An exoskeleton delay feedback control method based on gain and delay double adaptation
By employing a dual adaptive exoskeleton control method that combines gain and delay, the stride frequency is estimated in real time and the delay and gain are dynamically adjusted. This solves the problems of traditional exoskeletons in terms of stride frequency variation and weight adaptation, and achieves efficient human-computer interaction and safe auxiliary torque output.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2026-04-07
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional lower limb exoskeletons suffer from phase deviation due to fixed delay time during dynamic changes in gait frequency. Their rigid gain parameters cannot adapt to users of different weights, and they are susceptible to sensor noise during low-frequency steady-state walking, resulting in undesirable reverse torque output and control command jitter.
A control method based on gain and delay dual adaptation is adopted. By estimating step frequency in real time, inverse compensation of filters and multidimensional physical characteristic mapping, spatiotemporal dual-dimensional adaptation is achieved. The delay time and gain are dynamically adjusted to match human movement, generate auxiliary torque and drive exoskeleton joints.
It improves control precision, enhances parameter adaptability, ensures precise alignment between auxiliary torque and human movement under variable step speed conditions, reduces undesirable reverse torque output, and improves safety and human-machine interaction efficiency.
Smart Images
Figure CN122143028A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of human-computer interaction control, and specifically to an exoskeleton delay feedback control method based on dual adaptation of gain and delay. Background Technology
[0002] Traditional Delayed Output Feedback Control (DOFC) in lower limb exoskeleton applications typically employs a fixed delay time. When the human gait frequency changes dynamically, this fixed delay time causes a phase deviation between the exoskeleton's auxiliary torque and the human's actual movement intention, resulting in an unwanted reverse torque output that hinders human movement during unintended force application. Furthermore, existing systems often neglect the inherent physical phase hysteresis introduced by the sensor's low-pass filter.
[0003] Regarding gain parameter configuration, the feedback gain of traditional DOFC is often a static empirical value or a manually fixed value. On the one hand, a fixed gain cannot be adapted to users of different weights, making it difficult to achieve parameter universality; on the other hand, when walking speed changes dynamically, a fixed gain will result in excessive torque when walking slowly (which can easily disrupt the body's balance) or insufficient assistance when walking at high speed (which cannot meet peak metabolic needs).
[0004] In addition, existing methods for extracting step frequency using an adaptive frequency oscillator (AFO) are susceptible to sensor noise during low-frequency steady-state walking, resulting in minute oscillations and frequent jitter in the underlying control commands.
[0005] In summary, existing technologies still have shortcomings in terms of dynamic matching of delay parameters, adaptive gain adjustment, and low-frequency stability. Summary of the Invention
[0006] In order to overcome the technical problems of human-machine timing mismatch and undesirable reverse torque output caused by fixed delay constant, rigid gain parameter and low-frequency observation jitter in the application of lower limb exoskeletons in variable walking speed and cross-group applications, the present invention provides an exoskeleton delay feedback control method based on dual adaptive gain and delay, wherein the lower limb exoskeleton includes hip joint, knee joint or multi-joint system.
[0007] The exoskeleton delay feedback control method of the present invention is a control architecture that combines real-time step frequency estimation, filter inverse compensation and multi-dimensional physical characteristic mapping to achieve spatiotemporal dual-dimensional adaptation of the lower limb exoskeleton, that is, to achieve precise phase matching in the time domain and compliant torque adaptive output in the amplitude domain.
[0008] The objective of this invention is achieved through the following technical solution:
[0009] A method for delay feedback control of an exoskeleton based on dual adaptive gain and delay includes:
[0010] The motion angle signals of the left and right legs of the exoskeleton are collected. After noise reduction, the equivalent phase lag time is calculated based on the phase difference of the filtered motion angle signals.
[0011] An adaptive frequency oscillator model is constructed to obtain the instantaneous gait frequency estimate of the human body based on the motion angle signal;
[0012] Calculate the feedback dynamic delay time based on the instantaneous step frequency estimate;
[0013] The feedback gain is dynamically calculated based on human body parameters and human motion state, specifically as follows:
[0014]
[0015] in: For reference gain, For the user's weight, For reference weight, This is a dynamic scaling factor based on step frequency. Adjust the coefficients to be personalized for each user;
[0016] An auxiliary torque is generated based on the motion angle signal, feedback gain, and equivalent phase lag time.
