Real-time evaluation method and device for coordination between humanoid robot trunk and limb end

Abstract

The application provides a real-time evaluation method and device for coordination between a humanoid robot trunk and a limb end, comprising: collecting a lumbar lordosis angle signal and a shoulder joint forward extension angle signal, obtaining a lumbar lordosis angle starting time stamp, a shoulder joint forward extension angle starting time stamp, a real-time pressure ratio of a front foot sole and a rear foot sole, and an arm target stretching distance set for a current movement task; using a Kalman filtering algorithm to process the lumbar lordosis angle signal and the shoulder joint forward extension angle signal, and obtaining an accurate leading value between smoothed lumbar lordosis starting time and shoulder joint forward extension starting time after denoising; according to an evaluation result of a fall prevention reflex trigger risk, analyzing a pressure ratio rise when entering a rapid rise interval and the arm target stretching distance, and calculating an adjusted leading value upper limit.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and apparatus for real-time evaluation of the coordination between the torso and limbs of a humanoid robot. Background Technology

[0002] In the field of humanoid robot research, real-time assessment of whole-body joint coordination is a key technology to ensure smooth and safe robot movement. The importance of this field is self-evident, as it directly relates to the robot's stability and task execution efficiency in complex environments, especially in scenarios requiring fine manipulation and dynamic balance, such as domestic service or industrial collaboration, where joint coordination becomes a core indicator of robot performance. However, current research and application methods often fail to comprehensively consider the real-time interactions between different body parts when facing dynamic environments. Many solutions focus more on optimizing single joints or local movements, neglecting the impact of whole-body posture changes on overall balance, especially in complex movements involving center of gravity adjustments and load changes, lacking in-depth consideration of the multi-part linkage effects. This limitation leads to problems such as unsmooth movements or insufficient stability when robots perform tasks. The robot's torso (especially the lumbar / pelvic region) pre-adjusts its posture before the arm rapidly extends forward to counteract inertial torque and maintain balance. The time difference between the initiation of lumbar forward tilting and the initiation of shoulder extension, known as the lead time, has a dual impact on the smoothness of hand movement trajectory and standing stability. Lumbar forward tilting is used to adjust the torso's center of gravity in advance, providing support for arm extension, while shoulder extension directly determines the range and speed of arm movement. If the lumbar forward tilt is too far ahead, although theoretically it can make arm movement more stable, in practice it will cause drastic changes in foot pressure distribution, especially a rapid increase in pressure on the forefoot, which may cause the robot to misjudge a fall risk and thus interrupt the movement. This contradiction between the time difference and pressure distribution has become a pressing problem to be solved in coordination assessment. Specifically, when a robot performs a simple object retrieval task, the lumbar spine needs to tilt forward in advance to stabilize the center of gravity, and then the shoulder joint leads the arm to extend towards the target object. If the lumbar forward tilt occurs too early, the pressure on the forefoot will increase rapidly, triggering the protection mechanism, causing the arm movement to stop and the task to fail. In this case, the setting of the time difference directly affects the task completion rate and the robot's stability. Therefore, how to reasonably control the time difference between lumbar forward tilting and shoulder extension in dynamic tasks, while avoiding the false triggering of the protection mechanism caused by the imbalance of foot pressure distribution, has become a key issue in improving the overall joint coordination assessment of humanoid robots. Summary of the Invention

[0003] This invention provides a real-time evaluation method for the coordination of the torso and extremities of a humanoid robot, mainly including: Collect lumbar lordosis angle signal and shoulder extension angle signal, and obtain lumbar lordosis angle start time stamp, shoulder extension angle start time stamp, real-time pressure ratio of forefoot and heel, and the target arm extension distance set for the current exercise task. The Kalman filter algorithm is used to process the lumbar lordosis angle signal and the shoulder extension angle signal, and after noise reduction, the smoothed lumbar lordosis initiation time and the shoulder extension initiation time are obtained as a precise lead value. The assessment determines whether the precision advance value exceeds the preset initial upper limit threshold. If it does, the analysis is based on the real-time pressure ratio between the forefoot and heel to determine the trend of the pressure ratio increase between the forefoot and heel, identify the critical point where the current pressure ratio enters the rapid increase zone, and obtain the assessment result of the risk of triggering the fall reflex. If it does not exceed, the current precision advance value is marked as normal and the assessment proceeds directly to the smoothness of the hand end trajectory. Based on the assessment results of the risk of fall reflex triggering, the pressure ratio and the target arm extension distance when entering the rapid rise zone are analyzed, and the adjusted allowable upper limit of the advance amount is calculated. By assessing whether the current lead value exceeds the adjusted upper limit of the lead, the necessity of interrupting the arm extension movement is determined. If no interruption is required, the smoothness of the mid-range velocity of the hand end trajectory is assessed to obtain the assessment result of the smoothness of the hand end trajectory. If an interruption is required, the movement is returned to the previous safe posture, and the interruption status and interruption reason are recorded. Based on the evaluation results of the smoothness of the hand end trajectory and the conclusion of the necessity of interrupting the arm extension movement, the coordination evaluation report of the humanoid robot's torso and limbs is generated by integrating the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, and the real-time pressure ratio of the forefoot and heel.

[0004] Furthermore, lumbar lordosis angle signals and shoulder extension angle signals are collected to obtain the lumbar lordosis angle initiation time stamp, shoulder extension angle initiation time stamp, real-time pressure ratio of the forefoot and heel, and the target arm extension distance set for the current exercise task, including: The system collects lumbar lordosis angle signals during the lumbar lordosis process. Based on the gradient change of the angular velocity signal, it identifies the starting moment when the lumbar lordosis angle changes from a static state to a dynamic state, and extracts the lumbar lordosis angle initiation timestamp corresponding to this starting moment. Simultaneously, it collects shoulder joint extension angle signals, identifies the starting moment when the shoulder joint extension angle changes from a static state to a dynamic state, and extracts the shoulder joint extension angle initiation timestamp corresponding to this starting moment. Based on the lumbar lordosis angle initiation timestamp and the shoulder joint extension angle initiation timestamp, it subtracts the lumbar lordosis angle initiation timestamp from the shoulder joint extension angle initiation timestamp to obtain the original advance value. At the same time, it collects the pressure values ​​output by the forefoot pressure sensor and the rearfoot pressure sensor through a plantar pressure array, and divides the forefoot pressure value by the rearfoot pressure value to obtain the real-time pressure ratio of the forefoot and rearfoot. Finally, it collects the preset target arm extension distance in the current movement task command.

[0005] Furthermore, the Kalman filter algorithm is used to process the lumbar lordosis angle signal and the shoulder extension angle signal, and after denoising, a smoothed, precise lead value between the lumbar lordosis initiation time and the shoulder extension initiation time is obtained, including: The angle value sequences of lumbar lordosis angle and shoulder extension angle are extracted separately within the sampling period. For the high-frequency noise components present in the angle value sequences, a Kalman filter algorithm state vector is constructed. This state vector contains two state components: the current angle value and the angular velocity value. A state transition matrix is ​​established based on the sampling interval of the inertial measurement unit, describing the recursive relationship between angle and angular velocity values ​​between adjacent sampling moments. Based on the state transition matrix, the process noise covariance and observation noise covariance are set, and the prior angle estimate for the current moment is calculated through the prediction phase of the Kalman filter algorithm. Based on the prior angle estimate and the current angle observation value, the Kalman filter is executed. In the update phase of the wavelet algorithm, the Kalman gain is calculated and the prior angle estimate is corrected to obtain the posterior angle estimate. Filtering is then performed on the lumbar lordosis angle signal and the shoulder extension angle signal to obtain smoothed lumbar lordosis angle sequences and smoothed shoulder extension angle sequences. Based on the smoothed lumbar lordosis angle sequence, the moment when the angle value jumps from a static state to a dynamic state is identified as the smoothed lumbar lordosis initiation moment. Based on the smoothed shoulder extension angle sequence, the moment when the angle value jumps from a static state to a dynamic state is identified as the smoothed shoulder extension initiation moment. The smoothed lumbar lordosis initiation moment is subtracted from the smoothed shoulder extension initiation moment to obtain the precise lead value.

