A robot motion control method, system and device based on physical traction

Abstract

The application relates to the technical field of robot control, and discloses a robot motion control method, system and device based on physical traction. The method comprises the following steps: performing channel decoupling control on the force of a traction rope to obtain an axial force control quantity and an angular deflection control quantity; performing online identification control on the two control quantities in a robot zero-intention window to obtain compliance drift compensation parameters; performing synchronous feedback compensation control on the two control quantities according to the compliance drift compensation parameters, screening out a control quantity sequence conforming to a preset coupling control pattern as an intention self-proving control signature; and performing instruction mapping control according to the intention self-proving control signature to obtain a robot motion control instruction. The application improves the accuracy of traction intention extraction and the stability of long-term use of the system.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to a robot motion control method, system and device based on physical traction. Background Technology

[0002] Physical traction-based robot motion control is an important control paradigm in the field of physical human-computer interaction. The operator directly applies a pulling force to the robot body using flexible connectors such as hand-held traction ropes or levers. The robot, on the other hand, uses force and angle sensors mounted on the traction structure to collect the magnitude and direction of the traction force in real time. These signals are then processed by a command mapping algorithm into motion control commands such as forward movement, turning, acceleration, and deceleration, driving the robot to move along the operator's desired trajectory. This control method is widely used in scenarios such as guide quadruped robots, rehabilitation traction robots, and collaborative robotic arms for drag-and-teach demonstrations. Its core advantages lie in its intuitive interaction, bidirectional perception, and inherent distance constraints and safety redundancy characteristics through physical connections.

[0003] However, existing physical traction-based robot motion control methods have significant shortcomings in engineering application. First, most existing solutions embed the signal acquisition logic of force and angle sensors into the robot's native controller, resulting in a deep coupling between the system structure and the specific robot body, making it difficult to migrate and reuse between different robot models. Second, the traction rope simultaneously bears the user's active traction force and the periodic vibration tension induced by the robot's own gait swaying. These two types of signals highly overlap in the frequency domain, and existing solutions, which use a single low-pass filter or impedance control mapping, cannot effectively suppress gait noise while preserving the bandwidth of the intention command. Third, the stiffness and damping characteristics of viscoelastic force transmission elements continuously drift with temperature changes, fatigue accumulation, and load differences. Existing solutions lack an online self-calibration mechanism to address this drift, leading to a degradation of the intention-command mapping relationship over time. Fourth, existing solutions lack an active confirmation mechanism for the user's motion intention. Natural hand tremors, rope weight, or wind disturbances may all be identified as valid commands, posing a risk of false triggering under abnormal operating conditions. Summary of the Invention

[0004] This application provides a robot motion control method, system, and device based on physical traction, which solves the problems of gait noise and active intent mixing causing control quantity distortion, viscoelastic element parameter drift causing decoupling relationship degradation, and lack of active intent confirmation mechanism leading to false triggering in existing physical traction control methods, thereby improving the accuracy of traction intent extraction and the stability of the system in long-term use.

[0005] In a first aspect, this application provides a robot motion control method based on physical traction, the robot motion control method based on physical traction comprising: Step S1: Perform channel-specific decoupling control on the traction rope force to obtain the axial force control quantity and angular deflection control quantity; Step S2: In the robot zero intention window, the axial force control quantity and the angular deflection control quantity are identified and controlled online to obtain the compliance drift compensation parameters; Step S3: Perform synchronous feedback compensation control on the axial force control quantity and the angular deflection control quantity according to the compliance drift compensation parameters. Select the control quantity sequence that is synchronously coupled in time and conforms to the preset coupling control pattern from the two compensated control quantities as the intention self-certification control signature. Step S4: Perform instruction mapping control on the compensated two-way control quantities according to the intention self-certification control signature to obtain robot motion control instructions.

[0006] Secondly, this application provides a physical traction-based robot motion control system, which includes: The control module is used to perform decoupled control of the force on the traction rope in separate channels to obtain the axial force control quantity and the angular deflection control quantity; The identification module is used to perform online identification and control of the axial force control quantity and the angular deflection control quantity within the robot's zero intention window to obtain compliance drift compensation parameters. The compensation module is used to perform synchronous feedback compensation control on the axial force control quantity and the angular deflection control quantity according to the compliance drift compensation parameters, and to select the control quantity sequence that is synchronously coupled in time and conforms to the preset coupling control pattern from the two compensated control quantities as the intention self-certification control signature. The mapping module is used to perform instruction mapping control on the compensated two-way control quantities according to the intention self-certification control signature to obtain robot motion control instructions.

[0007] Thirdly, a physical traction-based robot motion control device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the physical traction-based robot motion control device to execute the aforementioned physical traction-based robot motion control method.

[0008] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned physical traction-based robot motion control method.

[0009] The technical solution provided in this application decouples the force on the traction rope through separate channels, pre-separating the axial force control and angular deflection control at the physical level. This fundamentally breaks the frequency domain aliasing problem caused by a single force transmission channel simultaneously carrying intention signals and gait disturbances in existing technologies. This ensures that the subsequent control quantities naturally possess channel independence, avoiding the cyclical constraints of relying on software filtering to repeatedly balance noise suppression and response bandwidth. The axial force control and angular deflection control are identified and controlled online within the robot's zero-intention window. Using the interference-free period triggered by the robot's own state machine as the identification window, the compliance drift compensation parameter is obtained by subtracting the actual stiffness and damping parameters of the anisotropic compliant base from its nominal parameters. This mechanism enables the system to continuously self-calibrate according to its usage state, overcoming the problem of long-term degradation of the channel decoupling relationship caused by temperature, fatigue, and load changes in viscoelastic elements, ensuring the effectiveness of the separate channel decoupling control throughout its entire life cycle.