[0017] The auxiliary torque is converted into a motor drive control signal, which drives the exoskeleton joints to perform corresponding auxiliary actions.
[0018] Furthermore, it also includes a step frequency dead zone locking mechanism. When the step frequency change in adjacent cycles is less than the set dead zone threshold, the instantaneous step frequency estimate of the previous cycle is kept unchanged as the step frequency output value; when the step frequency change is greater than the set dead zone threshold, the current step frequency output value is updated.
[0019] Furthermore, the feedback dynamic delay time is calculated based on the instantaneous step frequency estimate, specifically as follows:
[0020]
[0021] in, The theoretical delay time required to achieve an ideal 90° phase lag; The inherent hysteresis constant of the filter, This is the estimated instantaneous step frequency.
[0022] Furthermore, the constraint condition for the feedback gain is:
[0023] When the user's weight increases, the feedback gain is adjusted positively accordingly; when the estimated instantaneous cadence of the human body is in the low to mid-frequency range, the gain increases with the increase of cadence or remains at a preset low limit; when the estimated instantaneous cadence of the human body reaches the high-frequency threshold, the feedback gain remains at a preset saturation limit.
[0024] Furthermore, the auxiliary torque calculation process is as follows:
[0025]
[0026] in: For dynamic feedback gain, y represents the dynamic delay time, and y represents the difference in the angles of the left and right leg joints.
[0027] Furthermore, it also includes a safety monitoring step that immediately cuts off the auxiliary torque output when any of the following conditions are detected:
[0028] The joint angular velocity is lower than the preset threshold and the output torque is consistently higher than the safety threshold; the joint angle reaches the deep flexion limit and is accompanied by high torque output.
[0029] Furthermore, the step frequency scaling factor adopts a piecewise function form, specifically:
[0030] hour, ;
[0031] interval Depend on Linear change to ;
[0032] interval Depend on Linear change to ;
[0033] hour, .
[0034] Furthermore, the dead zone threshold is set to 0.05Hz.
[0035] Furthermore, the noise reduction process employs a first-order low-pass filter.
[0036] Furthermore, the torque command output to the left leg motor and the torque command output to the right leg motor are of the same magnitude but opposite in phase.
[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0038] Improved control precision: By observing the step frequency and the filter inverse compensation logic through the adaptive oscillator, the assist phase misalignment under variable step speed conditions is effectively avoided, and the undesirable reverse torque output (negative work) generated by the exoskeleton during human-computer interaction is significantly reduced.
[0039] Enhanced parameter adaptation capability: A gain calculation model integrating static weight normalization and dynamic cadence mapping was constructed, enabling the system to have the ability to generalize across populations without manual trial and error, and to automatically balance the needs of high-frequency bursts and low-frequency stability under varying cadence conditions.
[0040] Enhanced safety: A dual watchdog monitoring module with high engineering practical value has been introduced. Through clear logical judgment conditions, it effectively avoids secondary injuries to the human body caused by deep buckling overload or motor stall. Attached Figure Description
[0041] Figure 1 This is a flowchart of the process of this invention;
[0042] Figure 2 This is a diagram illustrating the AFO step frequency dynamic following effect of the present invention;
[0043] Figure 3 This is a diagram illustrating the effect of the spatiotemporal collaborative parameter adaptive mapping of the present invention;
[0044] Figure 4 This is a diagram illustrating the effect of adaptive torque output and timing alignment under variable frequency operating conditions of the present invention. Detailed Implementation
[0045] The present invention is further described below through specific embodiments, but the scope of protection of the present invention is not limited thereto.
[0046] Example
[0047] like Figure 1 As shown, an exoskeleton delay feedback control method based on dual adaptive gain and delay includes the following steps:
[0048] S1 acquires the motion angle signals of the left and right legs of the exoskeleton. After noise reduction, the equivalent phase lag time is calculated based on the phase difference of the filtered motion angle signals, and the motion angle of the left leg joint is determined. and the angle of movement of the right leg joint .