[0006] Furthermore, the assessment determines whether the precision advance value exceeds a preset initial upper limit threshold. If it does, the analysis of the real-time pressure ratio between the forefoot and heel is used to determine the trend of the pressure ratio increase, identifying the critical point where the current pressure ratio enters the rapid increase zone, and obtaining the assessment result of the risk of triggering the fall reflex. If it does not exceed the threshold, the current precision advance value is marked as within the normal range, and the assessment of the smoothness of the hand's end-effector trajectory is directly initiated, including: The precise lead value is compared with a preset initial upper limit threshold. If the precise lead value does not exceed the initial upper limit threshold, the current precise lead value is marked as normal and the process directly proceeds to the hand end-effector trajectory smoothness evaluation process. If the precise lead value exceeds the initial upper limit threshold, the real-time pressure ratio sequence of the forefoot and heel within the current sampling period is extracted, and the pressure ratio change trend judgment process is initiated. Based on the real-time pressure ratio sequence of the forefoot and heel, the difference in pressure ratio between adjacent sampling times is calculated to obtain the pressure ratio rise rate. The pressure ratio rise rate is compared with a preset rapid rise rate threshold. When the pressure ratio rise rate exceeds the rapid rise rate threshold, the process is initiated. When the rate threshold is reached, the current pressure ratio is determined to have entered a rapid increase range. The sampling time when it first enters the rapid increase range is recorded as the rapid increase critical point, and the time length from the rapid increase critical point to the moment when the pressure ratio increase rate drops below the rapid increase rate threshold is recorded as the duration of the rapid increase range. Based on the pressure ratio value corresponding to the rapid increase critical point and the duration of the rapid increase range, the risk level of the fall reflex triggered by the rapid increase in forefoot pressure is determined. If the duration of the rapid increase range exceeds a preset duration threshold, the assessment result of the fall reflex trigger risk is output as high risk; if the duration of the rapid increase range does not exceed the duration threshold, the assessment result of the fall reflex trigger risk is output as low risk.

[0007] Furthermore, based on the assessment results of the risk triggered by the fall reflex, the pressure ratio and the target arm extension distance when entering the rapid rise zone are analyzed to calculate the adjusted upper limit of the allowable advance, including: Based on the risk assessment results of the fall reflex triggering, if the assessment result is high risk, the pressure rise rate value corresponding to the entry into the rapid rise zone is extracted as the critical pressure rise rate. Simultaneously, the target arm extension distance in the current exercise task is obtained, and a numerical correspondence table between the critical pressure rise rate and the target arm extension distance is established to obtain the mapping relationship between pressure rise rate and extension distance. Based on the mapping relationship between pressure rise rate and extension distance, the target arm extension distance is compared with a preset extension distance benchmark value. If the target arm extension distance is greater than the benchmark value, the benchmark value is divided by the target arm extension distance to obtain the advance amount adjustment coefficient. Based on the advance amount adjustment coefficient and the critical pressure rise rate, the initial upper limit threshold is multiplied by the advance amount adjustment coefficient to obtain a scaled threshold. The critical pressure rise rate is multiplied by a preset pressure correction coefficient to obtain a pressure correction amount. The scaled threshold is subtracted from the pressure correction amount to obtain the adjusted allowable upper limit of the advance amount.

[0008] This invention provides a real-time assessment device for the coordination of the torso and extremities of a humanoid robot, mainly comprising: The signal acquisition module is used to acquire the lumbar spine forward tilt angle signal and the shoulder joint forward extension angle signal, and obtain the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, the real-time pressure ratio of the forefoot and the heel, and the target arm extension distance set for the current exercise task. The signal processing module is used to process the lumbar lordosis angle signal and the shoulder extension angle signal using the Kalman filter algorithm, and after noise reduction, obtain the smoothed accurate lead value between the lumbar lordosis initiation time and the shoulder extension initiation time. The advanced measurement assessment module is used to assess whether the precise advanced measurement value exceeds the preset initial upper limit threshold. If it does, it analyzes the trend of the pressure ratio increase between the forefoot and heel based on the real-time pressure ratio between the forefoot and heel, identifies the critical point where the current pressure ratio enters the rapid increase zone, and obtains the assessment result of the risk of triggering the fall reflex. If it does not exceed, the current precise advanced measurement value is marked as normal range and directly enters the assessment of the smoothness of the hand end trajectory. The upper limit adjustment module is used to analyze the pressure ratio and the target arm extension distance when entering the rapid rise zone based on the assessment results of the risk of fall reflex triggering, and calculate the adjusted upper limit of the advance amount; The interruption judgment module is used to determine the necessity of interrupting the arm extension movement by evaluating whether the current advance value exceeds the adjusted upper limit of the advance. If no interruption is required, the smoothness of the mid-segment velocity of the hand end trajectory is evaluated to obtain the evaluation result of the smoothness of the hand end trajectory. If an interruption is required, the system returns to the previous safe posture and records the interruption status and interruption reason. The report generation module is used to generate a coordination assessment report of the humanoid robot's torso and limbs based on the evaluation results of the smoothness of the hand end trajectory and the conclusion of the necessity of interrupting the arm extension movement, by integrating the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, and the real-time pressure ratio of the forefoot and heel.

[0009] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a real-time assessment method and device for the coordination of the torso and extremities of a humanoid robot. By deploying inertial measurement units and plantar pressure arrays on the robot's torso and lower limbs, data such as lumbar lordosis angle, shoulder extension angle, and plantar pressure ratio are collected. A complete solution is proposed to comprehensively assess the coordination of torso and extremity movements, fall prevention risk, and the smoothness of hand trajectories. First, this invention utilizes a Kalman filter algorithm to denoise the angle signals, accurately calculates the lead of torso and upper limb movements, and assesses fall prevention risk by combining the trend of pressure ratio changes. The upper limit of the lead is dynamically adjusted to determine the necessity of interrupting arm movements. Finally, multi-source data is fused to generate a full-body joint coordination assessment report. Through real-time data processing and dynamic threshold adjustment, this invention effectively improves the robot's movement coordination and stability, significantly reduces the risk of falls, and provides technical assurance for the safe movement of humanoid robots in complex environments. Attached Figure Description

[0010] Figure 1 This is a flowchart of a real-time evaluation method for the coordination of the torso and extremities of a humanoid robot according to the present invention.

[0011] Figure 2 This is a schematic diagram of the structure of a real-time evaluation device for the coordination of the torso and limbs of a humanoid robot according to the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0013] like Figure 1 This embodiment of a method and apparatus for real-time evaluation of the coordination between the torso and limbs of a humanoid robot may specifically include: S101. Collect lumbar lordosis angle signal and shoulder extension angle signal, obtain lumbar lordosis angle start time stamp, shoulder extension angle start time stamp, real-time pressure ratio of forefoot and heel, and the target arm extension distance set for the current exercise task.

