perception error bound. k Occupied area O j,k All obstacles at the predicted time t k The occupied section is represented as: .

[0068] For dynamic obstacles, predictions can be made for multiple possible motion patterns, and the union of the predicted regions corresponding to each motion pattern can be calculated. For environmental areas where passage is not confirmed, they are incorporated into the obstacle prediction occupancy domain under a safety degradation state.

[0069] Define the minimum distance between spatial sets as: ; When all prediction time points satisfy: ; And: In When the time comes, it is determined that the joint control command to be evaluated meets the braking and obstacle avoidance constraints under the operating state.

[0070] Traction spacetime mechanical domain: The traction spatiotemporal mechanics domain includes the traction component's predicted occupancy domain, as well as the recoverable elastic energy storage, the traction impulse in the imbalance direction, and the traction impulse in the recovery direction associated with each prediction moment.

[0071] The status of the traction components includes the position and speed of the intelligent guide dog side connection point and the user side connection point, the traction length and its rate of change, the traction force vector and its rate of change, the tension or slack state, and the elastic and damping parameters.

[0072] In the tensioned state, the safety controller determines the connection segment between the intelligent guide dog-side connection point and the user-side connection point based on the positions of the connection points at each predicted time. It then expands this connection segment outwards using the actual cross-sectional size of the traction component and its position estimation error, forming the spatial cross-section of the traction component for the corresponding predicted time. For two adjacent predicted times, the area swept by the continuous movement of the connection segment and the traction component cross-section between these two times is defined as the predicted occupancy domain of the traction component within the corresponding time interval. This avoids missing sweeping collisions between adjacent times by only performing collision checks on discrete times.

[0073] In the relaxed state, the safety controller determines the traction length margin based on the difference between the traction length and the geometric distance between the two connection points. It then determines the gravitational sag and lateral oscillation amplitude of the traction component based on the unit length mass of the traction component, the height difference between the two connection points, gravity, the relative speed between the human and machine, and the body's steering motion. The uncertain region of sag and oscillation formed by the traction length margin, gravitational sag, and lateral oscillation amplitude is merged with the predicted sweep region of the traction path between the two connection points to obtain the predicted occupancy region of the traction component in the relaxed state.

[0074] The amount of gravity-induced sag can be calculated using a catenary model, or a sag lookup table calibrated through experiments can be established based on the traction length margin and the distance between the two connection points. The lateral oscillation amplitude can be predicted based on the body's steering angular velocity, the lateral relative velocity between the human and machine, the length of the traction component, and historical oscillation states. When the sag or oscillation direction cannot be accurately determined, a conservative spatial envelope covering all possible sag and oscillation positions is used.

[0075] The recoverable elastic energy storage is determined by integrating the elastic deformation along the elastic deformation path of the series elastic components using the elastic restoring force. For linear elastic components, it can be determined based on the elastic stiffness and deformation; for nonlinear elastic components, it can be determined based on a pre-calibrated restoring force-deformation relationship curve or a lookup table. The energy consumed by the damping components during movement, as well as the mechanical work output or absorbed by the traction actuator in actively retracting or extending the traction assembly, are not included in the recoverable elastic energy storage.

[0076] The safety controller determines the corresponding recoverable elastic energy storage at each prediction time and determines the maximum value in the prediction time domain as the predicted recoverable elastic energy storage.

[0077] The traction impulse in the direction of imbalance is determined by accumulating the orthogonal projection of the traction force vector in the direction of imbalance over time within the prediction time domain. When the projection of the traction force in the direction of imbalance is negative, this portion of the traction force is not included in the traction impulse in the direction of imbalance. The traction impulse in the direction of recovery is determined by accumulating the orthogonal projection of the traction force vector in the direction of recovery over time. Each prediction time corresponds to the accumulated traction impulse in the direction of imbalance and the traction impulse in the direction of recovery from the current time to that prediction time. The accumulated traction impulse in the direction of imbalance at the end of the prediction time domain is taken as the predicted traction impulse in the direction of imbalance.

[0078] The safety controller determines the traction force change rate based on the traction force vector change and the corresponding time interval between adjacent prediction times, and determines the maximum value of the traction force change rate amplitude in the prediction time domain as the predicted traction force change rate amplitude.

[0079] Energy storage margin is the difference between the energy storage absorption threshold and the predicted recoverable elastic energy storage; impulse margin is the difference between the limit of traction impulse in the direction of imbalance and the predicted traction impulse in the direction of imbalance; rate of change margin is the difference between the limit of traction rate of change and the predicted amplitude of traction rate of change.

[0080] The predicted rebound direction is the initial motion direction of the traction component or connection point predicted based on the direction of the elastic restoring force, the mass and motion state of the two connection points, and the human-machine interface constraints, assuming that the traction actuator stops its active output and allows the series elastic components to release. When multiple possible rebound directions exist, the directional envelope formed by each possible rebound direction is used as the predicted rebound direction region.

[0081] When the ratio of any of the energy storage margin, impulse margin, or rate of change margin to its corresponding limit is less than a preset margin threshold, it indicates that the traction component is approaching the corresponding mechanical safety boundary. The safety controller expands the predicted occupancy domain of the traction component along the predicted rebound direction and reduces the upper limit of the traction length adjustment speed, the limit of the traction force rate of change, and the upper limit of the intelligent guide dog acceleration. The smaller the margin, the greater the expansion of the occupancy domain and the greater the reduction of the corresponding control quantity.

[0082] When the predicted lateral sway amplitude exceeds the lateral sway amplitude threshold, the predicted occupancy domain of the traction component is expanded along the lateral sway direction. Thus, the traction spatiotemporal mechanics domain can simultaneously characterize the spatial sweep risk, elastic rebound risk, and directional traction impulse risk of the traction component.

[0083] A set of traction force vectors can be realized: The safety controller determines the actual achievable tension vector set of the traction component by adjusting the position or angle of the anchor point, the relative position of the intelligent guide dog and the user, the position of the connection points on both sides, and the tension state of the traction component.

[0084] The traction assembly can only apply a non-negative traction force in the direction of tension acting on the user, and cannot apply a thrust force to the user in the opposite direction. For each permissible position or angle of the direction adjustment anchor point, the safety controller determines the corresponding tension direction based on the direction from the user-side connection point to the intelligent guide dog-side traction outlet, and determines the tension amplitude range in that direction based on the output capacity of the traction actuator, the permissible deformation range of the series elastic components, the permissible tension range of the traction assembly, and the individual traction safety limit of the user.

[0085] The tension vectors corresponding to all permissible positions or angles of the directional adjustment anchor points together constitute the set of achievable tension vectors. When the traction component is in a slack state, before the traction length margin is eliminated, the current set of achievable tension vectors may only include zero tension; when the rope can be pulled back and the slack eliminated within the prediction time domain, the achievable non-zero tension vectors are determined based on the remaining prediction time after tensioning is completed.

[0086] The safety controller determines the set of permissible traction force vectors based on the recoverable equilibrium domain, the direction of imbalance, the recovery direction, and the individual traction safety limit of the user, and then intersects the set of permissible traction force vectors with the set of achievable tension vectors.

[0087] From the intersection, select the traction force vector that can maintain or improve the user's balance margin as the target traction force vector. Priority can be given to traction force vectors that maximize the user's minimum balance margin in the prediction time domain, minimize the capture point recovery time, or minimize the deviation of the target traction force from the normal guiding traction force.

[0088] When the intersection is empty, a target traction force vector that is physically impossible to achieve with the traction component is not generated. Instead, the action is taken by adjusting the direction of the anchor point, lengthening the traction component, reducing the speed of the intelligent guide dog, or switching to an emergency braking state.

[0089] Joint feasible instruction domain: The safety controller determines the physically executable body motion control quantities and traction control quantities based on the kinematic and dynamic constraints of the body, the foot contact friction cone constraints, the joint position and torque constraints, the capabilities of the body motion actuator, the capabilities of the traction actuator, the energy absorption capacity, and the actuariability constraints of the adjustable traction component.

[0090] For each joint control command to be evaluated, predict the corresponding body braking occupancy domain, obstacle prediction occupancy domain, traction component prediction occupancy domain, recoverable elastic energy storage, traction impulse in the direction of imbalance, and user status.