[0049] The acquired angle signal is subjected to noise reduction processing using a first-order or higher-order low-pass filter. Its update format is as follows: Where β is the filter coefficient, with a value range of 0 < β < 1. The equivalent phase lag time generated by the filter is also calculated. Used for subsequent delay compensation calculations.
[0050] S2 constructs an adaptive frequency oscillator model and obtains the instantaneous step frequency estimate of the human body based on the motion angle signal. The instantaneous step frequency estimate is used for subsequent joint adaptive control of feedback delay and gain.
[0051] An adaptive frequency oscillator (AFO) updates its state variables in real time using the input signal to obtain an estimate of the human body's instantaneous gait frequency. .
[0052] To suppress frequency jitter caused by low-frequency noise, a step frequency dead zone locking mechanism is introduced in this step: when the step frequency change of adjacent control cycles meets the following conditions...
[0053] If the current step frequency is locked, then the instantaneous step frequency output value is updated. The dead zone threshold is set to a value of [value]. .
[0054] To further explain, dead-zone locking prevents unstable assist torque caused by real-time fluctuations in step frequency. For example, when the step frequency is between 1.4 and 1.45 Hz, it maintains the step frequency value at the moment of first entering that range, such as 1.42 Hz. This way, the output is calculated based on 1.42 Hz in real-time unless the frequency exceeds this range.
[0055] S3 dynamically calculates the feedback dynamic delay time based on the instantaneous cadence estimate. Delayed Output Feedback Control (DOFC) in human gait dynamics systems can generate stable periodic assist output through appropriate delay parameters. Theoretical analysis shows that when the delay time is close to one-quarter of the gait cycle, an energy injection effect with a phase difference of nearly 90° can be obtained. Based on this principle, this invention dynamically calculates the delay time through real-time cadence estimation, thereby achieving stable assist at different gait speeds.
[0056] The specific formula is as follows:
[0057]
[0058] in: This is the current instantaneous step frequency estimate. The equivalent phase lag time generated by the filter. The theoretical delay time required to achieve an ideal 90° phase lag.
[0059] The feedback dynamic delay time is used to compensate for system phase lag, so that the auxiliary torque maintains a preset phase relationship with human body movement. The calculation results of this step are subjected to safety limiting processing to ensure system stability.
[0060] S4 dynamically calculates the feedback gain based on human body parameters and human motion state. The feedback gain adopts a multi-dimensional mapping model, specifically:
[0061] in:
[0062] Reference gain;
[0063] User's weight;
[0064] For reference weight (e.g.) );
[0065] This is a dynamic scaling factor based on step frequency;
[0066] Adjust the coefficients to be personalized for each user.
[0067] The A piecewise linear function is used to achieve adaptive adjustment of the step speed, maintaining the minimum safety factor in the low frequency band, maintaining the saturation factor in the high frequency band, and linear interpolation in the mid frequency band.
[0068] Specifically:
[0069] hour, ;
[0070] interval Depend on Linear change to ;
[0071] interval Depend on Linear change to ;
[0072] hour, .
[0073] The constraint condition for the feedback gain is:
[0074] When the user's weight increases, the feedback gain is adjusted positively accordingly; when the estimated instantaneous cadence is in the low to mid-frequency range, the gain increases with the cadence or remains at a preset low limit; when the real-time cadence reaches the high-frequency threshold, the feedback gain remains at a preset saturation limit.
[0075] The effects of the above constraints are as follows: when walking slowly at low frequency, setting a low gain limit (such as 3 Nm) can prevent the torque from being too small and losing its auxiliary significance; when walking fast at high frequency, setting a saturation limit (such as 15 Nm) can not only meet the peak metabolic assistance needs, but also serve as an absolute safety upper limit to avoid system divergence and ensure the wearer's balance and safety.
[0076] Further explanation: In the low-to-mid frequency range, if the step frequency is less than 0.7Hz, the dynamic scaling factor based on the step frequency is maintained at 0.6. If the calculated feedback gain is less than 3, the output is maintained at 3. In the high frequency range, if the step frequency is greater than 1.3Hz (high frequency threshold), the dynamic scaling factor based on the step frequency is maintained at 1.4. If the calculated feedback gain is greater than 15, the output is maintained at 15 (saturation limit).