[0014] An inertial measurement unit (IMU) deployed in the lumbar spine region of the robot's torso collects lumbar tilt angle signals during the lumbar tilting process. Based on the gradient change in the angular velocity signal, the starting moment of the lumbar tilt angle transitioning from a static to a dynamic state is identified, and the corresponding lumbar tilt angle initiation timestamp is extracted. Simultaneously, an IMU deployed at the shoulder joint collects shoulder extension angle signals, identifying the starting moment of the shoulder extension angle transitioning from a static to a dynamic state, and extracting the corresponding shoulder extension angle initiation timestamp. Based on the lumbar tilt angle initiation timestamp and the shoulder extension angle initiation timestamp, the shoulder extension angle initiation timestamp is subtracted from the lumbar tilt angle initiation timestamp to obtain the original lead value. Simultaneously, a plantar pressure array collects pressure values ​​output from forefoot pressure sensors and heel pressure sensors. The forefoot pressure value is divided by the heel pressure value to obtain the real-time pressure ratio between the forefoot and heel. Based on the preset target arm extension distance parameter in the current motion task instruction, the original advance value, the real-time pressure ratio between the forefoot and the heel, and the target arm extension distance parameter are aligned and arranged according to the time sequence to form advance data, and the time difference analysis results of the start time of trunk and upper limb movement in the current action sequence are output.

[0015] During the upper limb extension movement of a humanoid robot, the pre-adjustment of the torso posture and the timing of the upper limb movement initiation directly affect the smoothness of the movement and the stability of the standing posture. By deploying an inertial measurement unit in the lumbar spine region of the robot's torso, it is possible to collect three-axis angular velocity signals in real time during the forward tilting process of the lumbar spine. The three-axis angular velocity signals include angular velocity components of rotation around the robot's front-back axis, left-right axis, and vertical axis.

[0016] Specifically, the gyroscope built into the inertial measurement unit outputs a sequence of angular velocity values ​​at a fixed sampling frequency. When the robot's torso is stationary, the angular velocity values ​​fluctuate around zero. When the lumbar spine begins to tilt forward, the angular velocity component rotating around the left and right axes increases significantly, forming a gradient from rest to motion. By setting a preset gradient threshold, when the rate of change of angular velocity at several consecutive sampling points exceeds the threshold, this moment is determined as the starting moment when the lumbar spine tilt angle changes from a stationary state to a moving state, and the corresponding lumbar spine tilt angle start timestamp is extracted.

[0017] In one embodiment, an inertial measurement unit deployed at the shoulder joint uses the same signal processing method to acquire angular velocity signals during shoulder extension. When the shoulder joint begins to extend forward from a rest state, the angular velocity component around the shoulder joint rotation axis exhibits a gradient jump, thereby identifying the starting moment of the shoulder extension angle and extracting the corresponding shoulder extension angle initiation timestamp. Based on the above two timestamps, the shoulder extension angle initiation timestamp is subtracted from the lumbar lordosis angle initiation timestamp, and the difference obtained is the original lead value.

[0018] For example, if the lumbar lordosis angle initiation timestamp is 120 milliseconds after the start of the movement, and the shoulder extension angle initiation timestamp is 200 milliseconds after the start of the movement, then the original lead value is 80 milliseconds, representing the time interval between the trunk intervention adjustment movement and the upper limb extension movement. Simultaneously, a plantar pressure array is used to collect pressure distribution data on the support surface. The plantar pressure array includes pressure sensors distributed in the forefoot area and pressure sensors distributed in the heel area, with each sensor outputting the pressure value borne by its corresponding area. Dividing the pressure value output by the forefoot pressure sensor by the pressure value output by the heel pressure sensor yields the real-time pressure ratio between the forefoot and heel, which reflects the distribution of the robot's center of gravity in the forward-backward direction. Further, based on the preset target arm extension distance parameter in the current motion task instruction, the original lead value, the real-time pressure ratio between the forefoot and heel, and the target arm extension distance parameter are time-series aligned according to the sampling time. The time series alignment refers to combining the lead value, pressure ratio, and extension distance parameter obtained at the same sampling time into a single data record. Data records from multiple sampling times are arranged in chronological order to form lead data, and the time difference analysis results of the start time of trunk and upper limb movement in the current action sequence are output.

[0019] S102. The Kalman filter algorithm is used to process the lumbar lordosis angle signal and the shoulder extension angle signal, and after noise reduction, the smoothed lumbar lordosis initiation time and the shoulder extension initiation time are obtained as a precise lead value.

[0020] The angle value sequences of the lumbar lordosis angle and shoulder extension angle signals within the sampling period are extracted separately. For the high-frequency noise components present in the angle value sequences, a Kalman filter algorithm state vector is constructed. This state vector contains two state components: the current angle value and the angular velocity value. A state transition matrix is ​​established based on the sampling interval of the inertial measurement unit (IMU), describing the recursive relationship between angle and angular velocity values ​​between adjacent sampling moments. Based on the state transition matrix, process noise covariance and observation noise covariance are defined. The process noise covariance reflects the degree of random disturbance during signal transmission, and the observation noise covariance reflects the degree of measurement error output by the IMU. The prior angle estimate for the current moment is calculated through the prediction phase of the Kalman filter algorithm. Based on the prior angle estimate, combined with the actual angle observation value output by the inertial measurement unit at the current moment, the update phase of the Kalman filter algorithm is executed. The Kalman gain is calculated and the prior angle estimate is corrected to obtain the posterior angle estimate. The above filtering process is then applied to the lumbar lordosis angle signal and the shoulder extension angle signal respectively, resulting in smoothed lumbar lordosis angle sequences and smoothed shoulder extension angle sequences. Based on the smoothed lumbar lordosis angle sequence, the moment when the angle value jumps from a static state to a dynamic state is identified as the smoothed lumbar lordosis initiation moment. Similarly, based on the smoothed shoulder extension angle sequence, the moment when the angle value jumps from a static state to a dynamic state is identified as the smoothed shoulder extension initiation moment. Subtracting the smoothed lumbar lordosis initiation moment from the smoothed shoulder extension initiation moment yields the precise lead value.

[0021] During the upper limb extension movements of a humanoid robot, the angle signal output by the inertial measurement unit (IMU) typically contains high-frequency noise components introduced by the sensor's own characteristics and external vibrations. These noise components can cause deviations in the motion initiation time extracted from the original signal. The Kalman filter algorithm, as a recursive state estimation method, establishes a state-space model of the signal and corrects it by combining the predicted and observed values ​​at each sampling time, thereby achieving a smooth estimate of the true angle value.

[0022] Specifically, constructing the state vector is a fundamental step in the Kalman filter algorithm. In this implementation, the state vector contains two components: the first component is the angle value at the current moment, reflecting the actual position of lumbar lordosis or shoulder extension; the second component is the angular velocity value at the current moment, reflecting the rate of angle change. By combining the angle and angular velocity values ​​into a state vector, the positional state and motion trend of the joint can be estimated simultaneously.

[0023] In one embodiment, the state transition matrix is ​​established based on the assumption of uniform motion. It is assumed that the angular velocity remains constant between two adjacent sampling moments. Therefore, the angle value at the next moment is equal to the current angle value plus the product of the angular velocity value and the sampling interval, and the angular velocity value at the next moment remains unchanged. The state transition matrix is ​​a second-order square matrix describing this recursive relationship, where the element in the first row and first column is 1, the element in the first row and second column is the sampling interval, the element in the second row and first column is zero, and the element in the second row and second column is 1. The sampling interval of the inertial measurement unit is typically determined by its hardware specifications.

[0024] For example, when the sampling frequency is 100 Hz, the sampling interval is 10 milliseconds.