[0091] The set of body motion control variables and traction control variables that meet the following conditions constitutes a joint feasible command domain: The cross sections of the body braking occupancy domain and the obstacle prediction occupancy domain do not intersect at each prediction time, and the minimum distance between each corresponding cross section is not less than the preset obstacle avoidance distance limit. The predicted occupancy domain of the traction component and the predicted occupancy domain of the obstacle do not intersect at the corresponding cross sections at each prediction time; The predicted recoverable resilient energy storage will not exceed the energy storage absorption threshold; The predicted traction impulse in the direction of imbalance shall not exceed the limit value of the traction impulse in the direction of imbalance. Each predicted state of the user belongs to a recoverable equilibrium domain that matches the corresponding gait phase and target support area; Both the motion control quantity and the traction control quantity of the body are within the allowable range of the corresponding actuator, and the corresponding traction force vector belongs to the set of realizable tension vectors.

[0092] The joint feasible command domain can be obtained through constrained model predictive control, sequential quadratic programming, interior point optimization, or sampling search. In one embodiment, the feasible control sequence of the previous control cycle is used as the initial solution for the current control cycle, and constrained optimization is performed on the control quantities near the candidate joint control commands. For the obstacle prediction occupancy domain and the traction component prediction occupancy domain, continuous collision detection or a conservative convex hull method is used to determine whether they intersect.

[0093] like Figure 3 As shown, candidate joint control commands and normal control: The primary blind controller generates candidate body motion commands based on environmental perception data and navigation information. The safety controller predicts the movement of the intelligent guide dog's side connection point based on the candidate body motion commands, and generates candidate traction control quantities that match the candidate body motion commands based on the traction component status, the user's balance status, and the relative motion between the human and the machine.

[0094] Candidate body motion commands and candidate traction control quantities have the same control cycle number and together form candidate joint control commands.

[0095] When a candidate joint control command is within the joint feasible command domain, the safety controller outputs a safety verification signal corresponding to that candidate joint control command. The safety arbitrator allows the execution of the first control input of the candidate joint control command, and then proceeds to the next control cycle and reacquires safety sensor data.

[0096] When a candidate joint control instruction is located outside the joint feasible instruction domain, but the joint feasible instruction domain is not empty, the safety controller determines the modified joint control instruction from within the joint feasible instruction domain.

[0097] Hierarchical optimization of joint control instructions: The modified joint control instructions are determined using a four-level hierarchical optimization approach. The joint feasible instruction domain serves as the initial solution set for the first level of optimization. After completing each of the first three levels of optimization, solutions whose absolute difference between the objective value and the optimal value of that level does not exceed the corresponding tolerance are retained, and these retained solutions are used as the solution set for the next level of optimization.

[0098] The first level of optimization aims to maximize the minimum user balance margin within the prediction time domain. This level prioritizes increasing the user balance margin at the most unfavorable prediction time, avoiding subsequent risk optimizations that compromise user balance safety.

[0099] The second level aims to minimize the overall traction risk. This overall traction risk is represented by a weighted sum of the following normalized risk quantities: the ratio of recoverable elastic energy storage to the energy storage absorption threshold, the ratio of the traction impulse in the imbalance direction to its limit, the ratio of the traction force change rate to its limit, and the sweep risk represented by the reciprocal of the predicted minimum distance from the predicted occupancy region of the traction component to the predicted occupancy region of the obstacle. The smaller the predicted minimum distance, the greater the corresponding sweep risk.

[0100] The third level optimizes by maximizing the weighted sum of the ratio of the intelligent guide dog's obstacle avoidance distance to the operational obstacle avoidance distance limit, and the ratio of the attitude stability margin to the attitude stability margin threshold. The intelligent guide dog's obstacle avoidance distance is the minimum predicted distance from the body's braking occupancy domain to the obstacle prediction occupancy domain.

[0101] The fourth level aims to minimize the deviation between the corrected joint control command and the candidate joint control command. For each control component, the absolute value of the deviation between the corrected value and the candidate value is divided by the difference between the upper and lower limits of that control component, and then summed according to preset positive weights. The optimal solution of the fourth level is determined as the corrected joint control command.

[0102] The weighted sums involved in the second, third and fourth levels of optimization mentioned above are all calculated according to preset positive weights. Each weight can be predetermined based on the user's safety priority, the sensitivity to traction risks and the stability requirements of the intelligent guide dog itself.

[0103] Thus, hierarchical optimization sequentially ensures user balance and safety, reduces traction risks, improves the body's obstacle avoidance and attitude stability capabilities, and preserves the original navigation intent of the dominant blind controller as much as possible while ensuring that the above safety objectives are not significantly affected.

[0104] Emergency recovery command field: When the joint feasible command domain is empty, it indicates that there is no joint control command that simultaneously satisfies all safety constraints under normal operating limits. The safety controller constructs the emergency recovery command domain based on the preset emergency obstacle avoidance distance limit, the preset emergency balance margin limit, the space occupancy constraint in the traction spatiotemporal mechanics domain, the energy storage absorption threshold, the traction impulse limit in the imbalance direction, and the current braking capacity constraint.

[0105] For each emergency joint control command to be evaluated, the safety controller re-predicts the corresponding body braking occupancy domain, traction component predicted occupancy domain, recoverable elastic energy storage, traction impulse in the direction of imbalance, and user balance state.

[0106] Only when the minimum distance between the cross sections of the main braking occupancy domain and the obstacle prediction occupancy domain at each prediction time is not less than the emergency obstacle avoidance distance limit, the corresponding cross sections of the traction component prediction occupancy domain and the obstacle prediction occupancy domain do not intersect, and the emergency balance margin constraint, traction impulse constraint, energy storage constraint and current braking capacity constraint are satisfied at the same time, will the corresponding emergency joint control command be included in the emergency recovery command domain.

[0107] When the current recoverable elastic energy storage is not higher than the energy storage absorption threshold, the emergency recovery joint control command should ensure that the recoverable elastic energy storage within the predicted time domain does not exceed the energy storage absorption threshold. When the current recoverable elastic energy storage is already higher than the energy storage absorption threshold, the emergency recovery joint control command should activate the energy absorption component to prevent the recoverable elastic energy storage from continuing to increase and reduce it to below the energy storage conversion threshold within the traction energy shaping time limit.

[0108] The emergency recovery joint control command should also enable the actual traction force vector to enter the intersection of the emergency-permitted traction force vector set and the achievable tension vector set.

[0109] Recovery startup conditions and safe handling time: The safety controller determines the prediction time when the body braking occupancy domain and the obstacle prediction occupancy domain first intersect at each prediction time based on the cross sections at each prediction time, and determines the difference between the prediction time and the current time as the predicted collision remaining time.

[0110] When no intersection occurs within the prediction time domain, the remaining time for predicted collision can be set to a value greater than the preset value in the prediction time domain; when the current cross section has already intersected, the remaining time for predicted collision can be set to zero.

[0111] The safety controller determines the required traction length adjustment and anchor point adjustment based on the difference between the target traction force vector and the current actual traction force vector.

[0112] The most unfavorable time required for traction length adjustment is determined by dividing the required traction length adjustment amount by the maximum allowable adjustment speed of the traction length; the most unfavorable time required for anchor point adjustment is determined by dividing the required anchor point position or angle adjustment amount by the maximum allowable adjustment speed of the anchor point; the most unfavorable time required for elastic energy storage to decrease to the energy storage conversion threshold is determined based on the energy to be absorbed and the guaranteed absorption power of the energy absorption components.

[0113] The sum of the maximum value among the above times and the traction force stabilization confirmation time is determined as the most unfavorable time required for traction energy shaping.

[0114] The first-stage deceleration time is determined based on the current speed, the first deceleration, the road gradient, and the ground adhesion conditions. The second-stage limited-jump deceleration time is determined by predicting the limited-jump deceleration process based on the current speed, the second deceleration, and the deceleration rate of change limit. Both the first and second decelerations represent deceleration amplitudes, with the second deceleration being greater than the first deceleration, and neither exceeding the maximum deceleration allowed by the current braking capacity.

[0115] The support adjustment time is determined based on the maximum movement distance of the foot to be adjusted, the maximum safe movement speed of the foot, and the foot contact confirmation time. The dwell time is determined based on the remaining speed after the support adjustment is completed and the current available braking capacity.

[0116] When it is necessary to continuously detect the stability of the body's posture and the user's balance to confirm a safe stop, the corresponding stop confirmation time is included in the stop time.

[0117] Traction energy shaping and the first-stage deceleration are performed in parallel. Therefore, the duration of each stage is the larger of the most unfavorable time required for traction energy shaping and the most unfavorable time required for the first-stage deceleration. This duration, the second-stage limit deceleration time, the support adjustment time, and the parking time are added together to obtain the staged guaranteed parking time.

