Pneumatic muscle multi-axis manipulator obstacle avoidance track correction method and control method

Abstract

The invention relates to the technical field of humanoid soft robot control, and provides a pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction method and control method.The pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction method comprises the steps that a target trajectory of a pneumatic muscle multi-axis manipulator is obtained; for the target track, a Mercator projection method is adopted, and the target track is converted into a two-dimensional plane track; the relative position and shape of an obstacle are obtained, external force is obtained through calculation, after the external force is introduced into a track generation equation of a dynamic motion primitive algorithm, based on a two-dimensional plane track, an obstacle avoidance track of the pneumatic muscle multi-axis manipulator is generated through the dynamic motion primitive algorithm, and inverse transformation is conducted through a Mercator projection method. The problem that traditional trajectory planning is lagged in adjustment and insufficient in precision in a complex environment is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of humanoid soft robot control, and particularly relates to an obstacle avoidance trajectory correction method and a control method for a pneumatic muscle multi-axis manipulator. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] PAM (Pneumatic Artificial Muscle) has the advantages of light weight, high power, high power-to-weight ratio, no need to consider mechanical friction / wear, high reliability, etc., and exhibits good bionic characteristics and environmental compatibility. Compared with rigid actuators, PAM has significant advantages in human-computer interaction, can more realistically cooperate with / replace humans to complete precision operation tasks, and has the characteristics of high compliance controllability and high safety.

[0004] Currently, the pneumatic muscle soft robot has the following important problems to be solved urgently: (1) Traditional trajectory planning cannot avoid obstacles in a timely manner in unstructured scenarios (such as multiple obstacles and irregular obstacles), and there are problems of adjustment lag and insufficient accuracy.

[0005] (2) The problem of flexible structure vibration and overshoot. Although the flexible driving characteristics of pneumatic muscles endow them with bionic compliance advantages, the low-stiffness structure is easily affected by the step change of air pressure, multi-degree-of-freedom coupling interference and external impact, resulting in low-frequency mechanical resonance and trajectory tracking overshoot. Traditional control methods (such as linear feedback) cannot dynamically adjust the air pressure response rate, resulting in phase lag and high-frequency chatter in the actuator output, which exacerbates the vibration amplitude of the flexible structure and even causes mechanical fatigue damage. In addition, the delay characteristics and non-linear coupling of pneumatic muscles further amplify the overshoot of trajectory tracking, which is likely to cause target deviation or safety hazards in precision operations (such as grasping and interaction). There is an urgent need for a control mechanism that takes into account both dynamic suppression and tracking accuracy.

[0006] (3) Difficulties in modeling non-linear dynamic characteristics. Due to the complex physical characteristics of pneumatic muscles, such as high non-linearity and anisotropy, the traditional Lagrangian model cannot meet the requirements of modeling accuracy. For example, when describing the instantaneous pressure-deformation characteristics of pneumatic muscles, due to their hysteretic characteristics, the Lagrangian model cannot accurately capture their instantaneous stress-strain response. At the same time, due to the significant differences in different directions caused by load and disturbance, the Lagrangian model of pneumatic muscles has strong asymmetry and cannot be controlled by traditional modeling methods. In addition, parameters such as the equivalent stiffness and damping of pneumatic muscles drift in real time with air pressure fluctuations, material aging, and external load changes, resulting in a decline in stability and deterioration of tracking accuracy in the actual application of control strategies based on fixed parameter models (such as PID and sliding mode control). Existing methods rely on high-precision modeling or complex parameter identification and are difficult to adapt to dynamic working conditions (such as sudden load changes and interactive force disturbances), severely restricting the application reliability of pneumatic muscles in high-precision bionic actuators (such as bionic arms and rehabilitation robots).

[0007] (4) Insufficient motion protection mechanism. Pneumatic muscles have a one-way constraint, and excessive contraction may cause excessive tension on the antagonist muscles, leading to mechanical system damage or even instability. In actual applications, excessive contraction will cause excessive gas pressure inside the muscle, resulting in material fatigue or even rupture, while excessive relaxation may cause a decline in the elastic recovery performance of the muscle, weakening the motion control accuracy. For example, when a bionic robotic arm is used for precise operation tasks (such as industrial assembly or medical surgery), if the tension on the antagonist muscle is too large, it may not only cause the actuator to exceed the control range, but also damage the operation target or threaten the safety of patients due to out-of-control end force. To prevent similar problems, an effective motion protection mechanism needs to be designed, such as angle constraint and speed constraint. Summary of the Invention