[0010] This application performs synchronous feedback compensation control on two control quantities based on the drift compensation parameters. From the compensated control quantities, a sequence of control quantities that is synchronously coupled in time and conforms to a preset coupling control pattern is selected as the intention self-certification control signature. This signature mechanism transfers the intention confirmation problem from the signal processing layer to the interaction protocol layer. It uses the closed coupling trajectory formed by the axial and angular control quantities on the phase plane as a physical token. Due to the strong damping characteristics of the anisotropic compliant base for lateral vibration, the robot's own gait vibration cannot physically reproduce a signal amplitude sufficient to form a closed phase trajectory in the angular channel, thus ensuring the signature's unforgeability and completely eliminating the risk of false triggering caused by the difficulty in distinguishing rhythmic gait disturbance leakage from rhythmic traction intentions mathematically. Finally, based on the intention self-certification control signature, the compensated two control quantities are subjected to instruction mapping control. After dead zone determination, linear gain reduction, and protocol encapsulation, robot motion control commands are issued. The input of each link in the entire control chain comes from the clearly physically defined processing results of the preceding links, resulting in clear data flow and interpretable parameter meanings. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of one embodiment of the robot motion control method based on physical traction in this application. Figure 2This is a schematic diagram of the online identification and convergence process of the recursive least squares identification model within the zero intent window for the axial stiffness and damping parameters of the anisotropic conforming base in the embodiments of this application. Figure 3 This is a schematic diagram showing the distribution of confidence in pattern matching between legitimate intent self-certification signatures and illegitimate traction actions in the embodiments of this application. Detailed Implementation

[0013] This application provides a method, system, and apparatus for robot motion control based on physical traction. The terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0014] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the robot motion control method based on physical traction in this application includes: Step S1: Perform channel-specific decoupling control on the traction rope force to obtain the axial force control quantity and angular deflection control quantity; Specifically, the force on the traction rope during robot walking naturally includes two components: the longitudinal traction force applied by the user along the direction of travel, and the lateral deflection force generated when the user guides the robot to turn. Channel-based decoupling control refers to pre-separating these two components at the physical structural level through an anisotropic compliant base. The anisotropic compliant base arranges high-stiffness, low-damping elastic elements along the traction rope axis and low-stiffness, high-damping viscoelastic elements in two directions perpendicular to the axis. This allows the lateral vibrations generated by the robot's own gait to be preferentially absorbed by the viscoelastic elements, while the axial traction force actively applied by the user is transmitted with high fidelity by the high-stiffness elements, thus achieving physical-level channel isolation before the signal enters the controller. The axial force control quantity reflects the strength of the user's forward intention, and the angular deflection control quantity reflects the direction of the user's turning intention; both are always used as independent channel control inputs in subsequent processing.

[0015] Step S2: In the robot's zero intention window, perform online identification and control of the axial force control quantity and the angular deflection control quantity to obtain the compliance drift compensation parameters; Specifically, the zero-intention window refers to the time period during which the robot is walking without any active traction applied by the user. This is determined by the real-time output of gait phase and body speed states from the robot's native controller. When the gait phase is in a periodic stable phase and the body speed remains constant, if the axial force control and angular deflection control values ​​are both below a preset dead zone threshold, this period is marked as the zero-intention window. The dead zone threshold is set based on the background noise amplitude generated by the anisotropic compliant base under its own weight and the elastic reset of the ropes when no one is operating it. It is typically twice the root mean square value of the background noise obtained from factory no-load calibration to ensure that normal walking vibrations are not misinterpreted as user commands. Within this window, the micro-vibration of the base collected by the inertial measurement unit is used as a known excitation. It forms an excitation-response paired sample set with the two control variables at the same time. The actual stiffness and damping parameters are then calculated by the recursive least squares identification model. The difference between the actual stiffness and damping parameters and the factory nominal values ​​is used to obtain the compliance drift compensation parameters. These parameters quantitatively describe the characteristic offset of the viscoelastic element caused by changes in temperature, fatigue, or load.

[0016] Step S3: Perform synchronous feedback compensation control on the axial force control quantity and angular deflection control quantity according to the compliance drift compensation parameters. Select the control quantity sequence that is synchronously coupled in time and conforms to the preset coupling control pattern from the two compensated control quantities as the intention self-certification control signature. Specifically, the physical essence of the intended self-verifying control signature is that before issuing a continuous movement command, the user actively applies a specific micro-motion coupled axially and angularly to the traction rope. This micro-motion forms a time-synchronized two-dimensional coupled trajectory in the axial force control quantity and the angular deflection control quantity. After projecting the two control quantities as horizontal and vertical coordinates respectively, a closed curve with a definite closure degree, amplitude ratio, and rotation direction is presented on the phase plane, i.e., the coupled phase trajectory pattern. The preset coupled control pattern is a standard trajectory template obtained by collecting and statistically summarizing data through repeated operations by multiple users before the system leaves the factory. Its closure degree characteristic value, amplitude ratio characteristic value, and rotation direction characteristic value are all stored in the controller in interval form. Due to the strong damping characteristics of the anisotropic conforming base to lateral vibration, the signal of the robot's own gait vibration in the angular channel is significantly attenuated, making it physically impossible to reproduce a coupled trajectory with the same degree of closure on the phase plane, thus ensuring the unforgeability of the signature.