[0118] The safe handling time will be determined by the sum of the phased guaranteed dwell time, the total delay of perception-decision-execution, and the preset safety margin.

[0119] In one embodiment, when the balance margin of each control cycle is reduced compared to the balance margin of the previous control cycle within a consecutive preset number of control cycles, and the reduction exceeds the margin change tolerance used to eliminate measurement noise, it is determined that the user's balance margin has been continuously decreasing within a preset duration.

[0120] When the predicted remaining collision time is not less than the safety handling time, the user's balance margin does not continuously decrease within the preset duration, and the safety sensor data used to generate the emergency recovery joint control command is valid, the preset recovery start conditions are determined to be met.

[0121] like Figure 4 As shown, when the emergency recovery command field is not empty and the preset recovery start conditions are met, the safety controller executes the emergency recovery joint control command; when the emergency recovery command field is empty or the preset recovery start conditions are not met, an emergency braking strategy limited by the current braking capacity is executed, and the energy absorption component is activated, specifically: When the emergency recovery command field is not empty and the preset recovery start conditions are met, the safety controller re-determines the direction of imbalance based on the latest obtained user balance state and selects the emergency recovery joint control command from the emergency recovery command field.

[0122] The emergency recovery process includes traction energy shaping and first-stage deceleration, second-stage limit deceleration, support adjustment, and parking.

[0123] During traction energy shaping and the first stage of deceleration, the intelligent guide dog is controlled to decelerate at the first deceleration rate. The anchor point is adjusted by regulating the length and direction of the traction component, so that the actual traction force vector enters the intersection of the emergency allowable traction force vector set and the achievable tension vector set. During the adjustment process, the traction force component in the direction of imbalance, the rate of change of traction force, and the cumulative traction impulse in the direction of imbalance are limited.

[0124] The safety controller compares the current recoverable resilient energy storage with the energy absorption threshold. When the recoverable resilient energy storage is not lower than the energy absorption threshold, the energy absorption component is activated. After activation, the energy absorption component maintains or adjusts its energy absorption function until the recoverable resilient energy storage decreases below the energy conversion threshold, in order to avoid frequent start-stop cycles near the threshold.

[0125] When the actual traction force vector remains within the set of emergency allowable traction force vectors, the recoverable elastic energy storage is below the energy storage conversion threshold, and the user's balance margin remains unchanged for a preset duration, the second stage of limited jump deceleration state is entered.

[0126] In the second stage of limited deceleration, the vehicle continues to decelerate at a second deceleration with an amplitude greater than that of the first deceleration, and the rate of change of deceleration is limited to prevent the braking deceleration from suddenly increasing and transmitting an excessive rate of change of traction force and an unbalanced traction impulse to the user through the traction component.

[0127] When the intelligent guide dog's walking speed decreases below the support transition speed threshold, it enters the support adjustment state. The safety controller lowers the body's center of gravity and controls the body posture adjustment mechanism and foot support mechanism to adjust the body posture and foot support state, improving posture stability margin.

[0128] The system enters a parking state when both the body attitude stabilization condition and the user balance condition are simultaneously met. The body attitude stabilization condition includes at least an attitude stabilization margin not less than the attitude stabilization margin threshold, and the body pitch and roll angular velocities not exceeding their corresponding limits. The user balance condition includes at least a user capture point located within the current support area or the predicted support area, and the user balance margin not less than the parking balance margin threshold.

[0129] like Figure 5 As shown, during the execution of the emergency recovery joint control command, the safety controller determines whether the preset exit conditions are met and whether the intelligent guide dog has reached the safe parking condition according to the control cycle, in order to determine whether to execute emergency braking, safe parking, or enter the next control cycle. Specifically: During the execution of the emergency recovery joint control command, the safety controller updates the predicted remaining collision time, remaining safety handling time, direction of imbalance, and emergency recovery command domain according to the control cycle.

[0130] The remaining safety response time will be recalculated based on the current emergency recovery phase and any subsequent phases that are not yet completed. Phases that have already been completed will not be included in the remaining safety response time.

[0131] The preset exit conditions are met when any of the following occurs: the updated predicted collision time is less than the remaining safety handling time, the traction energy shaping exceeds the preset time limit, the user balance margin continuously decreases within the preset duration, or the emergency recovery command field becomes empty.

[0132] When the emergency recovery command field is empty, the preset recovery start condition is not met, or the preset exit condition is met during the execution of the emergency recovery joint control command, the safety controller stops the phased emergency recovery, outputs an emergency braking command limited by the current braking capacity, and starts the energy absorption component.

[0133] When the recoverable elastic energy storage exceeds the energy absorption threshold or the traction impulse in the unbalanced direction exceeds the corresponding limit, the traction component is still released within the mechanical travel limit range. This release includes either a limited extension of the traction component or a sliding motion of the mechanical travel-limited release mechanism that creates a preset travel distance; it does not indicate a direct disconnection between the intelligent guide dog and the user.

[0134] If the intelligent guide dog has not yet reached the safe stopping condition, it will enter the next control cycle and re-execute the safety sensor data acquisition and safety domain update; when the safe stopping condition is reached, it will maintain the foot support state and end the current emergency response procedure.

[0135] like Figures 1 to 5 As shown, this embodiment also provides an intelligent guide dog traction safety control system for performing the above control method.

[0136] The system includes a main body, an adjustable traction component connecting the main body and the user, an environmental sensing component, a dominant blind controller, an independent safety sensor, a safety controller, a safety arbitrator, and an actuator.

[0137] Overall System Architecture The environmental perception component is connected to the dominant blind controller. The dominant blind controller generates candidate body motion commands based on environmental perception data and navigation information.

[0138] Environmental perception components may include navigation lidar, navigation vision sensors, positioning devices, and navigation map interfaces, and may also include inertial navigation devices or other sensing devices used to support road recognition, target detection, path planning, and navigation positioning.

[0139] Candidate body motion commands may include one or more of the following: target linear velocity, target angular velocity, target acceleration, target motion direction, body target attitude, target gait parameters, or target foot motion trajectory.

[0140] The independent safety sensor includes an environmental obstacle detection unit, a body status detection unit, a traction component status detection unit, and a user status detection unit, and is connected to the safety controller through a data channel independent of the master blind controller.

[0141] The environmental perception component and the environmental obstacle detection unit in the independent safety sensor can use different sensors, or they can use independent data acquisition, time synchronization, and processing channels. The safety controller's safety judgment does not depend on the processing results of the environmental perception data by the dominant blind controller.

[0142] The safety controller uses a different processor than the dominant blind controller and has the authority to veto candidate body motion commands.

[0143] The safety controller can also be configured with an independent watchdog, an independent communication interface, an independent safety task scheduler, and an independent power supply status monitoring interface.

[0144] Based on the data output by the independent safety sensors, the safety controller executes the specific steps described in the above embodiments, including state determination, safety domain construction, candidate instruction verification, corrective control, emergency recovery, and emergency braking.

[0145] Transmission of candidate joint control commands The dominant blind controller sends candidate body motion commands with control cycle numbers and timestamps to the safety controller and the safety arbitrator, respectively.

[0146] The safety controller generates candidate traction control quantities with the same control cycle number based on candidate body motion commands, traction component status, and user balance status, and matches the candidate body motion commands with the candidate traction control quantities to form candidate joint control commands.

[0147] The safety controller sends candidate joint control commands to the candidate command buffer channel of the safety arbitrator. After entering the buffer, the candidate joint control commands are not sent directly to the actuator, but wait for the corresponding safety verification signal or alternative control command.

[0148] When a candidate joint control command is located within the joint feasible command domain, the safety controller sends a safety verification signal with the same control cycle number.

[0149] When a candidate joint control instruction is located outside the joint feasible instruction field but the joint feasible instruction field is not empty, the safety controller outputs a modified joint control instruction.

[0150] When the joint feasible instruction field is empty and the emergency recovery instruction field is not empty and the preset recovery start conditions are met, the safety controller outputs the emergency recovery joint control instruction.

[0151] When the emergency recovery command field is empty, the preset recovery start condition is not met, or the preset exit condition is met during the emergency recovery process, the safety controller outputs an emergency braking command and an energy absorption command.

[0152] Determination of safety sensor data and status variables The safety controller acquires safety sensor data output by independent safety sensors according to a preset control cycle, and determines the body status and braking capability, environmental obstacle status, traction component status and user balance status based on the safety sensor data.

[0153] The body state detection unit includes at least a safety inertial sensor for obtaining the body's inertial state, and may include one or more of the following: a joint state detection unit, a foot contact detection unit, a plantar force detection unit, and a drive state detection unit. When the corresponding state can be reliably obtained through other independent safety detection devices or multi-source state estimation, the corresponding equivalent state estimation result may also be used.