[0008] In order to solve the technical problems existing in the above background technology, the present invention provides a method for correcting the obstacle avoidance trajectory and a control method for a pneumatic muscle multi-axis manipulator, which are designed based on the fusion of the Mercator projection algorithm and the DMP algorithm, and an external force is introduced into the motion generation equation of the DMP algorithm to adjust the trajectory to meet the obstacle avoidance requirements, realizing the efficient trajectory mapping from the joint space of the bionic hand to the two-dimensional plane and the real-time obstacle avoidance adjustment, and solving the problems of lag and insufficient accuracy in traditional trajectory planning in complex environments.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions: The first aspect of the present invention provides a method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator, which includes: Obtain the target trajectory of the pneumatic muscle multi-axis manipulator; For the target trajectory, use the Mercator projection method to convert the target trajectory into a two-dimensional plane trajectory; Obtain the relative position and shape of the obstacle, calculate the external force, introduce the external force into the trajectory generation equation of the dynamic movement primitive algorithm, and based on the two-dimensional plane trajectory, generate the obstacle avoidance trajectory of the pneumatic muscle multi-axis manipulator through the dynamic movement primitive algorithm, and perform inverse transformation through the Mercator projection method.

[0010] Further, the trajectory generation equation after introducing the external force is:

[0011]

[0012] Where, x is the current position of the pneumatic muscle multi-axis manipulator, v is the speed of the pneumatic muscle multi-axis manipulator, is the time scaling factor, K and D are the elastic and damping coefficients of the pneumatic muscle multi-axis manipulator, g is the target position, x 0 is the initial position, is a non-linear function, F obstacle is the external force.

[0013] Further, the external force is calculated using a repulsive force model.

[0014] Further, the external force is expressed as:

[0015] Where, is a constant representing the intensity of the force; is the current position of the pneumatic muscle multi-axis manipulator; x nearest is the point on the obstacle surface closest to the system; is the decay exponent.

[0016] The second aspect of the present invention provides a control method for a pneumatic muscle multi-axis manipulator, including: For the obstacle avoidance trajectory obtained by the obstacle avoidance trajectory correction method for a pneumatic muscle multi-axis manipulator as described in the first aspect, after performing trajectory discretization and kinematic inverse solution calculation, obtain the target angle values of each joint; Based on the target angle values of each joint, obtain a control signal through an adaptive controller to control the movement of each joint.

[0017] Further, the adaptive controller is constructed by combining a deep neural network and a backstepping control method.

[0018] The third aspect of the present invention provides an obstacle avoidance trajectory correction system for a pneumatic muscle multi-axis manipulator, including: A target acquisition module, which is configured to: acquire the target trajectory of the pneumatic muscle multi-axis manipulator; A projection module, which is configured to: for the target trajectory, use the Mercator projection method to convert the target trajectory into a two-dimensional plane trajectory; An obstacle avoidance module, which is configured to: acquire the relative position and shape of the obstacle, calculate the external force, introduce the external force into the trajectory generation equation of the dynamic motion primitive algorithm, and based on the two-dimensional plane trajectory, generate the obstacle avoidance trajectory of the pneumatic muscle multi-axis manipulator through the dynamic motion primitive algorithm, and perform inverse transformation through the Mercator projection method.

[0019] The fourth aspect of the present invention provides a control system for a pneumatic muscle multi-axis manipulator, including: A trajectory discretization module, which is configured to: for the obstacle avoidance trajectory obtained by using the obstacle avoidance trajectory correction system for a pneumatic muscle multi-axis manipulator described in the third aspect, perform trajectory discretization and inverse kinematics calculation to obtain the target angle values of each joint; A control module, which is configured to: based on the target angle values of each joint, obtain a control signal through an adaptive controller to control the movement of each joint.

[0020] The fifth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in an obstacle avoidance trajectory correction method for a pneumatic muscle multi-axis manipulator or a control method for a pneumatic muscle multi-axis manipulator as described above are implemented.

[0021] The sixth aspect of the present invention provides a computer device, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor. When the processor executes the program, the steps in an obstacle avoidance trajectory correction method for a pneumatic muscle multi-axis manipulator or a control method for a pneumatic muscle multi-axis manipulator as described above are implemented.

[0022] Compared with the prior art, the beneficial effects of the present invention are: The present invention is based on the fusion design of the Mercator projection algorithm and the DMP (Dynamic Movement Primitives) algorithm. By introducing an external force into the motion generation equation of the DMP algorithm to adjust the trajectory to meet the obstacle avoidance requirements, it realizes the efficient trajectory mapping from the joint space of the bionic hand to the two-dimensional plane and real-time obstacle avoidance adjustment. It solves the problems of lagging adjustment and insufficient accuracy of traditional trajectory planning in complex environments. Moreover, through the dynamic trajectory planning mechanism, while ensuring the accuracy of the target tracking error, it can complete the obstacle avoidance response at different set speeds, significantly improving the operation efficiency and safety of the bionic manipulator in unstructured scenarios (such as multiple obstacles and irregular obstacles), laying a solid foundation for the flexible application of the manipulator in complex environments.