[0017] Step S4: Perform instruction mapping control on the compensated two control quantities according to the intention self-certification control signature to obtain the robot motion control command.

[0018] Specifically, the received axial force control and angular deflection control are both compensated control quantities. Unless otherwise specified, the terms "axial force control" and "angular deflection control" refer to the compensated quantities. During the command mapping control stage, the axial force control quantity is compared to the axial dead zone threshold; any portion exceeding the threshold is converted into a linear velocity command component using a linear gain coefficient, determined based on the ratio of the robot's rated maximum linear velocity to the sensor's range. The angular deflection control quantity is subtracted from the preset angular zero-position reference, and the deviation is converted into an angular velocity command component using an angular velocity gain coefficient, determined based on the ratio of the robot's maximum steering angular velocity to the sensor's maximum deflection angle range. Both command components are encapsulated according to the robot's native communication protocol format and sent to the robot's underlying controller. The underlying controller uses the linear velocity and angular velocity command components to perform gait planning and joint torque distribution, driving the robot to move along the user's desired trajectory.

[0019] In one specific embodiment, step S1 includes: By using a tension sensor and an angle sensor installed on the traction base, the force on the traction rope is sampled synchronously to obtain the original traction force sequence and traction deflection angle sequence. Based on the axial stiffness damping parameters and lateral stiffness damping parameters of the anisotropic compliant base, the original traction force sequence and traction deflection angle sequence are subjected to component projection processing of the axial channel and angular channel to obtain the original axial component and the original angular component. Based on the axial stiffness damping parameters and the lateral stiffness damping parameters, an inter-channel crosstalk suppression matrix is ​​constructed. Based on the inter-channel crosstalk suppression matrix, the original axial component and the original angular component are decoupled to obtain the decoupled axial component and the decoupled angular component. The axial component after decoupling is processed by axial control quantity reduction to obtain axial force control quantity, and the angular component after decoupling is processed by angular control quantity reduction to obtain angular deflection control quantity.

[0020] Specifically, the tension sensor and angle sensor synchronously sample the force on the traction rope at a sampling frequency of no less than 200Hz. The 200Hz setting is based on the fact that the highest frequency component of the human body's active traction action does not exceed 10Hz, and the Inexile theorem requires at least 20Hz sampling. However, the rope vibration frequency induced by the robot's gait is usually between 1.5Hz and 5Hz. Considering the timing resolution required for subsequent phase trajectory projection, the sampling frequency is taken as 10 times the aforementioned minimum value to ensure trajectory reconstruction accuracy. The original traction force sequence records the scalar tension value output by the tension sensor at each sampling moment, and the traction deflection angle sequence records the traction rope deflection angle value output by the angle sensor at each sampling moment. Since the axial elastic element and the transverse viscoelastic element of the anisotropic conformal base are not completely orthogonal in space, the resultant force measured by the tension sensor has physical coupling between the axial and angular channels. The component projection processing is based on the spatial attitude of the elastic element described by the axial stiffness damping parameter and the transverse stiffness damping parameter. The resultant force is decomposed by vector projection according to the actual force direction in the axial and angular directions to obtain the original axial component and the original angular component. The axial stiffness damping parameter and the transverse stiffness damping parameter are both taken from the average value obtained by multiple measurements under different temperature and load conditions during factory calibration and are pre-stored in the controller in the form of nominal values.

[0021] The crosstalk suppression matrix between channels is constructed based on the crosstalk coefficient between the two channels. The crosstalk coefficient describes the leakage ratio of the axial channel under force to the original angular component, and the leakage ratio of the angular channel deflection to the original axial component. Both crosstalk coefficients are derived from the factory calibration. Specifically, they are calculated by comparing the ratio of the response amplitude of the original angular component to the original axial component when a pure axial tensile force is applied alone, and the ratio of the response amplitude of the original axial component to the original angular component when a pure angular deflection is applied alone. These two ratios constitute the off-diagonal elements of the crosstalk suppression matrix, while the diagonal elements are all set to 1. After inverting the matrix, multiplying it on the left by the column vector formed by the original axial component and the original angular component completes the decoupling operation, yielding the decoupled axial component and the decoupled angular component. Axial control quantity reduction converts the decoupled axial component from force units to dimensionless control quantity, with the conversion factor being the reciprocal of the sensor's rated range, thus normalizing the control quantity to the 0 to 1 range. Angular control quantity reduction converts the decoupled angular component from angle units to dimensionless control quantity, with the conversion factor being the reciprocal of the angle sensor's maximum deflection angle range, also normalizing it to the -1 to 1 range. This normalization process decouples the subsequent dead zone threshold and gain coefficient settings from specific sensor models, facilitating cross-platform adaptation of different robot models without requiring recalibration of the mapping relationship.

[0022] In one specific embodiment, step S2 includes: Based on the gait phase state quantity and body velocity state quantity output by the robot's native controller, zero intention determination processing is performed on the axial force control quantity and angular deflection control quantity to obtain the zero intention window identifier sequence. By using an inertial measurement unit installed on the traction base, the micro-vibration of the base is synchronously sampled within the time period indicated by the zero intention window identifier sequence to obtain the base excitation signal sequence. The base excitation signal sequence is then time-aligned with the axial force control quantity and angular deflection control quantity during the same period to obtain an excitation-response paired sample set. The excitation-response paired sample set is input into the preset recursive least squares identification model to perform online stiffness and damping identification processing, so as to obtain the current actual axial stiffness and damping parameters and the current actual lateral stiffness and damping parameters. Based on the current actual axial stiffness and damping parameters and the current actual lateral stiffness and damping parameters, the nominal axial stiffness and damping parameters and the nominal lateral stiffness and damping parameters of the anisotropic compliant base are differentially compared to obtain the compliance drift compensation parameters.