[0154] Safety inertial sensors are used to acquire body acceleration, angular velocity, and attitude information; joint state detection units are used to acquire joint position, joint velocity, and driving state; foot contact detection units are used to determine the contact and support states of each foot; and plantar force detection units are used to acquire foot support force or equivalent ground force.

[0155] The safety controller determines the body position, speed, posture, foot support configuration, and sliding state based on the above data.

[0156] The safety controller determines the current braking capacity based on the current speed of the main body, the attitude of the main body, the contact state of the foot end, the ground adhesion coefficient, the road slope, and the braking capacity parameters of the main body motion actuator.

[0157] The braking capability parameters of the main body motion actuator may include one or more of the following: allowable output torque of the drive motor or joint actuator, allowable braking torque, power limit, and thermal state limit. These braking capability parameters can be obtained through factory calibration, system initialization calibration, or status parameters provided by the actuator controller, and are stored in the safety controller or a parameter storage unit that can be read by the safety controller.

[0158] The current braking capability includes at least the range of braking deceleration or equivalent braking force allowed under the current condition, and may also include the deceleration rate of change limit and the expected braking distance.

[0159] The ground adhesion coefficient can be obtained from adhesion parameters pre-stored according to the road type, or it can be corrected online based on the slippage state of the foot relative to the ground. The road type can be identified from environmental perception data, obtained from road attributes in the navigation map, or pre-set during system initialization.

[0160] The road slope can be determined based on the body posture measured by the safety inertial sensor and the foot contact plane.

[0161] The environmental obstacle detection unit includes a safe ranging sensor. The safe ranging sensor may include one or more of the following: lidar, millimeter-wave radar, depth sensor, or ultrasonic ranging sensor.

[0162] The safety controller determines the location and spatial occupancy of obstacles based on distance data, point cloud data, or target detection data output by the environmental obstacle detection unit, and determines the speed and direction of movement of the obstacle based on the positional changes of the same obstacle within multiple consecutive control cycles.

[0163] The obstacle perception error boundary is determined based on one or more of the following: measurement error of the safety ranging sensor, target boundary extraction error, time synchronization error, and obstacle motion state estimation error, thereby forming the environmental obstacle state.

[0164] The traction component status detection unit includes one or more of the following: traction force sensor, traction length encoder, direction adjustment anchor point position detection unit, and series elastic component deformation detection unit.

[0165] The position and speed of the intelligent guide dog's side connection point are determined by adjusting the position or angle of the anchor point and its installation geometry based on the body's posture and orientation; the position and speed of the user's side connection point are determined based on the user's status detection results; and the traction length and its rate of change are determined based on the traction length encoder output.

[0166] The safety controller determines the traction force amplitude based on the output of the traction sensor and determines the traction direction based on the actual output direction of the traction component at the user-side connection point.

[0167] When there is no intermediate guide structure between the intelligent guide dog side connection point and the user side connection point, the traction direction can be determined based on the relative positional relationship between the two connection points; when there is a direction adjustment anchor point, guide wheel or other intermediate guide structure, the traction direction can be determined based on the direction of the traction segment adjacent to the user side connection point.

[0168] The safety controller combines the traction force amplitude with the traction direction to obtain the traction force vector, and determines the traction force change rate based on the change of the traction force vector within a continuous control cycle.

[0169] The tension or slack state of the traction assembly is determined based on the relationship between the traction force amplitude and the actual traction length and the geometric path length between the two connection points.

[0170] The elastic parameters of the series elastic components and the damping parameters of the damping components can be obtained through pre-calibration and stored in the safety controller, or they can be corrected based on the force and displacement responses measured during the operation of the traction components.

[0171] The user status detection unit includes one or more of wearable inertial sensors, vision sensors, and connection point position detection devices.

[0172] Wearable inertial sensors can be placed at one or more locations on the user's torso, waist, lower limbs, or feet to acquire the posture, angular velocity, and motion acceleration of the corresponding body parts.

[0173] The visual sensor is used to acquire the location of at least some key points on the user’s body, including the torso, hips, knees, ankles, and feet, and can further determine the relative position between the user and the intelligent guide dog.

[0174] The connection point position detection device is used to directly detect or assist in determining the position of the user-side traction connection point.

[0175] When the status of some users cannot be directly obtained from a single detection source, the safety controller combines wearable inertial data, visual human key points, traction length, connection point position, and other user status detection data to perform multi-source fusion or status estimation.

[0176] The user's center of mass position can be determined based on the positions of key points on the human body and pre-stored or calibrated human segment parameters obtained by using individual user parameters, and the center of mass velocity can be determined based on the changes in the center of mass position during continuous control cycles or in combination with the output of wearable inertial sensors.

[0177] The gait phase is determined based on one or more of the following: foot position, foot speed, foot lift-off state, and body inertia state, and the current supporting foot or the predicted supporting foot is determined accordingly.

[0178] The safety controller combines the user's center of gravity position and velocity, torso tilt angle and rate of change of tilt angle, support foot position, gait phase, relative position of the user and the intelligent guide dog, and traction vector to form the user's balance state, and further determines the target support area, capture point, balance margin, and direction of imbalance based on the user's balance state.

[0179] Based on the determined body state and braking capacity, environmental obstacle state, traction component state and user balance state, the safety controller constructs the body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal mechanics domain and recoverable balance domain respectively according to the corresponding steps of the above embodiments.

[0180] Secure Arbitrator: The actuators include a body motion actuator, a traction actuator, and an energy absorption actuator, and receive control commands only through a safety arbitrator.

[0181] The security arbitrator performs arbitration based on the control cycle number, timestamp, and instruction type.

[0182] When a modified joint control command, an emergency recovery joint control command, or an emergency braking command is received, the candidate joint control commands for the corresponding period are cleared or frozen, and the corresponding safety control commands are transmitted to the actuators first.

[0183] When only a security verification signal with the same control cycle number as the candidate joint control instruction is received, the cached candidate joint control instruction is transmitted to the actuator.

[0184] The candidate joint control command is not allowed to pass if the security verification signal is inconsistent with the control cycle number of the candidate joint control command, the timestamp expires, or the data verification fails.

[0185] When neither a valid control command output by the safety controller nor a valid safety verification signal is received within a preset arbitration cycle, the safety arbitrator transmits the default emergency braking command and default energy absorption command pre-stored in the safety arbitrator to the actuator.

[0186] The default emergency braking command and default energy absorption command are stored locally in the safety arbitrator and are not calculated in real time by the dominant blind controller or the safety controller.

[0187] Adjustable traction components and actuators The adjustable traction assembly includes a traction length adjustment mechanism, a series elastic component, a damping component, an energy absorption component, a driveable direction adjustment anchor point, a traction force sensor, and a traction length encoder.

[0188] In one specific embodiment, the traction length adjustment mechanism includes a servo motor, a reducer, a winding wheel, a rope guide mechanism, and a controlled locking mechanism.

[0189] The servo motor drives the winding wheel to rotate via a reducer, thereby shortening or lengthening the traction component.

[0190] The traction length encoder is located at the winding wheel or the outlet of the traction assembly to detect the actual traction length and the rate of change of traction length.

[0191] The series elastic components are installed in the traction force transmission path and can be helical springs, leaf springs, elastic rope segments, or elastomer assemblies. The series elastic components are equipped with deformation detection units to detect the amount of elastic deformation.

[0192] Damping components are used to continuously or passively dissipate the vibration energy of the traction assembly during normal operation, and can be viscous dampers, fluid dampers, or friction dampers.

[0193] The energy absorption component is a component that can selectively start or adjust its absorption capacity according to the energy absorption command. It can be a controllable electromagnetic brake, a power generation feedback mechanism, a magnetorheological energy absorption mechanism, or a controllable frictional energy absorption mechanism.

[0194] The damping component mainly suppresses vibration during normal traction, while the energy absorption component mainly absorbs the recoverable elastic energy released by the series elastic component during emergency recovery or emergency braking.

[0195] The directional adjustment anchor point may include a linear motion slide, a rotary table, or a combination of both.

[0196] The anchor point drive unit can change the position of the connection point of the traction component on the side of the smart guide dog, the direction of the rope exit, or both.

[0197] The traction actuator includes a length adjustment drive unit for driving the traction length adjustment mechanism and an anchor point drive unit for changing the direction and adjusting the position or angle of the anchor point.