[0023] Through the in-depth integration of the intelligent neural network control based on the backstepping method of pneumatic muscles and the dynamic obstacle avoidance algorithm, the present invention successfully constructs an autonomous environment perception - decision - execution closed-loop for the bionic manipulator in the interactive scenarios of single obstacle avoidance operation tasks and multi - obstacle avoidance operation tasks. It can dynamically plan the obstacle avoidance path according to the obstacle information real - time feedback by sensors during the operation process, adjust the parameters of the optimal controller in real - time under state constraints, and track the monitoring error. This not only improves the environmental adaptability of the bionic manipulator, but also provides core technical support for its multi - target operation requirements in fields such as medical rehabilitation and dangerous operations, significantly enhancing the task collaborative control ability of the manipulator.

[0024] The multi - dimensional trajectory planning and dynamic obstacle avoidance technology proposed by the present invention, as well as the significant improvement in the multi - modal environment adaptability and task collaborative control ability, greatly enrich the application scope of the bionic manipulator and significantly enhance its operation flexibility. In fields such as medical rehabilitation that require extremely high operation precision, and in dangerous operation scenarios facing complex and changeable environmental challenges, the present invention can provide solid technical support for the bionic manipulator to ensure that it can efficiently and safely complete various tasks. This innovation not only significantly improves the practical value and market competitiveness of the manipulator, but also injects strong new impetus into the development of the intelligent and automated processes in related fields.

[0025] The present invention proposes a dynamic damping injection technology, which introduces a virtual damping term obtained based on the real - time state into the backstepping framework to dynamically adjust the air pressure response rate of the pneumatic muscle and suppress low - frequency mechanical resonance; a non - linear dynamic surface control (DSC), which uses a low - pass filter to replace the step - by - step differential operation of the traditional backstepping method to smooth the virtual control signal, avoid high - frequency noise amplification, and reduce phase lag; a neural network feed - forward compensation: using a neural network to predict the delayed response of the pneumatic muscle and generate a feed - forward air pressure command to compensate for the overshoot caused by non - linear coupling. It realizes the technical effects of suppressing resonance, dynamically adapting, and controlling overshoot.

[0026] The present invention proposes a neural network disturbance observer, which regards the unmodeled dynamics (hysteresis, asymmetry) of pneumatic muscles as lumped disturbances, approximates and compensates them online by a neural network, and does not rely on an accurate mathematical model; an adaptive backstepping framework, which recursively designs an adaptation law to update the neural network weights and parameter estimates in real time, and tracks and compensates for time-varying factors such as air pressure fluctuations and material aging. The technical effects of model independence and parameter adaptability are achieved.

[0027] The present invention proposes an adaptive adjustment of the air pressure safety threshold, which predicts the muscle contraction limit based on a neural network and dynamically adjusts the upper limit of the air pressure command to avoid material rupture or elastic failure; fault detection and fault-tolerant control: an integrated state observer monitors the muscle pressure and deformation in real time, and an integrated fault observer ensures that an emergency decompression or braking mechanism is triggered to ensure safe shutdown before instability. The technical effects of safety protection and precise dynamic constraint are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings forming a part of this invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0029] Figure 1 is a flowchart of a method for obstacle avoidance trajectory correction of a pneumatic muscle multi-axis manipulator according to Embodiment 1 of the present invention; Figure 2 is a schematic diagram of target trajectory conversion according to Embodiment 1 of the present invention; Figure 3 is a schematic diagram of obstacle avoidance trajectory planning in a two-dimensional plane according to Embodiment 1 of the present invention; Figure 4 is a schematic diagram of obstacle avoidance trajectory planning on a three-dimensional spherical surface according to Embodiment 1 of the present invention; Figure 5 is a control flowchart according to Embodiment 2 of the present invention; Figure 6 is a schematic diagram of the sine tracking result of training and simulation according to Embodiment 2 of the present invention; Figure 7 is a simulation tracking result effect diagram of the obstacle avoidance trajectory according to Embodiment 2 of the present invention; Figure 8 is a curve graph of error change according to Embodiment 2 of the present invention; Figure 9 is a structural schematic diagram of a computer device according to Embodiment 6 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0031] It should be noted that the following detailed description is illustrative and aims to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention pertains.

[0032] Embodiment 1 This embodiment provides a method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator.

[0033] In terms of trajectory planning, the goal of this embodiment is to design a high-quality path planning scheme suitable for a pneumatic artificial muscle arm, and at the same time, it can reflect the obstacle avoidance planning for different types of obstacles. For this research goal, the present invention has designed a constrained motion obstacle avoidance algorithm based on Dynamic Movement Primitives (DMP) through exploration.