[0025] It should be noted that the settings of the process noise covariance and observation noise covariance directly affect the filtering effect. The process noise covariance reflects the degree of deviation between the state transition model and the actual motion. When there are acceleration changes in the robot joint motion, the error introduced by the uniform velocity assumption is the main source of process noise. The observation noise covariance reflects the degree of measurement error in the output signal of the inertial measurement unit, including sensor zero-bias drift, quantization noise, and environmental vibration interference. In practical applications, the values ​​of the process noise covariance and observation noise covariance can be obtained through statistical analysis of the sensor output signal in a static state. Specifically, the variance of the static sampling sequence is calculated as the initial value of the observation noise covariance. Further, in the prediction stage of the Kalman filter algorithm, the prior angle estimate for the current moment is calculated based on the posterior state estimate of the previous moment through the state transition matrix. The prior angle estimate represents the prediction of the current angle based on the motion model before obtaining the current observation value. The prediction stage also calculates the prior error covariance to quantify the uncertainty of the prior estimate.

[0026] In one possible implementation, the update phase is the core of the Kalman filter algorithm for noise suppression. After the inertial measurement unit outputs the current angle observation value, the algorithm first calculates the observation residual, which is the difference between the observed angle value and the prior angle estimate. Then, it calculates the Kalman gain, a weighting coefficient between zero and one, used to balance the confidence levels of the prior estimate and the observed value. When the observation noise is high, the Kalman gain approaches zero, indicating greater confidence in the predicted value; when the process noise is high, the Kalman gain approaches one, indicating greater confidence in the observed value. The posterior angle estimate is equal to the prior angle estimate plus the product of the Kalman gain and the observation residual; this posterior estimate is the smoothed angle value after removing noise.

[0027] Understandably, after performing the above filtering processing on the lumbar lordosis angle signal and the shoulder extension angle signal respectively, two smoothed angle sequences are obtained. Compared with the original signal, the high-frequency jitter components of the smoothed angle sequences are significantly suppressed, and the angle change curves show smooth and continuous characteristics, which facilitates accurate identification of the moment of motion state transition.

[0028] Preferably, the rate of change threshold method is used to identify the initiation time from the smoothed angle sequence. In a static state, the smoothed angle values ​​fluctuate slightly around a certain benchmark value, and the rate of change of angle between adjacent sampling points is lower than a preset static threshold. When the joint begins to move, the angle values ​​show a continuous increasing or decreasing trend, and the rate of change of angle between adjacent sampling points exceeds a preset movement threshold. The sampling time at which the rate of change first exceeds the movement threshold is determined as the initiation time. The corresponding initiation times are extracted from the smoothed lumbar lordosis angle sequence and the smoothed shoulder extension angle sequence, respectively. The difference between the shoulder extension initiation time and the lumbar lordosis initiation time is the precise advance value, which reflects the precise time interval between the trunk intervention adjustment movement and the upper limb extension movement after noise suppression processing.

[0029] S103. Assess whether the precision advance value exceeds the preset initial upper limit threshold. If it does, analyze the pressure ratio of the forefoot and heel based on the real-time pressure ratio of the forefoot and heel, identify the critical point where the current pressure ratio enters the rapid increase range, and obtain the assessment result of the risk of triggering the fall reflex. If it does not exceed, mark the current precision advance value as normal range and directly proceed to the assessment of the smoothness of the hand end trajectory.

[0030] The precise leading value is compared with a preset initial upper limit threshold. If the precise leading value does not exceed the initial upper limit threshold, the current precise leading value is marked as normal and the hand end-effector trajectory smoothness evaluation process is directly initiated. If the precise leading value exceeds the initial upper limit threshold, the real-time pressure ratio sequence of the forefoot and heel within the current sampling period is extracted, and the pressure ratio change trend judgment process is initiated. Based on the real-time pressure ratio sequence of the forefoot and heel, the difference in pressure ratio between adjacent sampling times is calculated to obtain the pressure ratio rise rate. The pressure ratio rise rate is compared with a preset rapid rise rate threshold. When the pressure ratio rise rate exceeds the rapid rise rate threshold, the current pressure ratio is determined to have entered the rapid rise interval. The sampling time of the first entry into the rapid rise interval is recorded as the rapid rise critical point, and the time length from the rapid rise critical point to the moment when the pressure ratio rise rate falls back below the rapid rise rate threshold is recorded as the duration of the rapid rise interval. Based on the pressure ratio corresponding to the rapid increase critical point and the duration of the rapid increase interval, the risk level of triggering the fall prevention reflex due to a rapid increase in forefoot pressure is determined. If the duration of the rapid increase interval exceeds a preset duration threshold, the assessment result of the fall prevention reflex triggering risk is output as high risk; if the duration of the rapid increase interval does not exceed the duration threshold, the assessment result of the fall prevention reflex triggering risk is output as low risk.

[0031] During the upper limb extension movements of a humanoid robot, the time interval between the body intervention adjustment and the upper limb movement directly affects the magnitude of changes in plantar pressure distribution. When the precision advance value is too large, the lumbar spine forward tilting movement is initiated prematurely, causing the robot's center of gravity to shift forward beyond the normal range. This leads to a rapid increase in pressure on the forefoot, thereby increasing the risk of triggering the fall protection mechanism.

[0032] Specifically, the initial upper limit threshold is set based on the distribution range of lead values ​​when the robot performs upper limb extension movements in a standard standing posture. By collecting lead value samples during multiple normal movement executions, the upper boundary value of the sample distribution is taken as the initial upper limit threshold. When the precise lead value does not exceed this initial upper limit threshold, it indicates that the timing coordination between the trunk and upper limbs is within the normal range, and no additional judgment on the pressure change trend is required.

[0033] In one embodiment, when the precision advance value exceeds the initial upper limit threshold, the real-time pressure ratio sequence of the forefoot and heel within the current sampling period is extracted. The pressure ratio rise rate is calculated by taking the pressure ratio of two adjacent sampling moments, subtracting the pressure ratio of the previous moment from the pressure ratio of the later moment, and the difference is the pressure ratio rise rate within that time period. A positive value indicates an increase in the pressure ratio, meaning that the proportion of forefoot pressure relative to heel pressure is increasing.

[0034] It should be noted that the rapid rise rate threshold reflects the safety boundary of changes in plantar pressure distribution. When the pressure rise rate exceeds the rapid rise rate threshold, it indicates that the forefoot pressure is increasing rapidly, and the robot's center of gravity is shifting forward quickly. The rapid rise interval is determined by the moment when the pressure rise rate first exceeds the rapid rise rate threshold; this moment is the rapid rise critical point. From the rapid rise critical point, the pressure rise rate at subsequent sampling moments is continuously monitored until the pressure rise rate falls back below the rapid rise rate threshold. The time between these two moments is the duration of the rapid rise interval. Furthermore, the duration of the rapid rise interval reflects the duration of the rapid increase in forefoot pressure. If the duration of the rapid rise interval exceeds the preset duration threshold, it indicates that the robot is in a state of rapid forward shift of the center of gravity for a longer period, and the probability of the fall prevention reflex mechanism being falsely triggered is high, resulting in a high-risk assessment of the fall prevention reflex trigger risk. If the duration of the rapid rise interval does not exceed the duration threshold, it indicates that the rapid increase in the pressure ratio is a short-term fluctuation, resulting in a low-risk assessment. The assessment results provide a basis for subsequent adjustments to the allowable upper limit of advance quantities.

[0035] S104. Based on the assessment results of the risk of fall reflex triggering, analyze the pressure ratio and the target arm extension distance when entering the rapid rise zone, and calculate the adjusted upper limit of the advance amount.