[0198] The energy absorption actuator controls the energy absorption component according to the energy absorption command output by the safety controller via the safety arbitrator.

[0199] Traction length and directional control: The safety controller determines the set of permissible traction force vectors based on the recoverable balance domain and the direction of imbalance, and determines the actual achievable tension vector set by adjusting the position or angle of the anchor point, the relative position of the body and the user, and the tension state of the adjustable traction components according to the direction.

[0200] The target traction vector for maintaining or improving the user's balance margin is determined from the intersection of the set of permissible traction vectors and the set of achievable tension vectors.

[0201] The relative speed of the user moving away from the intelligent guide dog along the traction direction is set to a positive value, and the relative speed of approaching along the traction direction is set to a negative value.

[0202] The safety controller first limits the relative speed, then integrates the limited speed, and limits the traction length range of the integration result to obtain the basic compensation amount of the traction length.

[0203] When the user moves away from the intelligent guide dog, the leash should be extended appropriately; when the user moves closer to the intelligent guide dog, the leash should be shortened appropriately to compensate for normal human-machine relative movement.

[0204] The safety controller determines the traction length correction and anchor point adjustment based on the deviation between the target traction force vector and the actual traction force vector, as well as the energy margin determined by the difference between the energy absorption threshold and the current recoverable elastic energy.

[0205] The traction length correction and anchor point adjustment can be determined through proportional feedback, proportional-integral feedback, constraint optimization, or model predictive control.

[0206] When the energy storage margin is large, the traction effect in the target direction can be improved by appropriately shortening the traction components and adjusting the anchor points.

[0207] When the energy storage margin is close to the margin threshold, to increase the control cost of continuing to tighten the traction component, it is preferable to change the anchor point direction, limit the rate of change of traction force, or appropriately lengthen the traction component.

[0208] The basic compensation amount for traction length is superimposed with the correction amount for traction length to form a traction length adjustment command, and a traction direction adjustment command is formed based on the anchor point adjustment amount.

[0209] The traction length adjustment command and the traction direction adjustment command are subject to the traction length range, the upper limit of the traction length adjustment speed, the anchor point adjustment range, the upper limit of the anchor point adjustment speed, the limit of the traction force change rate, the limit of the traction impulse in the direction of imbalance, and the space occupancy limit of the traction components determined by the traction spatiotemporal mechanical domain.

[0210] Under normal operating conditions, the actual traction force is limited by the lower limit of the normal guiding tension.

[0211] In emergency recovery mode, an emergency guide tension lower limit lower than the normal guide tension lower limit is adopted, and this emergency guide tension lower limit can be set to zero.

[0212] In emergency braking, the lower limit of the guide tension is released, and the execution priority of the traction component release control and energy absorption control is made higher than that of the guide tension holding control.

[0213] Safety sensor effectiveness and failure handling The safety controller performs timestamp continuity, delay, range, and multi-source consistency checks on the data from each safety sensor.

[0214] When the timestamps of safety sensor data are discontinuous, the delay exceeds the corresponding threshold, the data exceeds the effective range, or the multi-source estimation difference of the same state quantity exceeds the consistency threshold, the corresponding sensor output is determined to be invalid.

[0215] When the traction sensor output is invalid, an estimated traction force is generated based on the drive current of the traction actuator and the deformation of the series elastic component.

[0216] Specifically, the first traction force estimate is determined based on the motor torque constant, transmission ratio, winding wheel radius, and drive current; the second traction force estimate is determined based on the deformation of the series elastic component and its restoring force-deformation calibration relationship.

[0217] When the difference between two traction force estimates does not exceed the estimation consistency threshold, the two are weighted and fused based on the current detection accuracy and elastic deformation detection accuracy to obtain the traction force estimate; when the difference exceeds the estimation consistency threshold, the traction force estimate is determined to be invalid.

[0218] When both the traction sensor and the traction force estimate are invalid, or the traction length encoder output is invalid, traction length adjustment based on traction force feedback is prohibited. The energy absorption component is activated, and the mechanical travel limit release mechanism is activated within the mechanical travel limit range to control the intelligent guide dog to stop.

[0219] The mechanical limited-travel release mechanism can employ a differential clutch with a maximum permissible slippage, a limited-travel leash release mechanism, or a limited-travel slippage mechanism after controlled release. Its release stroke is limited by a mechanical stop to prevent unrestricted leash release or direct disconnection between the intelligent guide dog and the user.

[0220] When the output of the environmental obstacle detection unit is invalid, the unverified environmental area is set as the obstacle-occupied area.

[0221] If a verified passable area exists, decelerate and stop along the verified passable area; if no verified passable area exists, perform emergency braking and activate the energy absorption components.

[0222] When the output of the body state detection unit or the user state detection unit is invalid, the corresponding state uncertainty area is expanded to the preset failure uncertainty area, and the allowable speed, allowable acceleration and traction force change rate are reduced to control the intelligent guide dog to decelerate and stop.

[0223] Only after the corresponding safety sensor data recovers to a valid state for multiple consecutive control cycles and passes the consistency check can the corresponding degraded state be exited.

[0224] User individual parameter calibration The safety controller includes a user-specific parameter calibration module.

[0225] Individual parameter calibration can be performed upon first use, when the user's physical condition changes significantly, or when the preset review period expires.

[0226] The calibration process is conducted within the traction force range determined by preset initial safety limits. The user individual parameter calibration module forms a calibration combination according to at least two non-collinear traction directions, at least two traction force amplitude levels, and at least two gait phases.

[0227] The traction direction may include the direction of travel and the lateral direction, or it may include other non-collinear directions determined according to the user's body orientation.

[0228] Gait phases may include at least two of the following: double support phase, left single support phase, right single support phase, or the phase before the swing foot lands.

[0229] For each calibration combination, the controller applies controlled traction to the user using a restricted ramp, short-time hold, or restricted pulse method, and collects the user's center of mass position and velocity, torso tilt angle and rate of change of tilt angle, and capture point change.

[0230] For each calibration combination, the peak value of trunk tilt angle, the peak value of trunk tilt angle change rate, the minimum balance margin, the recovery time of the capture point, the number of support adjustment steps used during the recovery process, the traction impulse in the direction of imbalance, and the traction impulse in the direction of recovery are extracted respectively.

[0231] When the torso tilt angle, tilt angle change rate, and traction impulse do not exceed the initial safety limits, and the capture point can enter the target support area within the allowable adjustment steps to meet the calibration balance margin, the corresponding calibration combination is determined as the safe calibration combination.

[0232] For each traction direction and gait phase, the maximum safe traction force in the safety calibration combination is multiplied by a preset safety factor to determine the corresponding upper limit of traction force; the limit of traction force change rate is determined based on the maximum traction force change rate that does not cause a continuous decrease in balance margin; and the limits of traction impulse in the direction of imbalance and the direction of recovery are determined by accumulating the orthogonal projections of traction force in the direction of imbalance and the direction of recovery.

[0233] The recovery time of the most unfavorable capture point in the safety calibration combination is added to the time safety margin to determine the capture point recovery time; the number of allowable adjustment steps is determined based on the number of support adjustment steps that can restore balance; and the lower limit of the balance margin is determined based on the confidence lower bound or the most unfavorable value of the safety calibration data.

[0234] The individual traction safety limits determined thereby include the traction force vector range, the traction force change rate limit, the traction impulse limit in the direction of imbalance, the traction impulse limit in the direction of recovery, the trunk tilt angle and its change rate limit, the capture point recovery time, the number of allowable adjustment steps, and the lower limit of the balance margin.

[0235] The safety controller adjusts the recoverable balance domain, the set of permissible traction force vectors, and the set of permissible emergency traction force vectors according to the individual user's traction safety limits.

[0236] Body posture and foot support adjustment The system also includes a body posture adjustment mechanism and a foot support mechanism.

[0237] The body posture adjustment mechanism changes the body height, pitch angle, and roll angle through the leg joints; the foot support mechanism is used to adjust the foot position and maintain the foot in contact.

[0238] The safety controller generates at least two candidate foot target configurations within the kinematic reach of each foot. Each candidate foot target configuration includes at least one foot target position, and may also include the body target height and the expected foot contact state.

[0239] Candidate foot-end target configurations can be generated by sampling within the reachable area of ​​the foot end, or by optimizing the configuration with the goal of increasing the area of ​​the foot-end support region, increasing the distance of the support boundary in the direction of imbalance, or improving the attitude stability margin.

[0240] For each candidate configuration, the safety controller examines the foot kinematic constraints, joint position and moment constraints, and contact friction cone constraints, and predicts the corresponding foot contact position and ground reaction force.