[0034] The theoretical basis of the method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator provided in this embodiment: the motion obstacle avoidance algorithm of DMP; the Mercator projection method.

[0035] The method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator provided in this embodiment, based on the idea of the Mercator projection method, transforms the target trajectory on the spherical surface in the joint space to a two-dimensional plane, then uses the obstacle avoidance algorithm of DMP to perform trajectory planning in the plane of the projected trajectory, and then performs inverse transformation to output joint angles, realizing precise control of the bionic wrist joint and precise obstacle avoidance at the fingertip.

[0036] The method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator provided in this embodiment, as Figure 1 shown, includes the following steps: Step 1, Initial setting.

[0037] In order to solve the motion planning and obstacle avoidance problems of the robotic arm using mathematical methods, the present invention selects the joint space for modeling. Currently, the wrist of the bionic robotic arm is fixed, and only the up-and-down and left-and-right rotations of the wrist joint are considered, that is, two angular variables are considered.

[0038] Step 2, Selection of the target trajectory.

[0039] In order to provide sufficient complexity to verify the obstacle avoidance algorithm and trajectory learning ability of the bionic hand, the present invention generates a target trajectory (example trajectory or original trajectory) through mathematical function calculation. For example, a curve trajectory on a three-dimensional spherical shell is generated through trigonometric functions.

[0040] Among them, the mathematical function is expressed as:

[0041] Among them, q d(t) is t the target joint angle vector (in radians) at a moment; θ 1 (t), θ 2 (t) are the left - right rotation and up - down rotation angles of the wrist joint; A 1 ,A 2 is the amplitude (typical value 0.5 - 1.5 rad); ω 1 , ω 2 is the angular frequency (rad / s), which determines the movement speed; and is the phase offset (initial phase angle); θ 10 , θ 20 is the joint center position (default zero position).

[0042] Step 3: Projection of the target trajectory.

[0043] Inspired by the method of mapping the Earth, the present invention selects the Mercator projection method to transform the target trajectory on the spherical surface into a two - dimensional planar trajectory, as Figure 2 shown.

[0044] The Mercator projection is an orthographic equiangular cylindrical projection method that can maintain angular relationships and is suitable for trajectory tracking and obstacle avoidance planning of three - dimensional spherical trajectories on a two - dimensional plane.

[0045] Through this projection, the original spherical trajectory can be simplified into a path in the two - dimensional plane, making the trajectory planning calculation more convenient.

[0046] Step 4: Trajectory planning based on the DMP algorithm.

[0047] DMP is a trajectory generation and control method based on a non - linear dynamic system that can generate a smooth path between a given initial position and a target position.

[0048] In this embodiment, the DMP algorithm is used to process the planar trajectory after Mercator projection, and an external force is introduced into the basic motion equation to adjust the trajectory to meet the obstacle avoidance requirements. This algorithm uses the basic framework of an elastic - damping system and combines a learned non - linear trajectory generation function and a dynamic obstacle avoidance external force to achieve obstacle avoidance and flexible motion planning.

[0049] The motion generation equation of DMP mainly includes the following form: ; ; Among them, x is the current position of the non - linear dynamic system (i.e., the pneumatic muscle multi - axis manipulator), v is the velocity of the non - linear dynamic system, is the time scaling factor, K and D are the elasticity and damping coefficients of the non - linear dynamic system, g is the target position, x 0 is the initial position, is a non - linear function responsible for introducing the non - linear behavior of learning, depending on the phase variable s .

[0050] (2) Phase variable and non - linear term.

[0051] The phase variable s controls the progress of the movement and is usually defined as ; among them, α s is a positive time constant that determines the speed at which the phase variable s decays from the initial value (usually 1) to 0.

[0052] The non - linear function is usually obtained by linearly combining Gaussian basis functions to approximate the shape of the training trajectory, i.e.:

[0053] Among them, is the basis function, is the weight, N represents the total number of Gaussian basis functions used to approximate the shape of the training trajectory .

[0054] (3) Introduction of the obstacle avoidance mechanism.

[0055] In the obstacle avoidance scenario, the external force F obstacle is introduced into the acceleration equation of the DMP to modify the original trajectory to avoid obstacles, i.e.:

[0056] (4) External force calculation.

[0057] The external force F obstacle is usually calculated based on the relative position and shape of the obstacle to ensure that the trajectory deviates from the dangerous area. For example, when the obstacle is near the preset trajectory, the external force is calculated using the repulsive force model as follows:

[0058] Among them, is a constant representing the intensity of the force; is the current position of the system; x nearest is the point on the obstacle surface closest to the system; x obstacle is the position of the obstacle; is the attenuation exponent; Shape is the obstacle shape, and d(x, Shape) represents the closest distance from the current position x of the system to the obstacle surface.