[0023] Specifically, the criteria for zero-intent determination are: the gait phase state quantity is in a stable period and the rate of change of the body velocity state quantity is lower than a preset velocity stability threshold; simultaneously, the absolute values ​​of the axial force control quantity and the absolute values ​​of the angular deflection control quantity are both lower than a preset dead zone threshold. The velocity stability threshold is set to 5% per second of the robot's rated walking speed. This percentage is based on the typical duration of the robot's normal acceleration and deceleration transition. A rate of change lower than this is considered to indicate that the body is in a uniform steady state. The dead zone threshold is set to twice the root mean square value of the noise floor output by the tension sensor and angle sensor when the anisotropic compliant base is in an unloaded, unattended state. This multiple ensures that the natural tension fluctuations of the rope caused by normal gait jolts do not trigger false zero-intent determinations. Continuous periods that meet the above three conditions are marked as valid zero-intent windows and written into the zero-intent window identifier sequence in the form of timestamps. The inertial measurement unit is fixed to a rigid part of the traction base. The sampling frequency is consistent with that of the tension sensor and the angle sensor, which is no less than 200Hz. The three signals are acquired in a hardware synchronous triggering mode. The time alignment processing is based on the hardware trigger timestamp. The base excitation signal sequence is matched point by point with the axial force control quantity and angular deflection control quantity of the same period on the time axis. After removing the sampling points with inconsistent timestamps, an excitation-response paired sample set is formed.

[0024] The recursive least squares identification model uses the base excitation signal sequence in the excitation-response paired sample set as the input excitation, and the axial force control quantity and angular deflection control quantity as the system response output. The four parameters of the anisotropic compliant base—axial stiffness, axial damping, lateral stiffness, and lateral damping—are used as the quantities to be identified. A regression relationship between the parameters and the excitation-response is established based on the linear dynamic equation of the spring-damped system. When each new excitation-response paired sample arrives, the recursive least squares algorithm updates the parameter identification results by weighting them according to a forgetting factor based on the parameter identification results of the previous moment and the prediction error of the current sample. The forgetting factor is set to 0.98, which gives higher weight to recent samples, thereby tracking the slow parameter drift of the viscoelastic element caused by temperature rise or fatigue accumulation, while retaining sufficient historical sample weights to suppress the interference of single-sampling noise on the identification results. After identification, the current actual axial stiffness and damping parameters and the current actual lateral stiffness and damping parameters are obtained. The differential comparison process performs algebraic difference calculations on the two parameters and the nominal axial stiffness and damping parameters and nominal lateral stiffness and damping parameters obtained from factory calibration. The difference is the compliance drift compensation parameter, which physically means the offset of the current actual characteristics of the viscoelastic element relative to the factory nominal characteristics. This offset is used in subsequent compensation control to offset the influence of drift on the control quantity in a reverse superposition manner.

[0025] Figure 2 This diagram illustrates the convergence process of the recursive least squares identification model within a zero-intent window for online identification of the axial stiffness and damping parameters of an anisotropic compliant base, as described in this embodiment. The horizontal axis represents the number of recursive iterations, and the vertical axis represents the normalized parameter identification values. The solid line represents the iterative trajectory of the axial stiffness identification values, the dashed line represents the iterative trajectory of the axial damping identification values, the dotted line represents the actual nominal values, and the gray band represents the ±3% convergence zone. Both identification curves converge to the convergence zone within approximately 125 iterations and remain stable, indicating that the recursive least squares identification model, with a forgetting factor of 0.98, can accurately identify the current actual axial stiffness and damping parameters and the current actual lateral stiffness and damping parameters within a finite zero-intent window sample. The difference between the identification result and the nominal value is the source of the compliance drift compensation parameters.

[0026] In one specific embodiment, step S3 involves synchronous feedback compensation control of the axial force control quantity and the angular deflection control quantity based on the compliance drift compensation parameters, including: The compliance drift compensation parameter is decomposed into axial drift compensation component and angular drift compensation component; The axial drift compensation component and the axial force control quantity are differentially superimposed to obtain the compensated axial force control quantity. The angular drift compensation component and the angular deflection control quantity are differentially superimposed to obtain the compensated angular deflection control quantity.

[0027] Specifically, the compliance drift compensation parameters are output from the differential comparison processing in step S2. These parameters contain two independent sets of values: the stiffness damping offset of the axial channel and the stiffness damping offset of the lateral channel. The decomposition process extracts these two sets of offsets according to their channel affiliation. The axial stiffness damping offset corresponds to the axial drift compensation component, and the lateral stiffness damping offset corresponds to the angular drift compensation component. Physically, these two components describe the characteristic offsets of the anisotropic compliant base's axial elastic element and lateral viscoelastic element relative to their factory nominal state, respectively. The decomposition process requires no additional calculations; it only involves separating the parameters according to their channel affiliation. The axial drift compensation component is converted to a dimensionless compensation value using the same dimensional conversion relationship as the axial control quantity in step S1. Similarly, the angular drift compensation component is converted to a dimensionless compensation value. The upper limit of the absolute value of both compensation values ​​is clamped at 0.1. This clamping value is set based on the maximum possible drift amplitude of the viscoelastic element within its normal service life. Exceeding this range triggers a sensor malfunction flag, prompting maintenance personnel to recalibrate.