[0241] The safety controller determines the equivalent zero moment point based on the predicted traction force vector and its position, the body's inertial force, gravity, and the ground support reaction forces at each foot and their positions, according to the balance relationship between the resultant force and resultant moment in the support plane of the body's feet.

[0242] The support area at the foot end of the body is determined by the convex hull of the current contact position of the foot end or by the support polygon after considering the contact area of ​​the foot end.

[0243] The attitude stability margin is the directed distance from the equivalent zero moment point to the boundary of the foot support region of the body. The attitude stability margin is positive when the equivalent zero moment point is located inside the foot support region of the body; it is zero when it is located on the boundary; and it is negative when it is located outside the boundary.

[0244] The safety controller selects the configuration with the largest attitude stability margin from the candidate configurations with an attitude stability margin not less than the attitude stability margin threshold, and controls the foot support mechanism to adjust the corresponding foot.

[0245] When no candidate configuration meets the posture stability margin threshold, maintain the current foot support state, lower the body's center of gravity, and control the intelligent guide dog to slow down, so as to avoid forcibly changing the support state when there is no stable footing position.

[0246] When the travel speed is lower than the support conversion speed threshold, the body attitude adjustment mechanism begins to lower the body's center of gravity.

[0247] When the walking speed is lower than the parking speed threshold, the posture stability margin determined based on the current foot support state is not less than the posture stability margin threshold for two consecutive control cycles, and the user capture point is located in the user's current support area or predicted support area selected based on the gait phase, and the distance from the capture point to the boundary of the selected area is not less than the parking balance margin threshold, the foot support mechanism locks the foot support state.

[0248] Therefore, the parking state of the intelligent guide dog simultaneously meets the requirements of both the dog's posture stability and the user's balance stability, avoiding the situation where the dog's speed drops to zero as the condition for completing an emergency response.

[0249] In summary, this application uses a safety controller to acquire state data from the guide dog, environmental obstacles, traction components, and the user, respectively constructing a braking occupancy domain, an obstacle prediction occupancy domain, a traction spatiotemporal dynamics domain, and a recoverable balance domain, and forming a joint feasible command domain based on these. This joint feasible command domain simultaneously constrains the intelligent guide dog's braking and obstacle avoidance capabilities, the traction component's spatial occupancy and elastic energy storage, the traction impulse in the direction of imbalance, and the user's balance recovery ability, thereby jointly assessing the guide dog's motion safety, traction safety, and user balance safety.

[0250] The primary blind controller generates candidate motion commands based on navigation information and environmental perception data, while the safety controller generates candidate traction control quantities that match these commands. Candidate joint control commands are executed if they satisfy the joint feasible command domain; otherwise, if they do not satisfy the domain but the joint feasible command domain is not empty, the corrected joint control command is determined according to the priority order of user balance, traction risk, vehicle stability, and command deviation.

[0251] When the joint feasible command field is empty, the safety controller determines whether an emergency recovery can be performed based on the predicted remaining collision time, safety handling time, current braking capacity, and user balance status. If the recovery conditions are met, the traction force is brought within the emergency allowable range and the intelligent guide dog is safely parked by adjusting the length and direction of the traction component, absorbing elastic energy, and performing phased deceleration. If the recovery conditions are not met or the safety status continues to deteriorate during the recovery process, emergency braking, energy absorption, and restricted traction release are performed.

[0252] The system of this application includes a guide dog body, an adjustable traction component, an environmental sensing component, a dominant blind controller, independent safety sensors, a safety controller, a safety arbitrator, and an actuator. The safety controller has the authority to verify and reject candidate body movement commands, and in the event of sensor failure, it performs degraded control by substituting traction force estimation, obstructing unknown areas, or expanding the uncertain state area. The system can also adjust traction safety limits through calibration using individual user parameters, and improve the support stability of the intelligent guide dog during traction and deceleration through body posture adjustment and foot support control.

[0253] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application's patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application. These modifications and improvements are equivalent alterations and evolutions of the above embodiments based on the essential technology of this application, and all fall within the protection scope of this application.

Claims

1. A method for controlling the safety of a smart guide dog's pulling, characterized in that, Applications in intelligent guide dogs include: Acquire safety sensor data from the intelligent guide dog; Based on the acquired safety sensor data, the body state, braking capacity, environmental obstacle state, traction component state, user balance state, and direction of imbalance are determined, and the corresponding body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal mechanics domain, and recoverable balance domain are constructed. Based on the body braking occupancy domain, the obstacle prediction occupancy domain, the recoverable balance domain, the traction spatiotemporal dynamics domain, and the preset conditions, the set of body motion control quantities and traction control quantities that satisfy the preset conditions is determined as the joint feasible command domain. Based on the candidate body movement commands generated by the dominant blind controller of the intelligent guide dog, the state of the traction component, and the balance state of the user, candidate traction control quantities are generated, and the candidate body movement commands and candidate traction control quantities are combined to form candidate joint control commands. When the candidate joint control instruction is located within the joint feasible instruction domain, the intelligent guide dog is controlled to execute the candidate joint control instruction. When the candidate joint control instruction is located outside the joint feasible instruction domain and the joint feasible instruction domain is not empty, the intelligent guide dog is controlled to execute the modified joint control instruction selected from the joint feasible instruction domain. When the joint feasible command domain is empty, an emergency recovery command domain is constructed based on the preset emergency obstacle avoidance distance limit, the preset emergency balance margin limit, the space occupancy constraint in the traction spatiotemporal mechanics domain, the energy storage absorption threshold, the traction impulse limit in the imbalance direction, and the braking capacity constraint. When the emergency recovery command field is not empty and the preset recovery start conditions are met, an emergency recovery joint control command is determined and executed from the emergency recovery command field. This command controls the intelligent guide dog to adjust the length of the traction component and the direction of traction, so that the actual traction force vector enters the set of emergency permissible traction force vectors determined according to the user's balance state and the user's individual traction safety limit. It also limits the traction impulse in the direction of imbalance and controls the intelligent guide dog to decelerate and stop. During the execution of the emergency recovery joint control command, the energy absorption component is controlled according to the comparison result between the recoverable elastic energy storage and the energy absorption threshold. When the emergency recovery command field is empty, the preset recovery start condition is not met, or the preset exit condition is met during the execution of the emergency recovery joint control command, the intelligent guide dog is controlled to perform emergency braking and the energy absorption component is activated.