[0059] After the above calculations, taking an elliptical obstacle as an example, a reasonable obstacle avoidance trajectory is planned as follows Figure 3 as shown.

[0060] Step 5: Inverse transformation of the coordinates after trajectory planning, map it back to the three-dimensional joint space to generate the target joint angles of the bionic robotic arm. Specifically, the two-dimensional trajectory optimized by obstacle avoidance is restored to the target trajectory on the sphere through the inverse transformation of the Mercator projection to generate the motion path of the wrist joint; based on the projected trajectory, generate the target joint angle values to guide the movement of the pneumatic muscles.

[0061] As Figure 4 shown, on the three-dimensional sphere, Trajectory1 is the target trajectory for learning, and Trajectory2 is the actual trajectory after planning obstacle avoidance.

[0062] To verify the effectiveness of the algorithm, this embodiment conducts a preliminary simulation. By adjusting the dynamic parameters of the DMP, the motion effects of the bionic arm under obstacles of different shapes are observed. The preliminary results show that the Mercator projection method combined with the DMP obstacle avoidance algorithm can effectively guide the bionic arm to avoid obstacles, with smooth path generation and high controllability.

[0063] A method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator provided in this embodiment is designed based on the fusion of the Mercator projection algorithm and the DMP algorithm, and an external force is introduced into the motion generation equation of the DMP algorithm to adjust the trajectory to meet the obstacle avoidance requirements. It realizes the efficient trajectory mapping from the joint space to the two-dimensional plane and real-time obstacle avoidance adjustment of the bionic hand, breaking through and solving the problems of lagging adjustment and insufficient accuracy of traditional trajectory planning in complex environments. Through the dynamic trajectory planning mechanism, while ensuring the accuracy of the target tracking error, it can complete the obstacle avoidance response at different set speeds, significantly improving the operation efficiency and safety of the bionic robotic arm in unstructured scenarios (such as multiple obstacles and irregular obstacles).

[0064] Embodiment 2 This embodiment provides a control method for a pneumatic muscle multi-axis manipulator.

[0065] To build a stable controller for a typical non - linear system such as a pneumatic artificial muscle arm, this embodiment intends to combine a deep neural network (DNN) and a backstepping control method to build an intelligent controller, so as to achieve the purpose of tracking and controlling the trajectory of a Lagrangian system.

[0066] Its core idea is to use a deep neural network (DNN) to approximate the Lagrangian function. In addition, different convolutional neural networks and feed - forward convex neural networks (FCNN) are used to design the virtual control law in the controller to compensate for the time - delay error and external interference in the system, simulate the damping term of the controller, and combine to obtain the control law. In this way, through the data real - time feedback by the platform, during the movement process, the controller will autonomously adjust the parameters of the virtual control rate and the damping term to achieve the effect of adaptive control, improve the robustness and stability of the non - linear system, and effectively solve the challenges brought by characteristics such as pneumatic muscle hysteresis and creep to precise control.

[0067] A control method for a pneumatic muscle multi - axis manipulator provided by this embodiment has the following theoretical basis: Lagrangian equation; fault observer; deep neural network, Lagrangian neural network, convex neural network.

[0068] The design idea of a control method for a pneumatic muscle multi - axis manipulator provided by this embodiment is as follows: (1) Control system modeling and fault detection.

[0069] First, according to the physical structure, a dynamic model of a PAM - driven robotic hand is established, and the relationship between system faults and time is considered. The model is as follows:

[0070] Among them, is the mass matrix of the fitted Lagrangian model, describing the inertial characteristics of the system; is the Coriolis matrix, describing the velocity - like management generated due to the system movement; is the gravity vector, describing the generalized force caused by potential energy; D is the damping force vector, including dissipation effects such as friction; is the joint angle vector, is the control input, u is the state of the system, that is, the control input, is the parameter matrix, represents the relationship between system faults and time, is the real - time fault function.

[0071] For the convenience of stability analysis, the above model is rewritten as: ; .

[0072] Among them, It is the converted fault function. Considering that it is difficult to obtain the physical characteristics of pneumatic muscles due to their complex structure, a deep neural network is used to fit the Lagrangian model for modeling. By obtaining the system states and velocities under natural conditions and different input air pressures with different degrees of freedom, noise processing and normalization are performed on the data, and a dataset is constructed for training to obtain an offline controlled object model to solve the physical modeling difficulty of Challenge 3.

[0073] (2) Introduce a trajectory corrector based on Dynamic Movement Primitives (DMP).