[0036] Based on the risk assessment results of the fall reflex trigger, if the assessment result is high risk, the pressure rise rate value corresponding to the entry into the rapid rise zone is extracted as the critical pressure rise rate. Simultaneously, the target arm extension distance in the current exercise task is obtained, and a numerical correspondence table between the critical pressure rise rate and the target arm extension distance is established to obtain the mapping relationship between pressure rise rate and extension distance. According to the mapping relationship between pressure rise rate and extension distance, the target arm extension distance is compared with a preset extension distance benchmark value. If the target arm extension distance is greater than the benchmark value, the benchmark value is divided by the target arm extension distance to obtain an advance adjustment coefficient; if the target arm extension distance is less than or equal to the benchmark value, no adjustment is made, and the original preset threshold is maintained as the scaled threshold. The initial upper limit threshold is multiplied by the advance adjustment coefficient to obtain the scaled threshold, the critical pressure rise rate is multiplied by a preset pressure correction coefficient to obtain the pressure correction amount, and the scaled threshold is subtracted from the pressure correction amount to obtain the adjusted allowable upper limit of advance.

[0037] When a humanoid robot performs an upper limb extension movement, the target extension distance of the arm directly affects the required amplitude of the torso intervention adjustment movement. When the arm extension distance is large, the forward tilt of the torso increases accordingly, and the peak pressure on the forefoot also increases. In this case, the allowable range of forward extension should be narrowed to reduce the risk of the fall prevention reflex being falsely triggered.

[0038] Specifically, when the risk assessment result for triggering the fall prevention reflex is high, the corresponding pressure rise rate value is extracted from the starting moment of the rapid rise zone. This value reflects the rate of increase in forefoot pressure relative to heel pressure. By establishing a numerical correspondence table between the pressure rise rate and the target arm extension distance, a mapping relationship is formed between the two. This mapping relationship records the typical pressure rise rate range when the rapid rise zone is triggered under different extension distance conditions.

[0039] In one embodiment, the extension distance reference value is set based on the distance parameters when the robot performs a moderate extension movement in a standard standing posture. When the target arm extension distance is greater than the extension distance reference value, it indicates that the current task requires a large extension range, and the allowable space for advance should be reduced. In this case, the extension distance reference value is divided by the target arm extension distance, and the quotient is less than one, which is used as the advance adjustment coefficient. When the target arm extension distance is less than or equal to the extension distance reference value, it indicates that the current task requires a small extension range, and the allowable space for advance is relatively large. In this case, the target arm extension distance is divided by the extension distance reference value, and the quotient is less than or equal to one, which is used as the advance adjustment coefficient.

[0040] It should be noted that the pressure correction factor reflects the degree to which the rate of pressure rise affects the upper limit of the allowable advance. A higher critical rate of pressure rise indicates a more dramatic increase in forefoot pressure when entering the rapid rise zone, increasing the likelihood of triggering the fall prevention reflex. Therefore, a larger safety margin needs to be deducted from the upper limit of the allowable advance. Multiplying the critical rate of pressure rise by the pressure correction factor yields the pressure correction amount, expressed in units of time and having the same dimensions as the advance. Further, multiplying the initial upper limit threshold by the advance adjustment factor yields a scaled threshold adjusted for the extension distance factor. Subtracting the pressure correction amount from this scaled threshold gives the adjusted upper limit of the allowable advance. This upper limit comprehensively considers both arm extension distance and the degree of rapid increase in plantar pressure, providing a dynamic boundary for determining whether the current advance value is within a safe range.

[0041] S105. By assessing whether the current advance value exceeds the adjusted upper limit of the advance, a conclusion is drawn regarding the necessity of interrupting the arm extension movement. If no interruption is required, the smoothness of the mid-range velocity of the hand end trajectory is assessed to obtain the assessment result of the smoothness of the hand end trajectory. If an interruption is required, the movement is returned to the previous safe posture, and the interruption status and interruption reason are recorded.

[0042] The precise lead value is compared with the adjusted allowable lead limit. If the precise lead value exceeds the adjusted allowable lead limit, the necessity of interrupting the arm extension movement is determined to be interrupted, triggering a posture reversal command. The previous safe posture data stored before the movement is executed is read, and each joint of the robot is driven to restore its posture according to the lumbar lordosis angle and shoulder extension angle in the previous safe posture data. The interruption status and interruption reason are recorded, such as exceeding the maximum range of motion of the joint. If the precise lead value does not exceed the adjusted allowable lead limit, the necessity of interrupting the arm extension movement is determined to be interrupted without interruption. The position coordinate sequence of the hand end in the middle section of the movement trajectory is extracted. The middle section is the interval where the distance from the hand end to both the starting position and the target position exceeds a preset distance ratio during the movement from the starting position to the target position. The change in the position coordinates of the hand end between adjacent sampling times is calculated and divided by the sampling interval to obtain the instantaneous velocity value at each sampling time in the middle section. Based on the instantaneous velocity values ​​at each sampling time within the middle section, the difference between adjacent instantaneous velocity values ​​is calculated as the velocity fluctuation amount. The proportion of the number of times the velocity fluctuation amount exceeds a preset fluctuation threshold to the total number of samplings is counted as the fluctuation ratio. If the fluctuation ratio is lower than a preset smoothness threshold, the output evaluation result of the hand end trajectory smoothness is smooth; if the fluctuation ratio is higher than or equal to the smoothness threshold, the output evaluation result of the hand end trajectory smoothness is not smooth.

[0043] During the upper limb extension movement of a humanoid robot, if the precise advance value exceeds the adjusted allowable upper limit of the advance, it indicates that the time interval between the body intervention adjustment and the upper limb movement is too large. Continuing to execute the movement risks triggering the fall prevention reflex mechanism. In this case, the robot should interrupt the current movement and return to a safe posture. The previous safe posture data is pre-stored before each movement, including the angle values ​​and position coordinates of each joint of the robot before the movement starts. When the conclusion that interruption is necessary is determined, the stored data is read, and the lumbar spine joint and shoulder joint are driven to move in the opposite direction according to the stored angle values, so that the robot's torso and upper limbs return to the posture state before the movement was executed.

[0044] In one embodiment, when the precise lead value does not exceed the adjusted allowable lead limit, it is determined that there is no need to interrupt the movement, and the evaluation process for the smoothness of the hand end-effector trajectory begins. The movement trajectory of the hand end-effector can be divided into three parts: the initial interval, the middle interval, and the end interval. The middle interval is the main stage in which the arm moves at a relatively stable speed, and it best reflects the smoothness of the movement execution.

[0045] It should be noted that the definition of the middle section is based on a preset distance ratio parameter.

[0046] For example, if the preset distance ratio is 20%, then the middle section is the movement range where the distance from the hand's end-effector to the starting position exceeds 20% of the total travel distance, and the distance from the target position also exceeds 20% of the total travel distance. Within this middle section, the three-dimensional position coordinates of the hand's end-effector at each sampling moment are extracted, the magnitude of the change in position coordinates between adjacent sampling moments is calculated, and then divided by the sampling interval to obtain the instantaneous velocity value at each sampling moment. Furthermore, the velocity fluctuation reflects the degree of drastic change in the speed of the hand's end-effector within the middle section. The absolute value of the difference between the instantaneous velocity values ​​of two adjacent sampling moments is taken to obtain the velocity fluctuation within that time period. If the velocity fluctuation exceeds a preset fluctuation threshold, it indicates that there is a sudden change in speed at that moment. The number of sampling points in the middle section where the velocity fluctuation exceeds the fluctuation threshold is counted, and this number is divided by the total number of sampling points in the middle section to obtain the fluctuation ratio.