2. The intelligent guide dog's traction safety control method according to claim 1, wherein, The preset conditions include: The minimum distance between the body braking occupancy domain and the obstacle prediction occupancy domain at each prediction time section is not less than the preset obstacle avoidance distance limit. The predicted occupancy region of the traction component does not intersect with the corresponding cross section of the predicted occupancy region of the obstacle; The recoverable elastic energy storage and the traction impulse in the direction of imbalance shall not exceed the energy storage absorption threshold and the preset limit value of the traction impulse in the direction of imbalance, respectively. The user's state is in a recoverable equilibrium zone. 3.The intelligent leash safety control method of a guide dog according to claim 1, wherein, The main body braking occupancy domain is composed of the brakeable occupancy time space tube in the prediction time domain; For each braking switching moment in the prediction time domain, the pose sequence of the intelligent guide dog executing the joint control command to be evaluated and switching to the braking strategy limited by the braking capability at the corresponding braking switching moment until parking is predicted; at each prediction moment, the Minkowski summation is performed on the ontological geometric model of the predicted pose transformation and the state uncertainty region, and the resulting region and the corresponding prediction moment form a spatiotemporal section. The union of each spatiotemporal section is obtained to obtain the brakeable occupancy spatiotemporal tube; Based on the position, velocity, and perception error boundary of environmental obstacles, an obstacle prediction occupancy domain is constructed. The cross-sections of the brakeable occupancy space tube and the obstacle prediction occupancy domain at each prediction time are not intersecting and the minimum distance between the cross-sections is not less than the running obstacle avoidance distance limit, which is determined as the braking obstacle avoidance constraint. Wherein: the predicted pose is determined based on the intelligent guide dog's walking speed, braking ability, ground adhesion coefficient, road slope, foot contact state, and total delay from perception to execution; the state uncertainty region is expanded along the walking direction, the lateral slip direction, and the spatial offset direction caused by pitch error and roll error, respectively, according to the corresponding state error boundary; the joint control command to be evaluated is a candidate joint control command or a joint control command to be optimized. 4.The intelligent leash safety control method of a guide dog according to claim 1, wherein, The user's balance state includes the user's center of mass position and velocity, trunk tilt angle and rate of change of tilt angle, supporting foot position, gait phase, relative position of the user and the intelligent guide dog, and traction force vector; The current support area or the predicted support area is determined as the target support area based on the gait phase; the recoverable balance domain is a set of user states that meet the following conditions: there exists a traction control sequence that meets the user's individual traction safety limit, center of mass dynamics constraint and the reachability range of the support foot landing point, so that the user's torso tilt angle and tilt angle change rate do not exceed the corresponding limit, and the capture point enters the target support area within the allowed number of adjustment steps with a balance margin not less than the running balance margin limit; The operational balance margin limit is a preset non-negative value; The balance margin is the directed distance from the capture point to the boundary of the target support area. It is positive when the capture point is within the target support area and negative when it is outside the target support area. The outward normal direction of the target support area at the boundary point closest to the capture point is determined as the direction of imbalance, and its opposite direction is determined as the direction of recovery. The set of emergency permissible traction force vectors is jointly defined by the upper and lower limits of the traction force component in the recovery direction, the upper limit of the traction force component in the imbalance direction, the limit of the traction force change rate, and the limit of the traction impulse in the imbalance direction. When the balance margin is less than the lower limit of the balance margin, the upper limit of the traction force component in the imbalance direction, the limit of the traction force change rate, and the limit of the traction impulse in the imbalance direction shall be emergency limits that are less than the corresponding preset operating limits. 5.The intelligent leash safety control method for a guide dog according to claim 4, wherein, The traction component status includes the position and speed of the intelligent guide dog side connection point and the user side connection point, the traction length and its rate of change, the traction force vector and its rate of change, the tensioned or relaxed state, and the elastic and damping parameters. The traction spatiotemporal mechanical domain includes the traction component's predicted occupancy domain and the recoverable elastic energy storage, the traction impulse in the imbalance direction, and the traction impulse in the recovery direction associated with each prediction moment. In the tensioned state, the traction component's predicted occupancy domain includes the area swept between adjacent prediction moments by the connecting segment formed by the intelligent guide dog side connection point and the user side connection point at each prediction moment, together with the traction component's cross-section. In the relaxed state, the traction component's predicted occupancy domain also includes a sagging and swaying uncertainty region determined by the traction length margin, gravity sag, and lateral sway amplitude. The recoverable elastic energy storage is the integral of the elastic restoring force along the elastic deformation path with respect to the elastic deformation amount, and does not include the energy consumed by the damping component or the mechanical work output or absorbed by the traction actuator; the traction impulse in the imbalance direction and the traction impulse in the recovery direction are the time integrals of the orthogonal projection of the traction force vector in the imbalance direction and the recovery direction in the prediction time domain, respectively. The predicted recoverable elastic energy storage is the maximum value of recoverable elastic energy storage within the prediction time domain; the predicted traction impulse in the direction of imbalance is the cumulative traction impulse in the direction of imbalance at the end of the prediction time domain; and the predicted traction force change rate amplitude is the maximum value of the traction force change rate amplitude within the prediction time domain. The energy storage margin, impulse margin, and change rate margin are the differences between the energy storage absorption threshold and the predicted recoverable elastic energy storage, the difference between the traction impulse limit in the direction of imbalance and the predicted traction impulse in the direction of imbalance, and the difference between the traction force change rate limit and the predicted traction force change rate amplitude, respectively. When the ratio of any margin to the corresponding limit is less than the margin threshold, the predicted occupancy domain of the traction component is expanded along the predicted rebound direction, and the upper limit of the traction length adjustment speed, the limit of the traction force change rate, and the upper limit of the intelligent guide dog acceleration are reduced; when the predicted lateral swing amplitude exceeds the lateral swing amplitude threshold, the predicted occupancy domain of the traction component is expanded along the lateral swing direction. 6.The intelligent walking stick dog harness safety control method of claim 5, wherein, The modified joint control instructions are generated through hierarchical optimization. Using the joint feasible instruction domain as the initial solution set, after completing the optimization of each of the first three levels, the solution whose absolute value of the difference between the objective value and its optimal value does not exceed the corresponding tolerance is determined as the next level solution set, and then executed sequentially: The first-level optimization objective is to maximize the minimum value of the user balance margin in the prediction time domain. The second-level optimization objective is to minimize the weighted sum of the following: the ratio of recoverable elastic energy storage to energy storage absorption threshold, the ratio of traction impulse in the imbalance direction to its limit, the ratio of traction force change rate to its limit, and the reciprocal of the predicted minimum distance from the predicted occupied area of ​​the traction component to the predicted occupied area of ​​the obstacle. The third-level optimization objective is to maximize the weighted sum of the ratio of the obstacle avoidance distance of the intelligent guide dog to the limit of the running obstacle avoidance distance and the ratio of the attitude stability margin to the attitude stability margin threshold. The attitude stability margin is the directed distance from the equivalent zero moment point to the boundary of the foot support area. It is positive when the equivalent zero moment point is within the foot support area and negative when it is outside the foot support area. The fourth-level optimization objective is to minimize the weighted sum of the ratios of the absolute values ​​of the deviations of each control component between the corrected joint control command and the candidate joint control command to the differences between the upper and lower limits of the corresponding control components, and the optimal solution of the fourth level is determined as the corrected joint control command; each weighted sum is calculated according to a preset positive weight. 7.The intelligent walking stick dog harness safety control method of claim 4, wherein, The process of determining the preset recovery start conditions and the preset exit conditions includes: Based on the cross sections of the body braking occupancy domain and the obstacle prediction occupancy domain at each prediction time, the prediction time at which the two first intersect is determined, and the difference between the prediction time at which the first intersection occurs and the current time is determined as the predicted collision remaining time. Based on the current status of the main body, braking capacity, traction component status, and user balance, and under the constraints of the first deceleration, second deceleration, deceleration change rate limit, traction adjustment capability, and support adjustment capability, the predicted phased guaranteed parking time required to sequentially complete traction energy shaping and the first stage deceleration, the second stage limited-jump deceleration, support adjustment, and parking is determined. The sum of the phased guaranteed parking time, the total delay from perception to decision to execution, and the preset safety margin is determined as the safe handling time. Among them, when traction energy shaping and the first stage deceleration are executed in parallel, the duration of the corresponding stage is taken as the larger value between the most unfavorable time required to complete traction energy shaping and the most unfavorable time required to complete the first stage deceleration. The preset recovery initiation conditions are determined as follows: the predicted remaining collision time is not less than the safety handling time, the user's balance margin does not continuously decrease within the preset duration, and the safety sensor data used to generate the emergency recovery joint control command is valid. When the preset recovery start-up conditions are met, the direction of imbalance is re-determined based on the latest obtained user balance state, and the traction energy shaping and first-stage deceleration state is entered. By adjusting the length of the traction component and the traction action direction, controlling energy absorption and applying the first deceleration, the actual traction force vector is brought into the emergency allowable traction force vector set, and the recoverable elastic energy is reduced to below the preset energy conversion threshold. When the actual traction force vector remains within the set of emergency allowable traction force vectors, the recoverable elastic energy is lower than the energy conversion threshold, and the user's balance margin remains unchanged for a preset duration, the system enters the second stage of limited deceleration state, continues to decelerate at a second deceleration with an amplitude greater than the first deceleration, and limits the rate of change of deceleration. When the walking speed of the intelligent guide dog decreases below the support transition speed threshold, it enters the support adjustment state, which improves the posture stability margin by adjusting the body posture and foot support state; when the body posture stability condition and the user balance condition are met at the same time, it enters the parking state. During the execution of the emergency recovery joint control command, the predicted remaining collision time, remaining safety handling time, imbalance direction, and emergency recovery command field are updated according to the control cycle. When the updated predicted remaining collision time is less than the remaining safety handling time, the traction energy shaping exceeds a preset time limit, the user's balance margin continuously decreases within a preset duration, or the emergency recovery command field becomes empty, the preset exit condition is determined to be met, the phased emergency recovery is terminated, emergency braking limited by the current braking capacity is executed, and the energy absorption component is activated. When the recoverable elastic energy storage exceeds the energy absorption threshold or the traction impulse in the imbalance direction exceeds the corresponding limit, the traction component is also released within the mechanical travel limit range.