[0028] In differential superposition processing, the axial drift compensation component is superimposed on the axial force control quantity with a negative sign. That is, the compensated axial force control quantity equals the axial force control quantity minus the axial drift compensation component. The physical meaning of the negative sign is: when the stiffness of the viscoelastic element decreases due to aging, the control quantity generated in the axial channel by the same user's traction force is too large. The compensation component is positive, and after superposition with a negative sign, the excessive control quantity is pulled back to the true level corresponding to the nominal stiffness. The angular deflection control quantity is treated similarly. The angular drift compensation component is superimposed with a negative sign to eliminate the influence of the lateral viscoelastic element characteristic drift on the angular channel control quantity. After compensation, both control quantities inherit the normalized range constraint. The axial direction remains in the range of 0 to 1, and the angular direction remains in the range of -1 to 1. The compensation result exceeding the range is output after saturation truncation processing to ensure that the control quantity received by subsequent phase trajectory projection and command mapping is always within the effective range.

[0029] In one specific embodiment, step S3 involves filtering out a sequence of control quantities that are synchronously coupled in time from the compensated two control quantities, including: The compensated axial force control quantity and the compensated angular deflection control quantity are processed frame by frame using a preset sliding window to obtain the axial window sequence and the angular window sequence. Cross-correlation operations are performed on the axial window sequence and the angular window sequence to obtain the time-series cross-correlation coefficient; The temporal cross-correlation coefficient is compared with the preset synchronization coupling threshold, and the frame segments whose temporal cross-correlation coefficient exceeds the preset synchronization coupling threshold are retained to obtain the temporal synchronization coupling control quantity sequence.

[0030] Specifically, the preset sliding window length is set to the number of sampling points corresponding to 0.5 seconds, i.e., 100 sampling points at a sampling frequency of 200Hz. This length is determined based on the shortest duration of the micro-movement intended to be verified. The 0.4 seconds is taken from the lower bound of the complete cycle distribution of axial and angular coupled micro-movements measured in the aforementioned calibration experiment with multiple users, corresponding to the shortest neuromuscular duration required for the wrist and arm to complete one round-trip coupled movement. The 0.5-second window length ensures that at least one complete micro-movement cycle is included. In view of the situation where the movement speed of different users varies greatly and the fixed window may truncate slow movements, the frame-by-frame truncation is based on a 90% overlapping sliding window. Multiple sliding windows of different lengths are used in parallel for cross-correlation calculations on the same frame segment. The truncation result corresponding to the window length with the highest peak cross-correlation coefficient is taken as the temporal synchronization coupling control sequence of that frame segment. Thus, even when the movement speed changes, a complete micro-movement cycle can still be included without being truncated or mixed with unnecessary noise. The sliding window step size is set to 0.05 seconds, corresponding to 10 sampling points. The step size is one-tenth of the window length, ensuring 90% overlap between adjacent frames and guaranteeing that the start and end boundaries of micro-motions are not truncated or missed due to frame boundary alignment issues. Frame-by-frame truncation processing moves the window progressively according to the step size. Each time, all sampling points of the compensated axial force control and compensated angular deflection control within the current window are captured, forming axial window sequences and angular window sequences respectively. The two sequences strictly correspond in time to sampling points within the same window range.

[0031] The cross-correlation operation calculates the normalized cross-correlation coefficient for the axial and angular window sequences at zero delay. The normalized cross-correlation coefficient is calculated by summing the pointwise products of the two sequences and dividing by the product of their respective root mean square values. The result falls within the range of -1 to 1. The closer the absolute value is to 1, the higher the temporal synchronization coupling between the two sequences. Positive values ​​indicate co-directional coupling, and negative values ​​indicate anti-directional coupling. The preset synchronization coupling threshold is 0.85. This value is determined based on the lower quartile of the temporal cross-correlation coefficient distribution obtained from measurements of multiple users applying legitimate intent to prove their micro-actions. Frames with a cross-correlation coefficient below 0.85 correspond to situations where the two control variables are not synchronously coupled, belonging to single-channel random fluctuations or residual gait noise, and are therefore discarded. Frames with a time-series cross-correlation coefficient exceeding 0.85 are retained. The sampling point series of the compensated axial force control quantity and the sampling point series of the compensated angular deflection control quantity in the frame segment together constitute the time-series synchronous coupling control quantity sequence. The two data maintain a point-by-point time correspondence in the sequence for use in phase trajectory projection processing.

[0032] In one specific embodiment, step S3 involves selecting a sequence of control quantities that conforms to a preset coupling control pattern as the intention self-verifying control signature, including: Using the compensated axial force control quantity in the time-synchronous coupling control quantity sequence as the horizontal axis and the compensated angular deflection control quantity as the vertical axis, a two-dimensional phase trajectory projection process is performed to obtain the coupled phase trajectory pattern. The closure feature value, amplitude ratio feature value, and rotation direction feature value are extracted from the coupled phase trajectory pattern. The closure feature value, amplitude ratio feature value, and rotation direction feature value are then matched with the preset coupled control pattern to obtain the pattern matching confidence level. The sequence of time-synchronized coupled control quantities whose pattern matching confidence exceeds a preset confidence threshold is identified as the intention self-certifying control signature.