[0047] Understandably, a lower fluctuation ratio indicates a smoother velocity change of the hand's extremities within the mid-range, resulting in better trajectory smoothness. The fluctuation ratio is compared to a preset smoothness threshold. If the fluctuation ratio is lower than the threshold, the output assessment result for the hand's trajectory smoothness is "smooth"; if the fluctuation ratio is higher than or equal to the threshold, the output assessment result is "unsmooth". This assessment result, along with the conclusion regarding the necessity of interruption, constitutes the basis for determining the current execution state of the action.

[0048] S106. Based on the evaluation results of the smoothness of the hand end trajectory and the conclusion of the necessity of interrupting the arm extension movement, the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, and the real-time pressure ratio of the forefoot and heel are integrated to generate a coordination evaluation report of the humanoid robot's torso and limbs.

[0049] Based on the evaluation results of the smoothness of the hand's end-effector trajectory and the conclusion regarding the necessity of interrupting the arm extension movement, combined with the lumbar spine forward tilt angle initiation time stamp, the shoulder joint forward extension angle initiation time stamp, and the real-time pressure ratio of the forefoot and heel, the above evaluation results and sensor data are integrated in chronological order to obtain a coordination evaluation dataset. Based on this coordination evaluation dataset, the smoothness evaluation results, the conclusion regarding the necessity of interruption, the difference between the lumbar spine forward tilt angle initiation time stamp and the shoulder joint forward extension angle initiation time stamp, and the real-time pressure ratio of the forefoot and heel are written into a preset report format, outputting a coordination evaluation report of the humanoid robot's torso and extremities.

[0050] After assessing the smoothness of the hand's end-effector trajectory and determining the necessity of interrupting arm extension movements, the assessment results were integrated with the raw sensor data to form a complete coordination assessment record. The coordination assessment dataset includes smoothness assessment results, conclusions on the necessity of interruption, timestamps of lumbar lordosis initiation, timestamps of shoulder extension initiation, and real-time pressure ratios of the forefoot and heel. These data are arranged and organized in chronological order of acquisition.

[0051] In one embodiment, the whole-body joint coordination assessment report is output using a preset report format. The report records the difference between the initiation time stamp of the lumbar lordosis angle and the initiation time stamp of the shoulder extension angle, which reflects the temporal coordination between the trunk intervention adjustment and the upper limb movement. The report also records the changes in the real-time pressure ratio between the forefoot and the heel, reflecting the stability of the support surface during the execution of the movement.

[0052] It should be noted that the smoothness assessment results and the conclusion on the necessity of interruption are included in the report as a comprehensive basis for judging the quality and safety of motion execution. When the smoothness assessment result is smooth and the conclusion on the necessity of interruption is that no interruption is needed, it indicates that the coordination of the whole body joints in the current motion sequence is good; when the smoothness assessment result is not smooth or the conclusion on the necessity of interruption is that an interruption is required, it indicates that there is room for improvement in coordination. This report provides data support for the adjustment of motion control parameters of the humanoid robot.

[0053] like Figure 2 This invention provides a real-time assessment device for the coordination of the torso and limbs of a humanoid robot, mainly comprising: The signal acquisition module is used to acquire the lumbar spine forward tilt angle signal and the shoulder joint forward extension angle signal, and obtain the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, the real-time pressure ratio of the forefoot and the heel, and the target arm extension distance set for the current exercise task. The signal processing module is used to process the lumbar lordosis angle signal and the shoulder extension angle signal using the Kalman filter algorithm, and after noise reduction, obtain the smoothed accurate lead value between the lumbar lordosis initiation time and the shoulder extension initiation time. The advanced measurement assessment module is used to assess whether the precise advanced measurement value exceeds the preset initial upper limit threshold. If it does, it analyzes the trend of the pressure ratio increase between the forefoot and heel based on the real-time pressure ratio between the forefoot and heel, identifies the critical point where the current pressure ratio enters the rapid increase zone, and obtains the assessment result of the risk of triggering the fall reflex. If it does not exceed, the current precise advanced measurement value is marked as normal range and directly enters the assessment of the smoothness of the hand end trajectory. The upper limit adjustment module is used to analyze the pressure ratio and the target arm extension distance when entering the rapid rise zone based on the assessment results of the risk of fall reflex triggering, and calculate the adjusted upper limit of the advance amount; The interruption judgment module is used to determine the necessity of interrupting the arm extension movement by evaluating whether the current advance value exceeds the adjusted upper limit of the advance. If no interruption is required, the smoothness of the mid-segment velocity of the hand end trajectory is evaluated to obtain the evaluation result of the smoothness of the hand end trajectory. If an interruption is required, the system returns to the previous safe posture and records the interruption status and interruption reason. The report generation module is used to generate a coordination assessment report of the humanoid robot's torso and limbs based on the evaluation results of the smoothness of the hand end trajectory and the conclusion of the necessity of interrupting the arm extension movement, by integrating the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, and the real-time pressure ratio of the forefoot and heel.

[0054] If the technical solution of this application involves the collection, processing, or application of personal information, the relevant products have strictly complied with the requirements of the "Personal Information Protection Law of the People's Republic of China" and other laws and regulations before implementing any personal information processing activities, clearly and explicitly informing individuals of the rules for personal information processing and obtaining their independent and voluntary authorization and consent. Specifically, if the information involved is sensitive personal information, the product has not only obtained the individual's separate consent before processing, but this consent is also an explicit consent made on the basis of full knowledge. For example, in areas where personal information collection devices such as cameras are deployed, prominent and eye-catching signs have been set up to clearly inform users that entering the area is considered as consenting to the collection of their personal information; or, on the personal information processing interface (such as applications, web pages, etc.), through pop-ups, checkboxes, or active uploads, the user is required to actively authorize the process after clearly displaying key rules such as the identity of the personal information processor, the purpose of processing, the processing method, and the types of information involved.

[0055] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application.

Claims

1. A method for real-time evaluation of the coordination between the torso and limbs of a humanoid robot, characterized in that, The method includes: Collect lumbar lordosis angle signal and shoulder extension angle signal, and obtain lumbar lordosis angle start time stamp, shoulder extension angle start time stamp, real-time pressure ratio of forefoot and heel, and the target arm extension distance set for the current exercise task. The Kalman filter algorithm is used to process the lumbar lordosis angle signal and the shoulder extension angle signal, and after noise reduction, the smoothed lumbar lordosis initiation time and the shoulder extension initiation time are obtained as a precise lead value. The assessment determines whether the precision advance value exceeds the preset initial upper limit threshold. If it does, the analysis is based on the real-time pressure ratio between the forefoot and heel to determine the trend of the pressure ratio increase between the forefoot and heel, identify the critical point where the current pressure ratio enters the rapid increase zone, and obtain the assessment result of the risk of triggering the fall reflex. If it does not exceed, the current precision advance value is marked as normal and the assessment proceeds directly to the smoothness of the hand end trajectory. Based on the assessment results of the risk of fall reflex triggering, the pressure ratio and the target arm extension distance when entering the rapid rise zone are analyzed, and the adjusted allowable upper limit of the advance amount is calculated. By assessing whether the current lead value exceeds the adjusted upper limit of the lead, the necessity of interrupting the arm extension movement is determined. If no interruption is required, the smoothness of the mid-range velocity of the hand end trajectory is assessed to obtain the assessment result of the smoothness of the hand end trajectory. If an interruption is required, the movement is returned to the previous safe posture, and the interruption status and interruption reason are recorded. Based on the evaluation results of the smoothness of the hand end trajectory and the conclusion of the necessity of interrupting the arm extension movement, the coordination evaluation report of the humanoid robot's torso and limbs is generated by integrating the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, and the real-time pressure ratio of the forefoot and heel.