8. A smart guide dog leash safety control system, characterized in that, It includes a main body, an adjustable traction component connecting the main body and the user, an environmental sensing component, a dominant blind controller, an independent safety sensor, a safety controller, a safety arbitrator, and an actuator; The environmental perception component is connected to the dominant blind controller, which is used to generate candidate body motion commands based on environmental perception data and navigation information. The independent safety sensor includes an environmental obstacle detection unit, a body status detection unit, a traction component status detection unit, and a user status detection unit, and is connected to the safety controller through a data channel independent of the dominant blind controller; The safety controller is used to determine the body state and braking capability, environmental obstacle state, traction component state and user balance state based on the safety sensor data output by the independent safety sensor, determine the direction of imbalance, and construct the body braking occupancy domain, obstacle prediction occupancy domain, traction spatiotemporal mechanical domain and recoverable balance domain. The safety controller is also used to determine the set of body motion control quantities and traction control quantities that satisfy the preset conditions as a joint feasible command domain based on the body braking occupancy domain, the obstacle prediction occupancy domain, the recoverable balance domain, the traction spatiotemporal mechanics domain, and preset conditions. The safety controller is further configured to: output a safety verification signal to the safety arbitrator when the joint feasible instruction domain is not empty and the candidate joint control instruction is located within the joint feasible instruction domain; output a modified joint control instruction determined by the joint feasible instruction domain when the joint feasible instruction domain is not empty but the candidate joint control instruction is located outside the joint feasible instruction domain; construct an emergency recovery instruction domain when the joint feasible instruction domain is empty; output an emergency recovery joint control instruction when the emergency recovery instruction domain is not empty and a preset recovery start condition is met; and output an emergency braking instruction and an energy absorption instruction when the emergency recovery instruction domain is empty, the preset recovery start condition is not met, or a preset exit condition is met during the execution of the emergency recovery joint control instruction. The actuator includes a body motion actuator, a traction actuator, and an energy absorption actuator, and receives control commands only through the safety arbitrator; The safety arbitrator is used to: preferentially transmit the corresponding control command to the actuator when it receives a modified joint control command, an emergency recovery joint control command, or an emergency braking command output by the safety controller; transmit the corresponding candidate joint control command to the actuator when it only receives the safety verification signal; and transmit the default emergency braking command and default energy absorption command pre-stored in the safety arbitrator to the actuator when it neither receives the control command output by the safety controller nor the safety verification signal within a preset arbitration period.

9. The intelligent guide dog pulling safety control system according to claim 8, wherein, The adjustable traction assembly includes a traction length adjustment mechanism, a series elastic component, a damping component, an energy absorption component, a driveable direction adjustment anchor point, a traction force sensor, and a traction length encoder. The traction actuator includes a length adjustment drive unit for driving the traction length adjustment mechanism and an anchor point drive unit for changing the position or angle of the direction adjustment anchor point. The energy absorption actuator is used to control the energy absorption component to absorb recoverable elastic stored energy. The safety controller determines the set of permissible traction force vectors based on the recoverable balance domain and the direction of imbalance, and adjusts the position or angle of the anchor point, the relative position of the body and the user, and the tension state of the adjustable traction component according to the direction to determine the set of tension vectors that the adjustable traction component can actually achieve; from the intersection of the set of permissible traction force vectors and the set of tension vectors, a target traction force vector for maintaining or improving the user's balance margin is determined, wherein the traction force in the set of tension vectors acts on the user along the tension direction of the adjustable traction component and has a non-negative amplitude; The safety controller sets the relative speed of the user moving away from the intelligent guide dog along the traction direction to a positive value and the relative speed of approaching the guide dog along the traction direction to a negative value. It performs amplitude-limited integration on the relative speed to obtain the basic compensation amount of traction length for compensating for the relative motion between the user and the machine. Based on the deviation between the target traction force vector and the actual traction force vector and the energy margin determined by the difference between the energy absorption threshold and the current recoverable elastic energy, it determines the traction length correction amount and the anchor point adjustment amount. The basic compensation amount of traction length and the correction amount of traction length are superimposed to form the traction length adjustment command, and the traction direction adjustment command is formed based on the anchor point adjustment amount. The traction length adjustment command and traction direction adjustment command are subject to the traction length range, the upper limit of the traction length adjustment speed, the anchor point adjustment range, the upper limit of the anchor point adjustment speed, the limit of the traction force change rate, the limit of the traction impulse in the direction of imbalance, and the space occupancy limit of the traction component determined by the traction spatiotemporal mechanical domain. Under normal operating conditions, the actual traction force is limited by the lower limit of the normal guide tension. Under emergency recovery conditions, the actual traction force is limited by the lower limit of the emergency guide tension, which is set based on the lower limit of the normal guide tension and is allowed to be zero. Under emergency braking conditions, the lower limit of the guide tension is released, and the execution priority of the traction component release control and energy absorption control is made higher than that of the guide tension holding control.

10. The intelligent guide dog pulling safety control system according to claim 9, wherein, The safety controller has the authority to reject candidate body movement commands; The environmental obstacle detection unit includes a safe ranging sensor, the body state detection unit includes a safe inertial sensor, and the user state detection unit includes at least one of a wearable inertial sensor and a visual sensor; the system also includes a traction actuator drive current detection unit, a series elastic component deformation detection unit, and a mechanical limit release mechanism; When the timestamps of independent safety sensor data are discontinuous, the delay exceeds the threshold, the data exceeds the range, or the multi-source estimation difference of the same state quantity exceeds the consistency threshold, the safety controller determines that the corresponding sensor output is invalid. When the traction force sensor output is invalid, a traction force estimate is generated based on the drive current of the traction actuator and the deformation of the series elastic component. When both the traction force sensor and the traction force estimate are invalid, or the traction length encoder output is invalid, the adjustment of the traction length based on traction force feedback is prohibited, the energy absorption component is activated, and the mechanical travel limit release mechanism is activated within the mechanical travel limit range to control the intelligent guide dog to stop. When the environmental obstacle detection unit outputs an invalid value, the unverified environmental area is set as an obstacle-occupied area, and the vehicle slows down and stops along the verified passable area; if no verified passable area exists, emergency braking is performed and the energy absorption component is activated. When the output of the body state detection unit or the user state detection unit is invalid, the corresponding state uncertainty area is expanded to the preset failure uncertainty area, and the intelligent guide dog is controlled to slow down and stop.

11. The intelligent guide dog pulling safety control system of claim 8, wherein, The safety controller includes a user individual parameter calibration module; Within the traction range determined by the preset initial safety limit, the user individual parameter calibration module forms calibration combinations according to at least two non-collinear traction directions, at least two traction amplitude levels, and at least two gait phases. For each calibration combination, it applies controlled traction to the user and collects the user's center of mass response, trunk tilt response, and capture point changes. The individual traction safety limits for users are determined based on the collected data. These limits include the traction force vector range, the traction force change rate limit, the traction impulse limit in the direction of imbalance, the traction impulse limit in the recovery direction, the trunk tilt angle and its change rate limit, the capture point recovery time, the allowable number of adjustment steps, and the lower limit of the balance margin. The safety controller adjusts the recoverable balance domain according to the individual traction safety limits for users.

12. The intelligent guide dog pulling safety control system of claim 8, wherein, It also includes a body posture adjustment mechanism and a foot support mechanism; The safety controller generates at least two candidate foot target configurations, each configuration including at least one foot target position; under the conditions of satisfying foot kinematics, contact friction cone and joint constraints, it predicts the foot contact position and ground support reaction force corresponding to each configuration, and determines the corresponding equivalent zero moment point based on the predicted traction force vector and its position of action, body inertial force, gravity, ground support reaction force and its position of action. The attitude stability margin is the directed distance from the equivalent zero moment point to the boundary of the foot support area. It is positive when the equivalent zero moment point is within the foot support area and negative when it is outside the foot support area. The safety controller selects the configuration with the largest attitude stability margin from candidate configurations with an attitude stability margin not less than the attitude stability margin threshold, and controls the foot support mechanism to adjust the corresponding foot. If there is no configuration that meets the attitude stability margin threshold, the current foot support state is maintained, the body's center of gravity is lowered, and the intelligent guide dog is controlled to decelerate. When the walking speed is lower than the support transition speed threshold, the body posture adjustment mechanism lowers the body's center of gravity; when the walking speed is lower than the parking speed threshold, the posture stability margin determined according to the current foot support state is not less than the posture stability margin threshold for two consecutive control cycles, and the user capture point is located in the current support area or predicted support area selected according to the gait phase and the distance from the boundary of the selected area is not less than the parking balance margin threshold, the foot support mechanism locks the foot support state.

13. A smart guide dog, characterized in that, The system includes the intelligent guide dog traction safety control system according to any one of claims 8 to 12; or, the intelligent guide dog performs the intelligent guide dog traction safety control method according to any one of claims 1 to 7.