[0033] Specifically, the two-dimensional phase trajectory projection processing uses the compensated axial force control quantity at each sampling moment in the time-series synchronous coupling control quantity sequence as the abscissa component on the phase plane at that moment, and the compensated angular deflection control quantity as the ordinate component. All sampling points are plotted sequentially on the two-dimensional plane in chronological order to form a continuous trajectory curve, i.e., the coupled phase trajectory pattern. The closure feature value describes the ratio of the distance between the first and last endpoints of the trajectory curve to the total arc length of the trajectory. The smaller the ratio, the closer the trajectory is to a closed curve. The legitimate intent self-validation micro-action requires the user to complete a complete round-trip action of first pulling and then releasing in the axial direction and first deflecting and then returning in the angular direction. This action will inevitably form a near-closed trajectory on the phase plane. The upper limit of the preset legitimate range of the closure feature value is 0.15, that is, the distance between the first and last endpoints does not exceed 15% of the total arc length. The amplitude ratio characteristic is the ratio of the range of the trajectory along the vertical axis to the range along the horizontal axis, reflecting the relative magnitude of the angular deflection amplitude and the axial tension amplitude. The preset legal range is 0.3 to 0.7, which is determined based on the 95% confidence interval of the amplitude ratio distribution of legal micro-motions in calibration experiments conducted by multiple users. The rotation direction characteristic is determined by calculating the sign of the sum of the directed areas formed by three adjacent points on the trajectory curve. A positive value corresponds to counterclockwise rotation, and a negative value corresponds to clockwise rotation. The preset coupling control pattern allows both rotation directions to be considered legal to accommodate differences in user hand habits.

[0034] The preset coupling control pattern is stored in the controller as a valid interval of three feature values. The interval boundaries are determined before leaving the factory by statistically analyzing the feature value distribution obtained from at least 20 users each performing the operation 50 times. The boundaries of the valid intervals are not taken from a single user, but are set according to the dispersion of the feature value distribution of the above multiple users. The upper and lower limits of the intervals for closure and amplitude ratio are taken as the mean of their respective distributions plus or minus two standard deviations, so that the feature values ​​corresponding to users of different heights and strengths fall into the same interval. The rotation direction feature value allows both clockwise and counterclockwise values ​​to accommodate the differences in operating habits between left-handed and right-handed users, thereby absorbing the influence of individual differences on the feature value distribution at the group level. Feature matching processing compares the three feature values ​​extracted from the current coupling phase trajectory pattern with the corresponding valid intervals. A score of 0.4 is assigned to the closure feature value falling into a valid interval, a score of 0.4 is assigned to the amplitude ratio feature value falling into a valid interval, and a score of 0.2 is assigned to the rotation direction feature value conforming to the preset direction. The sum of the three scores is the pattern matching confidence score, with a maximum score of 1.0. The allocation of the three scoring weights is based on the fact that closure degree and amplitude ratio contribute more to distinguishing between legitimate micro-movements and random disturbances, while rotation direction is more affected by individual differences and therefore has a lower weight. Specifically, the scoring ratios of 0.4, 0.4, and 0.2 are determined based on the ranking of the separability of the three feature values ​​for legitimate micro-movements and random disturbances in the calibration samples. The inter-class separation of closure degree and amplitude ratio is significantly higher than that of rotation direction. Rotation direction has the greatest inter-class overlap due to differences in user hand habits, so it is given the lowest weight. The preset confidence threshold is 0.8, which requires that both closure degree and amplitude ratio must be satisfied simultaneously, and the rotation direction can be inconsistent, or one of closure degree and amplitude ratio must be satisfied and the rotation direction must be satisfied. The timing synchronous coupling control quantity sequence with a pattern matching confidence of 0.8 is determined as the intention self-certification control signature, triggering the activation of the subsequent instruction receiving state.

[0035] Figure 3This diagram illustrates the distribution of pattern matching confidence levels between legitimate intent self-certification signatures and illegitimate traction actions in an embodiment of this application. The horizontal axis represents pattern matching confidence level, the vertical axis represents the number of samples, the light-colored diagonally filled bars represent the confidence level distribution of illegitimate traction actions, the dark-colored reverse diagonally filled bars represent the confidence level distribution of legitimate intent self-certification signatures, and the vertical dashed line represents a preset confidence threshold of 0.80. The confidence levels of illegitimate traction actions are mainly distributed below 0.80, while the confidence levels of legitimate intent self-certification signatures are mainly distributed above 0.80. There is some overlap between the two distributions near the threshold, and the overlapping area corresponds to a small number of misjudged samples. The preset confidence threshold of 0.80 is determined by balancing the false rejection rate and the false acceptance rate within the overlapping area. This keeps the proportion of illegal actions being misjudged as legitimate signatures and the proportion of legitimate signatures being missed at an acceptable low level. This verifies that the judgment rule of calculating the pattern matching confidence level by weighting the three features of closure degree feature value, amplitude ratio feature value and rotation direction feature value, and using 0.80 as the preset confidence threshold to screen intentionally self-verifying control signatures has reliable discrimination ability.

[0036] In one specific embodiment, step S4 includes: Based on the intention self-certification control signature, the compensated axial force control quantity and the compensated angular deflection control quantity are activated by the instruction receiving state to obtain the activated axial force control quantity and the activated angular deflection control quantity. The activated axial force control quantity is compared with the axial dead zone threshold. The activated axial force control quantity that exceeds the axial dead zone threshold is reduced by linear gain to obtain the linear velocity command component. After activation, the angular deflection control quantity is subjected to angular zero-position deviation extraction processing. The angular deflection control quantity after activation is subjected to differential operation with the preset angular zero-position reference to obtain the angular velocity command component. The linear velocity command component and the angular velocity command component are encapsulated using a protocol to obtain the robot motion control command.