2. The real-time evaluation method for the coordination of the torso and limbs of a humanoid robot according to claim 1, characterized in that, The acquisition of lumbar lordosis angle signals and shoulder extension angle signals, obtaining the lumbar lordosis angle initiation time stamp, shoulder extension angle initiation time stamp, real-time pressure ratio of the forefoot and heel, and the target arm extension distance set for the current exercise task, includes: The system collects lumbar lordosis angle signals during the lumbar lordosis process. Based on the gradient change of the lumbar lordosis angle signal, it identifies the starting moment when the lumbar lordosis angle changes from a static state to a dynamic state, and extracts the lumbar lordosis angle initiation timestamp corresponding to this starting moment. Simultaneously, it collects shoulder joint extension angle signals, identifies the starting moment when the shoulder joint extension angle changes from a static state to a dynamic state, and extracts the shoulder joint extension angle initiation timestamp corresponding to this starting moment. It also collects the pressure values ​​output by the forefoot pressure sensor and the rearfoot pressure sensor through a plantar pressure array, and divides the forefoot pressure value by the rearfoot pressure value to obtain the real-time pressure ratio between the forefoot and rearfoot. Finally, it collects the preset target arm extension distance in the current movement task command.

3. The real-time evaluation method for the coordination of the torso and limbs of a humanoid robot according to claim 1, characterized in that, The process employs a Kalman filter algorithm to process the lumbar lordosis angle signal and the shoulder extension angle signal, and after noise reduction, obtains a smoothed, precise lead value between the lumbar lordosis initiation time and the shoulder extension initiation time, including: The angle value sequences of lumbar lordosis angle and shoulder extension angle are extracted separately within the sampling period. For the high-frequency noise components present in the lumbar lordosis angle and shoulder extension angle signals, a Kalman filter algorithm state vector is constructed. This state vector contains two state components: the current angle value and the angular velocity value. A state transition matrix is ​​established by combining these two state components and the sampling interval of the inertial measurement unit. This state transition matrix describes the recursive relationship between angle and angular velocity values ​​between adjacent sampling moments. Based on the state transition matrix, and after setting the process noise covariance and observation noise covariance, the prior angle at the current moment is calculated through the prediction phase of the Kalman filter algorithm. The prior angle estimate is calculated and, based on the current angle observation, the Kalman filter algorithm is updated to calculate the Kalman gain and correct the prior angle estimate, resulting in a smoothed lumbar lordosis angle sequence and a smoothed shoulder extension angle sequence. The moment when the angle value changes from a static state to a dynamic state is identified as the smoothed lumbar lordosis initiation moment, and the moment when the angle value changes from a static state to a dynamic state is identified as the smoothed shoulder extension initiation moment. The smoothed lumbar lordosis initiation moment is subtracted from the smoothed shoulder extension initiation moment to obtain a precise lead value.

4. The real-time evaluation method for the coordination of the torso and limbs of a humanoid robot according to claim 1, characterized in that, The assessment determines whether the precise advance value exceeds a preset initial upper limit threshold. If it does, the trend of the pressure ratio increase between the forefoot and heel is analyzed based on the real-time pressure ratio of the forefoot and heel to identify the critical point where the current pressure ratio enters the rapid increase range, thus obtaining the assessment result of the risk of triggering the fall reflex. If it does not exceed the threshold, the current precise advance value is marked as normal and the assessment directly proceeds to the smoothness assessment of the hand's end-effector trajectory, including: The precise lead value is compared with a preset initial upper limit threshold. If the precise lead value does not exceed the initial upper limit threshold, the current precise lead value is marked as normal and the process directly proceeds to the hand end-effector trajectory smoothness evaluation process. If the precise lead value exceeds the initial upper limit threshold, the real-time pressure ratio sequence of the forefoot and heel within the current sampling period is extracted, and the pressure ratio change trend judgment process is initiated. Based on the real-time pressure ratio sequence of the forefoot and heel, the difference in pressure ratio between adjacent sampling times is calculated to obtain the pressure ratio rise rate. The pressure ratio rise rate is compared with a preset rapid rise rate threshold. When the pressure ratio rise rate exceeds the rapid rise rate threshold, the process is initiated. When the rate threshold is reached, the current pressure ratio is determined to have entered a rapid increase range. The sampling time when it first enters the rapid increase range is recorded as the rapid increase critical point, and the time length from the rapid increase critical point to the moment when the pressure ratio increase rate drops below the rapid increase rate threshold is recorded as the duration of the rapid increase range. Based on the pressure ratio value corresponding to the rapid increase critical point and the duration of the rapid increase range, the risk level of the fall reflex triggered by the rapid increase in forefoot pressure is determined. If the duration of the rapid increase range exceeds a preset duration threshold, the assessment result of the fall reflex trigger risk is output as high risk; if the duration of the rapid increase range does not exceed the duration threshold, the assessment result of the fall reflex trigger risk is output as low risk.

5. The real-time evaluation method for the coordination of the torso and limbs of a humanoid robot according to claim 1, characterized in that, Based on the risk assessment results of the fall prevention reflex, the pressure ratio and the target arm extension distance when entering the rapid rise zone are analyzed to calculate the adjusted allowable upper limit of the forward movement, including: Based on the risk assessment results of the fall reflex triggering, if the assessment result is high risk, the pressure rise rate value corresponding to the entry into the rapid rise zone is extracted as the critical pressure rise rate. Simultaneously, the target arm extension distance in the current exercise task is obtained, and a numerical correspondence table between the critical pressure rise rate and the target arm extension distance is established to obtain the mapping relationship between pressure rise rate and extension distance. Based on the mapping relationship between pressure rise rate and extension distance, the target arm extension distance is compared with a preset extension distance benchmark value. If the target arm extension distance is greater than the benchmark value, the benchmark value is divided by the target arm extension distance to obtain the advance amount adjustment coefficient. Based on the advance amount adjustment coefficient and the critical pressure rise rate, the initial upper limit threshold is multiplied by the advance amount adjustment coefficient to obtain a scaled threshold. The critical pressure rise rate is multiplied by a preset pressure correction coefficient to obtain a pressure correction amount. The scaled threshold is subtracted from the pressure correction amount to obtain the adjusted allowable upper limit of the advance amount.

6. A real-time evaluation device for the coordination of the torso and limbs of a humanoid robot, characterized in that, The device includes: The signal acquisition module is used to acquire the lumbar spine forward tilt angle signal and the shoulder joint forward extension angle signal, and obtain the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, the real-time pressure ratio of the forefoot and the heel, and the target arm extension distance set for the current exercise task. The signal processing module is used to process the lumbar lordosis angle signal and the shoulder extension angle signal using the Kalman filter algorithm, and after noise reduction, obtain the smoothed accurate lead value between the lumbar lordosis initiation time and the shoulder extension initiation time. The advanced measurement assessment module is used to assess whether the precise advanced measurement value exceeds the preset initial upper limit threshold. If it does, it analyzes the trend of the pressure ratio increase between the forefoot and heel based on the real-time pressure ratio between the forefoot and heel, identifies the critical point where the current pressure ratio enters the rapid increase zone, and obtains the assessment result of the risk of triggering the fall reflex. If it does not exceed, the current precise advanced measurement value is marked as normal range and directly enters the assessment of the smoothness of the hand end trajectory. The upper limit adjustment module is used to analyze the pressure ratio and the target arm extension distance when entering the rapid rise zone based on the assessment results of the risk of fall reflex triggering, and calculate the adjusted upper limit of the advance amount; The interruption judgment module is used to determine the necessity of interrupting the arm extension movement by evaluating whether the current advance value exceeds the adjusted upper limit of the advance. If no interruption is required, the smoothness of the mid-segment velocity of the hand end trajectory is evaluated to obtain the evaluation result of the smoothness of the hand end trajectory. If an interruption is required, the system returns to the previous safe posture and records the interruption status and interruption reason. The report generation module is used to generate a coordination assessment report of the humanoid robot's torso and limbs based on the evaluation results of the smoothness of the hand end trajectory and the conclusion of the necessity of interrupting the arm extension movement, by integrating the lumbar spine forward tilt angle start time stamp, the shoulder joint forward extension angle start time stamp, and the real-time pressure ratio of the forefoot and heel.