[0074] In the front-end design of the controller, the present invention introduces a trajectory corrector based on Dynamic Movement Primitives (DMP). The core idea is to model the trajectory generation problem as a spring-damper system perturbed by an external correction term. DMP dynamically adjusts the endpoints and shape of the trajectory through a non-linear function to achieve precise correction of the reference trajectory. The specific form is as follows:

[0075]

[0076] Among them, x is the current position of the non-linear dynamic system (i.e., the multi-axis manipulator with pneumatic muscles), v is the velocity of the non-linear dynamic system, is the time scaling factor, K and D are the elastic and damping coefficients of the non-linear dynamic system, g is the target position, x 0 is the initial position, is a non-linear function responsible for introducing the learned non-linear behavior, depending on the phase variable s .

[0077] (3) Design a fault observer.

[0078] In this embodiment, a fault observer is designed to detect faults in the system in real time.

[0079] The dynamics of the fault observer are as follows:

[0080] Among them, x is the actual state vector of the system (which can be directly measured), is the amount of control input, is the observer gain to be designed, is the parameter matrix, is the estimated value of, is a non - linear state function that describes the system dynamic model (based on the estimated state and the control input u).

[0081] (4) Development of an adaptive controller based on a deep neural network.

[0082] Thus, combining the previous physical modeling, trajectory corrector, and fault observer based on Lagrange's theorem, the present invention proposes an adaptive tracking controller based on a deep neural network (DNN) for simultaneously compensating for unknown dynamics / disturbances and adjusting the optimal control gain. The adaptive controller based on the backstepping method is designed as follows:

[0083] where, is the mass matrix of the fitted Lagrangian model, which describes the inertial characteristics of the system; is the Coriolis matrix, which describes the velocity - like management due to the system motion; is the gravity vector, which describes the generalized force caused by the potential energy; is the tracking error, is the error signal related to the velocity.

[0084] During the design process of the adaptive tracking controller, a Lagrangian neural network is introduced to approximate the unknown Lagrangian dynamics model of the system for estimating the kinetic and potential energies of the whole system, that is, the fully - connected - layer Lagrangian neural network is as follows:

[0085] Here, and are the estimated values of the kinetic energy and potential energy respectively, and are the parameters to be optimized. Considering the high - dimensional and non - linear nature of the model, the present invention selects the Kriging model and the surrogate - based optimization algorithm (SBO) to obtain the optimal parameters and , and obtains

[0086] In addition, in order to compensate for the disturbance, approximate the damping term, and dynamically adjust the optimal control gain in the controller, an auxiliary function composed of different FICNN networks and two virtual control rates , are introduced, which not only ensures the convexity of the neural network term but also realizes the effect of adaptive control.

[0087] For the design of the neural network in the controller, multiple fully connected layers such as LNN and FICNN, and deep convolutional neural networks, approximate Lagrangian models, virtual control rates, and auxiliary functions are constructed.

[0088] The present invention constitutes a neural network training dataset by collecting the states of the pneumatic muscle arm under different inputs. A fully connected layer neural network is built to model the LNN model, and multiple deep convolutional neural networks are introduced to fit the virtual control rate and damping matrix. While online adaptively adjusting parameters and performing simulations, it tracks the sine-schematic target trajectory, and the results are as Figure 6 shown.

[0089] Combined with the dynamic obstacle avoidance function in the input correction part, obstacle avoidance simulations are carried out in a single-obstacle environment in the spherical coordinate system. The planned trajectory and the actual tracking trajectory are as Figure 7 shown, and the recording of the tracking error is as Figure 8 shown.

[0090] In summary, a control method for a pneumatic muscle multi-axis manipulator provided in this embodiment is as Figure 5 shown, and specifically includes the following steps: Step 1, trajectory discretization: For the obstacle avoidance trajectory obtained by a method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator provided in Embodiment 1, the continuous trajectory is discretized into a series of discrete path points in the time domain, and parameters such as the angles, positions, or velocities of each joint of the manipulator at each path point are obtained. These parameters serve as the target values for subsequent control.

[0091] Step 2, inverse kinematics calculation: According to the coordinates of the discrete path points, using the inverse kinematics algorithm of the manipulator, the desired position and attitude of the end effector are converted into the target angle values of each joint.

[0092] Step 3, based on the target angle values (obstacle avoidance trajectory) of each joint and the actual state of the system, an error signal is obtained and input into the fault observer. Through the analysis and calculation of the residuals, a fault quantification signal is output to monitor the sensor deviation and the actuator saturation degree, so as to view and quantify the faults of the system in real time.

[0093] Step 4, the fault information output by the fault observer, combined with the feedback signal and the filtered state output by the closed-loop filter, through recursive identification and Lyapunov stability analysis, obtains the output of the time-varying parameter estimation link, that is, the estimated dynamic parameters (time-varying parameters). The fault observer and the time-varying parameters in Steps 3 and 4 jointly ensure the normal operation of the motion protection mechanism by monitoring the faults and parameter changes during the operation of the monitoring system.