[0037] Specifically, the instruction receiving state activation process starts from the detection time of the intention self-certification control signature, switching the controller state to the continuous instruction receiving state. The duration of the activation state is set to a cumulative 10 seconds from the signature detection time. If the axial force control quantity remains below the stationary dead zone threshold for more than 3 seconds within 10 seconds, the activation state is exited early. This dual exit mechanism ensures that the robot does not continuously respond to residual control quantities after the user releases the traction. During the activation state, the compensated axial force control quantity and the compensated angular deflection control quantity are directly output as the activated axial force control quantity and activated angular deflection control quantity. After exiting the activation state, both control quantities are forcibly reset to zero, and the linear velocity and angular velocity components in the robot motion control command are both set to zero. The axial dead zone threshold is set to 0.05, which corresponds to 5% of the force sensor's range. This value is set based on the maximum amplitude of the residual noise in the axial channel under normal walking vibration of the anisotropic compliant base. Activation of axial force control below this threshold is considered invalid traction and is directly truncated to zero without entering linear gain calculation, preventing the user's natural hand tremors from being misinterpreted as forward commands. The portion exceeding the axial dead zone threshold is converted into a linear velocity command component through linear gain calculation. The linear gain coefficient is the ratio of the robot's rated maximum linear velocity to the full scale of the normalized control quantity, i.e., rated maximum linear velocity divided by 0.95 (full scale 1.0 minus dead zone threshold 0.05), in meters per second, ensuring that the robot reaches its rated maximum linear velocity precisely when the user pulls to the sensor's full scale.

[0038] The preset angular zero-position reference is the output value of the angle sensor when the traction rope is in a natural suspension state. This value is automatically collected and stored during the static calibration phase after each system power-on. The calibration period is 2 seconds, during which the average value is continuously sampled to eliminate the influence of mechanical vibration at the moment of power-on. After activation, the angular deflection control quantity is differentially calculated with the preset angular zero-position reference to obtain the angular deviation quantity. The angular deviation quantity is then converted into the angular velocity command component by the angular velocity gain coefficient. The angular velocity gain coefficient is the ratio of the robot's rated maximum steering angular velocity to the normalized value of the angle sensor's maximum deflection angle range, in radians per second. Positive values ​​correspond to left turns, and negative values ​​correspond to right turns, consistent with the installation direction of the angle sensor. The linear velocity command component and the angular velocity command component are encapsulated according to the robot's native communication protocol format. The encapsulation content includes a velocity command type identifier byte, a linear velocity value field, an angular velocity value field, and a check byte. The encapsulated data frame is sent to the robot's underlying controller via the CAN bus or serial port at a frequency of not less than 50Hz. The sending frequency is one-quarter of the 200Hz sampling frequency to ensure the real-time performance of the command while matching the robot's underlying gait control cycle.

[0039] The above describes the robot motion control method based on physical traction in the embodiments of this application. The following describes the robot motion control system based on physical traction in the embodiments of this application. One embodiment of the robot motion control system based on physical traction in the embodiments of this application includes: The control module is used to perform decoupled control of the force on the traction rope in separate channels to obtain the axial force control quantity and the angular deflection control quantity; The identification module is used to perform online identification and control of the axial force control quantity and the angular deflection control quantity within the robot's zero intention window to obtain compliance drift compensation parameters. The compensation module is used to perform synchronous feedback compensation control on the axial force control quantity and the angular deflection control quantity according to the compliance drift compensation parameters, and to select the control quantity sequence that is synchronously coupled in time and conforms to the preset coupling control pattern from the two compensated control quantities as the intention self-certification control signature. The mapping module is used to perform instruction mapping control on the compensated two-way control quantities according to the intention self-certification control signature to obtain robot motion control instructions.

[0040] This invention also provides a physical traction-based robot motion control device, which can be a server. The physical traction-based robot motion control device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the physical traction-based robot motion control device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the physical traction-based robot motion control device stores the data corresponding to this embodiment. The network interface of the physical traction-based robot motion control device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0041] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the physical traction-based robot motion control method.

[0042] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0043] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a physical traction-based robot motion control device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A robot motion control method based on physical traction, characterized in that, The method includes: Step S1: Perform channel-specific decoupling control on the traction rope force to obtain the axial force control quantity and angular deflection control quantity; Step S2: In the robot zero intention window, the axial force control quantity and the angular deflection control quantity are identified and controlled online to obtain the compliance drift compensation parameters; Step S3: Perform synchronous feedback compensation control on the axial force control quantity and the angular deflection control quantity according to the compliance drift compensation parameters. Select the control quantity sequence that is synchronously coupled in time and conforms to the preset coupling control pattern from the two compensated control quantities as the intention self-certification control signature. Step S4: Perform instruction mapping control on the compensated two-way control quantities according to the intention self-certification control signature to obtain robot motion control instructions.