[0077] The low-to-mid frequency band is 0.5Hz to 1.0Hz; the high-frequency threshold is 1.2Hz to 1.5Hz; the gain saturation limit is 10 to 20; and the sampling frequency is set to 50-200Hz.
[0078] S5 generates an auxiliary torque based on the motion angle signal, feedback gain, and equivalent phase lag time;
[0079] Dynamically calculated and Import the torque generation module, calculate the basic auxiliary torque, and distribute it to the bilateral joint drive execution module. The torque output formula based on DOFC is:
[0080]
[0081]
[0082] in and This represents the real-time angle values of the left and right legs. For the calculated , For the calculated .
[0083] y(t) is a function representing the difference in angles between the left and right leg joints, defined as:
[0084]
[0085] The auxiliary torque τ(t) is used to drive the exoskeleton joint to output assistance.
[0086] After the auxiliary torque is calculated, the torque value output to the left leg is τ, and the torque value output to the right leg is -τ, that is, the torques are the same in magnitude but output in opposite phases.
[0087] S6 is used to convert the calculated auxiliary torque into a motor drive control signal to drive the exoskeleton joints to perform corresponding auxiliary actions. The various parts are connected by signals to form a closed-loop control structure.
[0088] The invention also includes a safety monitoring step, in which the system immediately cuts off the torque output when any of the following conditions are detected:
[0089] (1) When the joint angular velocity is lower than the preset threshold and the output torque is continuously higher than the safety threshold, it is judged as a stalled state; (2) When the joint angle reaches the deep buckling limit and is accompanied by high torque output, it is judged as an overload state.
[0090] For example, in a normal walking scenario, if there is a torque output and the movement suddenly stops, or if a large torque value is calculated due to a lunge motion of the front and back legs but no forward movement occurs, the output will be interrupted. That is, the safety monitoring step has a higher priority than the feedback gain adjustment step.
[0091] This invention employs an adaptive oscillator to acquire step frequency in real time. By combining the body's motion phase and calculating the delay of the compensation filter, the controller's delay can be precisely aligned to the moment of highest assist efficiency (90° delay). A feedback gain optimization method is designed, incorporating both body weight and step frequency, allowing the gain to have different values for different body weights and step speeds.
[0092] This embodiment is applied to a single-degree-of-freedom lower limb exoskeleton at the hip joint. The parameters of each part are as follows:
[0093] Sampling frequency: Left and right, sample the angle values of the left and right legs:
[0094] Signal preprocessing uses a first-order low-pass filter for noise reduction, and the filter update method is as follows:
[0095]
[0096] in For the filter coefficients, in this embodiment According to the system frequency response test, when the step frequency is approximately... At that time, the equivalent time lag introduced by the filter .
[0097] The effective range of step frequency is .
[0098] Dead zone threshold set to Frequency lock is maintained when the frequency change in adjacent cycles is less than the threshold.
[0099] The limit range is set as follows: .
[0100] : .
[0101] : .
[0102] : .
[0103] Step scaling factor Using piecewise function form:
[0104] hour, ;
[0105] interval Depend on Linear change to ;
[0106] interval Depend on Linear change to ;
[0107] hour, ;
[0108] When the joint angular velocity is less than And the output torque is greater than Continue to exceed When this occurs, the system is determined to be in a stalled state and the output is cut off.
[0109] To verify the effectiveness of the gain and delay dual adaptive feedback control method described in this invention, a variable walking speed simulation test condition was designed, accelerating from a slow walk (0.8Hz) to a fast run (1.5Hz). The test results are as follows: Figures 2 to 4 As shown.
[0110] like Figure 2 The AFO cadence dynamic following effect diagram shown demonstrates that, after incorporating the cadence dead zone locking mechanism and smoothing logic, the AFO can smoothly and accurately follow the dynamic changes of the human body's actual cadence, without jitter or spikes in the initial stage and steady-state range, effectively suppressing frequent jumps in the underlying control commands.
[0111] like Figure 3 The spatiotemporal collaborative parameter adaptive mapping effect diagram shown illustrates the feedback dynamic delay time during acceleration. As the step frequency increases, it automatically and smoothly converges to a lower value (to compensate for the shortened gait period at high frequencies); at the same time, the feedback gain increases significantly with the step frequency after crossing the mid-low frequency band, and smoothly transitions to the saturation limit after reaching the high frequency threshold, perfectly realizing the dynamic coordination of multi-dimensional parameters.