[0094] Step 5: Design and train the estimation network of the controlled object: In order to obtain a suitable backstepping control rate, a fully connected neural network is introduced to train the Lagrangian dynamics model that approximates the unknown system to estimate the kinetic energy and potential energy of the entire system. At the same time, an auxiliary function, which is essentially a convolutional neural network, is designed to be equivalent to an observer of the external disturbance torque and estimate the damping caused by the external disturbance.

[0095] The Lagrangian model estimation network is pre-trained offline using natural data from a pneumatic muscle multi-axis manipulator platform to obtain an offline fully connected layer Lagrangian neural network (LNN). In actual applications, the LNN inputs the feedback signal obtained through a closed-loop filter, namely the state and velocity signals, to obtain the inertia matrix, Coriolis force matrix, gravity matrix and other physical parameters of the Lagrangian estimation model.

[0096] The optimization and training of the external disturbance damping estimation network are synchronized with the actual movement of the pneumatic arm. The parameter initialization of this part of the network is completed offline, and the parameters of each level of the auxiliary function in the subsequent network are forwarded with the operation of the system and continuously updated and optimized. The auxiliary function in the external disturbance damping estimation network and the virtual control rate 1 constitute the external disturbance damping compensation term in the subsequent controller input. In actual application, the state and speed signal of the external disturbance damping estimation network are input, the parameters of the auxiliary function are updated, and the external disturbance damping compensation term is obtained by combining with the virtual control rate 1.

[0097] Step 6, controller design and intelligent control rate implementation: Based on the inertia matrix, Coriolis force matrix, gravity matrix and other physical parameters obtained by the Lagrangian neural network in step 3, the external disturbance damping compensation term obtained by the auxiliary function and the virtual control rate 1, the time delay compensation term obtained by the output signal of the closed-loop filter and the virtual control rate 2, the smooth feedback signal output by the closed-loop filter, and the real-time state of the controlled object obtained by the sensor, the control signal u is calculated through an adaptive controller based on the backstepping method as the input of the controlled object; the fault quantization signal obtained by the fault observer and the output of the time-varying parameter estimation are used as a constraint mechanism to cut off the operation of the control system when necessary.

[0098] Step 7: Convert the control signal u output by the controller into an appropriate drive signal to drive the movement of each joint of the pneumatic muscle multi-axis robotic arm (multi-muscle release soft robotic arm).

[0099] Step 8, Feedback and Correction: During the movement of the robotic arm, sensors (such as encoders, force sensors, vision sensors, etc.) are used to obtain real-time information about the actual position, posture, force, etc. of the robotic arm as feedback signals. And compare it with the desired trajectory (obstacle avoidance trajectory), and obtain an error signal through the feedback path; track the error signal, input the error signal and the feedback signal into a closed-loop filter to obtain a smoothed and denoised feedback signal, and then return it to the Lagrangian model estimation network and the external disturbance damping estimation network, continuously update and adjust the accurate state signals of the actuators in the system, as well as the virtual control rates 1 and 2 in the adaptive controller, so as to adjust the control signals in the next cycle and perform real-time correction on the movement of the robotic arm to make the steady-state error finally converge.

[0100] Embodiment III This embodiment provides an obstacle avoidance trajectory correction system for a pneumatic muscle multi-axis manipulator, which specifically includes: A target acquisition module, which is configured to: acquire the target trajectory of the pneumatic muscle multi-axis manipulator; A projection module, which is configured to: for the target trajectory, use the Mercator projection method to convert the target trajectory into a two-dimensional plane trajectory; An obstacle avoidance module, which is configured to: acquire the relative position and shape of the obstacle, calculate the external force, introduce the external force into the trajectory generation equation of the dynamic movement primitive algorithm, and generate the obstacle avoidance trajectory of the pneumatic muscle multi-axis manipulator based on the two-dimensional plane trajectory through the dynamic movement primitive algorithm, and perform inverse transformation through the Mercator projection method.

[0101] It should be noted here that each module in this embodiment corresponds to each step in Embodiment I one by one, and the specific implementation process is the same, so it will not be repeated here.

[0102] Embodiment IV This embodiment provides a control system for a pneumatic muscle multi-axis manipulator, which specifically includes: A trajectory discretization module, which is configured to: for the obstacle avoidance trajectory obtained by using the obstacle avoidance trajectory correction system for a pneumatic muscle multi-axis manipulator described in Embodiment III, after performing trajectory discretization and inverse kinematics calculation, obtain the target angle values of each joint; A control module, which is configured to: based on the target angle values of each joint, obtain control signals through an adaptive controller to control the movement of each joint.

[0103] It should be noted here that each module in this embodiment corresponds to each step in Embodiment II one by one, and the specific implementation process is the same, so it will not be repeated here.