2. The robot motion control method based on physical traction according to claim 1, characterized in that, Step S1 includes: By using a tension sensor and an angle sensor installed on the traction base, the force on the traction rope is sampled synchronously to obtain the original traction force sequence and traction deflection angle sequence. Based on the axial stiffness damping parameters and lateral stiffness damping parameters of the anisotropic compliant base, the original traction force sequence and the original traction deflection angle sequence are subjected to component projection processing of the axial channel and the angular channel to obtain the original axial component and the original angular component. Based on the axial stiffness damping parameter and the lateral stiffness damping parameter, an inter-channel crosstalk suppression matrix is ​​constructed. Based on the inter-channel crosstalk suppression matrix, the original axial component and the original angular component are decoupled to obtain the decoupled axial component and the decoupled angular component. The axial component after decoupling is processed by axial control quantity reduction to obtain the axial force control quantity, and the angular component after decoupling is processed by angular control quantity reduction to obtain the angular deflection control quantity.

3. The robot motion control method based on physical traction according to claim 1, characterized in that, Step S2 includes: Based on the gait phase state quantity and body velocity state quantity output by the robot's native controller, the axial force control quantity and the angular deflection control quantity are subjected to zero intent determination processing to obtain a zero intent window identifier sequence. By using an inertial measurement unit installed on the traction base, the micro-vibration of the base is synchronously sampled within the time period indicated by the zero intention window identifier sequence to obtain the base excitation signal sequence. The base excitation signal sequence is then time-aligned with the axial force control quantity and the angular deflection control quantity during the same period to obtain an excitation-response paired sample set. The excitation-response paired sample set is input into a preset recursive least squares identification model for online stiffness and damping identification to obtain the current axial actual stiffness and damping parameters and the current lateral actual stiffness and damping parameters. Based on the current actual axial stiffness damping parameter and the current actual lateral stiffness damping parameter, the nominal axial stiffness damping parameter and the nominal lateral stiffness damping parameter of the anisotropic compliant base are differentially compared to obtain the compliant drift compensation parameter.

4. The robot motion control method based on physical traction according to claim 1, characterized in that, Step S3 involves synchronous feedback compensation control of the axial force control quantity and the angular deflection control quantity based on the compliance drift compensation parameters, including: The compliance drift compensation parameter is decomposed into axial drift compensation component and angular drift compensation component; The axial drift compensation component and the axial force control quantity are differentially superimposed to obtain the compensated axial force control quantity. The angular drift compensation component and the angular deflection control quantity are differentially superimposed to obtain the compensated angular deflection control quantity.

5. The robot motion control method based on physical traction according to claim 4, characterized in that, Step S3 involves selecting a sequence of control quantities that are synchronously coupled in time from the compensated two control quantities, including: The compensated axial force control amount and the compensated angular deflection control amount are processed frame by frame using a preset sliding window to obtain an axial window sequence and an angular window sequence. Perform cross-correlation operations on the axial window sequence and the angular window sequence to obtain the time-series cross-correlation coefficient; The temporal cross-correlation coefficient is compared with a preset synchronization coupling threshold, and the frame segments whose temporal cross-correlation coefficient exceeds the preset synchronization coupling threshold are retained to obtain the temporal synchronization coupling control quantity sequence.

6. The robot motion control method based on physical traction according to claim 5, characterized in that, Step S3 involves selecting control quantity sequences that conform to a preset coupling control pattern as intentional self-verifying control signatures, including: Using the compensated axial force control quantity in the time-series synchronous coupling control quantity sequence as the horizontal axis and the compensated angular deflection control quantity as the vertical axis, a two-dimensional phase trajectory projection process is performed to obtain a coupled phase trajectory pattern. The closure feature value, amplitude ratio feature value, and rotation direction feature value are extracted from the coupled phase trajectory pattern. The closure feature value, amplitude ratio feature value, and rotation direction feature value are then matched with a preset coupling control pattern to obtain the pattern matching confidence level. The sequence of time-synchronized coupling control quantities whose pattern matching confidence exceeds a preset confidence threshold is determined as the intention self-certification control signature.

7. The robot motion control method based on physical traction according to claim 1, characterized in that, Step S4 includes: Based on the intent self-certification control signature, the compensated two control quantities are activated in the instruction receiving state to obtain the activated axial force control quantity and the activated angular deflection control quantity. The activated axial force control quantity is subjected to axial dead zone threshold comparison processing. The activated axial force control quantity that exceeds the axial dead zone threshold is subjected to linear gain reduction processing to obtain the linear velocity command component. The activated angular deflection control quantity is subjected to angular zero-position deviation extraction processing, and the activated angular deflection control quantity is subjected to differential operation with the preset angular zero-position reference to obtain the angular velocity command component. The linear velocity command component and the angular velocity command component are encapsulated using a protocol to obtain the robot motion control command.

8. A robot motion control system based on physical traction, characterized in that, For implementing the physical traction-based robot motion control method as described in any one of claims 1-7, the physical traction-based robot motion control system comprises: The control module is used to perform decoupled control of the force on the traction rope in separate channels to obtain the axial force control quantity and the angular deflection control quantity; The identification module is used to perform online identification and control of the axial force control quantity and the angular deflection control quantity within the robot's zero intention window to obtain compliance drift compensation parameters. The compensation module is used to perform synchronous feedback compensation control on the axial force control quantity and the angular deflection control quantity according to the compliance drift compensation parameters, and to select the control quantity sequence that is synchronously coupled in time and conforms to the preset coupling control pattern from the two compensated control quantities as the intention self-certification control signature. The mapping module is used to perform instruction mapping control on the compensated two-way control quantities according to the intention self-certification control signature to obtain robot motion control instructions.

9. A robot motion control device based on physical traction, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the physical traction-based robot motion control method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the physical traction-based robot motion control method as described in any one of claims 1 to 7.

Images