7. The real-time evaluation device for the coordination of the torso and extremities of a humanoid robot according to claim 6, characterized in that, The acquisition of lumbar lordosis angle signals and shoulder extension angle signals, obtaining the lumbar lordosis angle initiation time stamp, shoulder extension angle initiation time stamp, real-time pressure ratio of the forefoot and heel, and the target arm extension distance set for the current exercise task, includes: The system collects lumbar lordosis angle signals during the lumbar lordosis process. Based on the gradient change of the angular velocity signal, it identifies the starting moment when the lumbar lordosis angle changes from a static state to a dynamic state, and extracts the lumbar lordosis angle initiation timestamp corresponding to this starting moment. Simultaneously, it collects shoulder joint extension angle signals, identifies the starting moment when the shoulder joint extension angle changes from a static state to a dynamic state, and extracts the shoulder joint extension angle initiation timestamp corresponding to this starting moment. Based on the lumbar lordosis angle initiation timestamp and the shoulder joint extension angle initiation timestamp, it subtracts the lumbar lordosis angle initiation timestamp from the shoulder joint extension angle initiation timestamp to obtain the original advance value. At the same time, it collects the pressure values ​​output by the forefoot pressure sensor and the rearfoot pressure sensor through a plantar pressure array, and divides the forefoot pressure value by the rearfoot pressure value to obtain the real-time pressure ratio of the forefoot and rearfoot. Finally, it collects the preset target arm extension distance in the current movement task command.

8. The real-time evaluation device for the coordination of the torso and limbs of a humanoid robot according to claim 6, characterized in that, The process employs a Kalman filter algorithm to process the lumbar lordosis angle signal and the shoulder extension angle signal, and after noise reduction, obtains a smoothed, precise lead value between the lumbar lordosis initiation time and the shoulder extension initiation time, including: The angle value sequences of lumbar lordosis angle and shoulder extension angle are extracted separately within the sampling period. For the high-frequency noise components present in the angle value sequences, a Kalman filter algorithm state vector is constructed. This state vector contains two state components: the current angle value and the angular velocity value. A state transition matrix is ​​established based on the sampling interval of the inertial measurement unit, describing the recursive relationship between angle and angular velocity values ​​between adjacent sampling moments. Based on the state transition matrix, the process noise covariance and observation noise covariance are set, and the prior angle estimate for the current moment is calculated through the prediction phase of the Kalman filter algorithm. Based on the prior angle estimate and the current angle observation value, the Kalman filter is executed. In the update phase of the wavelet algorithm, the Kalman gain is calculated and the prior angle estimate is corrected to obtain the posterior angle estimate. Filtering is then performed on the lumbar lordosis angle signal and the shoulder extension angle signal to obtain smoothed lumbar lordosis angle sequences and smoothed shoulder extension angle sequences. Based on the smoothed lumbar lordosis angle sequence, the moment when the angle value jumps from a static state to a dynamic state is identified as the smoothed lumbar lordosis initiation moment. Based on the smoothed shoulder extension angle sequence, the moment when the angle value jumps from a static state to a dynamic state is identified as the smoothed shoulder extension initiation moment. The smoothed lumbar lordosis initiation moment is subtracted from the smoothed shoulder extension initiation moment to obtain the precise lead value.

9. A real-time evaluation device for the coordination of the torso and extremities of a humanoid robot according to claim 6, characterized in that, The assessment determines whether the precise advance value exceeds a preset initial upper limit threshold. If it does, the trend of the pressure ratio increase between the forefoot and heel is analyzed based on the real-time pressure ratio of the forefoot and heel to identify the critical point where the current pressure ratio enters the rapid increase range, thus obtaining the assessment result of the risk of triggering the fall reflex. If it does not exceed the threshold, the current precise advance value is marked as normal and the assessment directly proceeds to the smoothness assessment of the hand's end-effector trajectory, including: The precise lead value is compared with a preset initial upper limit threshold. If the precise lead value does not exceed the initial upper limit threshold, the current precise lead value is marked as normal and the process directly proceeds to the hand end-effector trajectory smoothness evaluation process. If the precise lead value exceeds the initial upper limit threshold, the real-time pressure ratio sequence of the forefoot and heel within the current sampling period is extracted, and the pressure ratio change trend judgment process is initiated. Based on the real-time pressure ratio sequence of the forefoot and heel, the difference in pressure ratio between adjacent sampling times is calculated to obtain the pressure ratio rise rate. The pressure ratio rise rate is compared with a preset rapid rise rate threshold. When the pressure ratio rise rate exceeds the rapid rise rate threshold, the process is initiated. When the rate threshold is reached, the current pressure ratio is determined to have entered a rapid increase range. The sampling time when it first enters the rapid increase range is recorded as the rapid increase critical point, and the time length from the rapid increase critical point to the moment when the pressure ratio increase rate drops below the rapid increase rate threshold is recorded as the duration of the rapid increase range. Based on the pressure ratio value corresponding to the rapid increase critical point and the duration of the rapid increase range, the risk level of the fall reflex triggered by the rapid increase in forefoot pressure is determined. If the duration of the rapid increase range exceeds a preset duration threshold, the assessment result of the fall reflex trigger risk is output as high risk; if the duration of the rapid increase range does not exceed the duration threshold, the assessment result of the fall reflex trigger risk is output as low risk.

10. A real-time evaluation device for the coordination of the torso and extremities of a humanoid robot according to claim 6, characterized in that, Based on the risk assessment results of the fall prevention reflex, the pressure ratio and the target arm extension distance when entering the rapid rise zone are analyzed to calculate the adjusted allowable upper limit of the forward movement, including: Based on the risk assessment results of the fall reflex triggering, if the assessment result is high risk, the pressure rise rate value corresponding to the entry into the rapid rise zone is extracted as the critical pressure rise rate. Simultaneously, the target arm extension distance in the current exercise task is obtained, and a numerical correspondence table between the critical pressure rise rate and the target arm extension distance is established to obtain the mapping relationship between pressure rise rate and extension distance. Based on the mapping relationship between pressure rise rate and extension distance, the target arm extension distance is compared with a preset extension distance benchmark value. If the target arm extension distance is greater than the benchmark value, the benchmark value is divided by the target arm extension distance to obtain the advance amount adjustment coefficient. Based on the advance amount adjustment coefficient and the critical pressure rise rate, the initial upper limit threshold is multiplied by the advance amount adjustment coefficient to obtain a scaled threshold. The critical pressure rise rate is multiplied by a preset pressure correction coefficient to obtain a pressure correction amount. The scaled threshold is subtracted from the pressure correction amount to obtain the adjusted allowable upper limit of the advance amount.

Images