[0104] Embodiment V This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in a method for correcting an obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator as described in Embodiment 1 above or a method for controlling a pneumatic muscle multi-axis manipulator as described in Embodiment 2 above.

[0105] Embodiment 6 This embodiment provides a computer device, as Figure 9 shown, including a display device, an input device, a computer-readable storage medium (volatile memory and non-volatile storage medium), a processor, a communication interface (i.e., a network interface), and a computer program stored on the computer-readable storage medium and executable on the processor. Among them, the processor, the communication interface, and the computer-readable storage medium can be connected through a bus or other means. Among them, the communication interface is used to receive and send data. When the processor executes the program, it implements the steps in a method for correcting an obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator as described in Embodiment 1 above or a method for controlling a pneumatic muscle multi-axis manipulator as described in Embodiment 2 above.

[0106] Among them, any reference to a memory, storage, database, or other medium provided in this application and used in the embodiments may include non-volatile and / or volatile memory. The non-volatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0107] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing the processFigure 1 means for the functions specified in one or more processes and / or blocks Figure 1 or blocks.

[0108] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means that implements the functions specified in one Figure 1 or more processes and / or blocks Figure 1 or blocks.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one Figure 1 or more processes and / or blocks Figure 1 or blocks.

[0110] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction method, characterized in that: include: Obtain the target trajectory of the pneumatic muscle multi-axis manipulator; For the target trajectory, the Mercator projection method is used to transform the target trajectory into a two-dimensional plane trajectory; The relative position and shape of the obstacle are obtained, and the external force is calculated. After the external force is introduced into the trajectory generation equation of the dynamic motion primitive algorithm, the obstacle avoidance trajectory of the pneumatic muscle multi-axis manipulator is generated based on the two-dimensional plane trajectory through the dynamic motion primitive algorithm, and the inverse transformation is performed through the Mercator projection method.

2. A pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction method as claimed in claim 1, characterized in that: The trajectory generation equation after the introduction of external force is: in, x is the current position of the pneumatic muscle multi-axis manipulator, v is the speed of the pneumatic muscle multi-axis manipulator, is the time scaling factor, K and D are the elasticity and damping coefficients of the pneumatic muscle multi-axis manipulator, g is the target location, x 0 is the initial position, is a nonlinear function, F obstacle For external force.

3. The method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator as claimed in claim 1, characterized in that: The external force is calculated using a repulsive force model.

4. The method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator according to claim 1, characterized in that: The external force is expressed as: in, is a constant, indicating the strength of the force; is the current position of the pneumatic muscle multi-axis manipulator; x nearest is the point on the obstacle surface closest to the system; is the decay exponent.

5. A pneumatic muscle multi-axis manipulator control method, characterized in that: include: For the obstacle avoidance trajectory obtained by the method for correcting the obstacle avoidance trajectory of a pneumatic muscle multi-axis manipulator as described in any one of claims 1 to 4, after performing trajectory discretization and kinematic inverse solution calculation, the target angle value of each joint is obtained; Based on the target angle value of each joint, a control signal is obtained through an adaptive controller to control the movement of each joint.

6. A pneumatic muscle multi-axis manipulator control method as claimed in claim 5, characterized in that: The adaptive controller is constructed by combining a deep neural network and a backstepping control method.

7. A pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction system, characterized in that: include: A target acquisition module is configured to: acquire a target trajectory of the pneumatic muscle multi-axis manipulator; The projection module is configured to: for the target trajectory, use the Mercator projection method to convert the target trajectory into a two-dimensional plane trajectory; The obstacle avoidance module is configured to: obtain the relative position and shape of the obstacle, calculate the external force, introduce the external force into the trajectory generation equation of the dynamic motion primitive algorithm, generate the obstacle avoidance trajectory of the pneumatic muscle multi-axis manipulator based on the two-dimensional plane trajectory through the dynamic motion primitive algorithm, and perform inverse transformation through the Mercator projection method.

8. A pneumatic muscle multi-axis manipulator control system, characterized in that: include: A trajectory discretization module is configured to: for the obstacle avoidance trajectory obtained by using the pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction system according to claim 7, after performing trajectory discretization and kinematic inverse solution calculation, obtain the target angle value of each joint; The control module is configured to obtain a control signal through an adaptive controller based on the target angle value of each joint to control the movement of each joint.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction method as described in any one of claims 1-4 or a pneumatic muscle multi-axis manipulator control method as described in any one of claims 5-6 are implemented.

10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored in the computer-readable storage medium and executable on the processor, characterized in that: When the processor executes the program, it implements the steps in a pneumatic muscle multi-axis manipulator obstacle avoidance trajectory correction method as described in any one of claims 1-4 or a pneumatic muscle multi-axis manipulator control method as described in any one of claims 5-6.

Images