[0112] like Figure 4 The diagram shown illustrates the effect of adaptive torque output and timing alignment under variable frequency operating conditions. The solid gray line represents the joint differential state, and the dashed black line represents the auxiliary torque generated by the system. This invention, through inverse delay compensation combined with the inherent hysteresis constant of the filter, ensures that the torque amplitude converges smoothly at low step frequencies, guaranteeing the smoothness of walking and human balance. At high step frequencies, the torque amplitude automatically increases, and the phase of the torque waveform remains precisely aligned with the joint state waveform throughout the entire speed change process (i.e., achieving perfect 90° hysteresis force application). This completely eliminates the undesirable reverse resistance (parasitic negative work) generated by traditional fixed delay control during frequency conversion, significantly improving the human-machine interaction efficiency of the exoskeleton.
[0113] Those skilled in the art will readily understand that the above description is merely an embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for delay feedback control of an exoskeleton based on dual adaptive gain and delay, characterized in that, include: The motion angle signals of the left and right legs of the exoskeleton are collected. After noise reduction, the equivalent phase lag time is calculated based on the phase difference of the filtered motion angle signals. An adaptive frequency oscillator model is constructed to obtain the instantaneous gait frequency estimate of the human body based on the motion angle signal; Calculate the feedback dynamic delay time based on the instantaneous step frequency estimate; The feedback gain is dynamically calculated based on human body parameters and human motion state, specifically as follows: in: For reference gain, For the user's weight, For reference weight, This is a dynamic scaling factor based on step frequency. Adjust the coefficients to be personalized for each user; An auxiliary torque is generated based on the motion angle signal, feedback gain, and equivalent phase lag time. The auxiliary torque is converted into a motor drive control signal, which drives the exoskeleton joints to perform corresponding auxiliary actions.
2. The exoskeleton delay feedback control method according to claim 1, characterized in that, It also includes a step frequency dead zone locking mechanism, which keeps the step frequency estimate of the previous cycle unchanged as the step frequency output value when the step frequency change in adjacent cycles is less than the set dead zone threshold; and updates the current step frequency output value when the step frequency change is greater than the set dead zone threshold.
3. The exoskeleton delay feedback control method according to claim 1, characterized in that, The feedback dynamic delay time is calculated based on the instantaneous step frequency estimate, specifically as follows: in, The theoretical delay time required to achieve an ideal 90° phase lag; The inherent hysteresis constant of the filter, This is the estimated instantaneous step frequency.
4. The exoskeleton delay feedback control method according to claim 1, characterized in that, The constraint condition for the feedback gain is: When the user's weight increases, the feedback gain is adjusted positively accordingly; when the instantaneous cadence estimate is in the low to mid-frequency range, the gain increases with the cadence or remains at a preset low limit; when the instantaneous cadence estimate reaches the high-frequency threshold, the feedback gain remains at a preset saturation limit.
5. The exoskeleton delay feedback control method according to claim 1, characterized in that, The auxiliary torque calculation process is as follows: in: For dynamic feedback gain, y represents the dynamic delay time, and y represents the difference in the angles of the left and right leg joints.
6. The exoskeleton delay feedback control method according to claim 1, characterized in that, It also includes a safety monitoring step that immediately cuts off the auxiliary torque output when any of the following conditions are detected: The joint angular velocity is lower than the preset threshold and the output torque is consistently higher than the safety threshold; the joint angle reaches the deep flexion limit and is accompanied by high torque output.
7. The exoskeleton delay feedback control method according to claim 1, characterized in that, The step frequency scaling factor adopts a piecewise function form, specifically: hour, ; interval Depend on Linear change to ; interval Depend on Linear change to ; hour, .
8. The exoskeleton delay feedback control method according to claim 2, characterized in that, The dead zone threshold is set to 0.05Hz.
9. The exoskeleton delay feedback control method according to claim 1, characterized in that, The noise reduction process uses a first-order low-pass filter.
10. The exoskeleton delay feedback control method according to claim 5, characterized in that, The torque command output to the left leg motor is the same in magnitude but opposite in phase as the torque command output to the right leg motor.