A sparse motion planning method and system for robotic arms based on cross-species optimization

Abstract

The present invention discloses a sparse motion planning method and system for a manipulator based on cross-species optimization, which belongs to the field of manipulator control technology, and includes: S1: taking the joint angle angular velocity space energy of the manipulator as the objective function, constructing a constrained sparse motion planning model of the manipulator, and converting it into an unconstrained sparse motion planning model based on the Lagrange multiplier method; S2 constructing an error momentum neural network for solving the unconstrained sparse motion planning model in the current iteration, which includes a hidden layer and an output layer; S3: solving the output layer variables of the error momentum neural network model at different initial points, and using each output layer variable as a particle, searching for the optimal output layer variable using a particle swarm algorithm; S4: if the maximum number of iterations is not reached, then transfer to S2; otherwise, the sparse motion planning of the manipulator is realized according to the optimal output layer variables in each iteration. The present invention can quickly search for the global optimal solution and improve the solution quality of the sparse motion planning of the manipulator.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot arm control, and more specifically, relates to a robot arm sparse motion planning method and system based on cross-species optimization. Background Art

[0002] In the industrial field, industrial robots play a vital role in efficiently performing repetitive and demanding tasks. Redundant manipulators are a special type of industrial robot that provide enhanced flexibility and adaptability through additional degrees of freedom. This flexibility enables them to adapt to complex working environments and meet various task requirements, including obstacle avoidance, collision detection and multi-target operation, bringing advantages such as high productivity, reduced costs and less reliance on manual labor to industrial production.

[0003] Due to the additional degrees of freedom, redundant manipulators show remarkable flexibility and adaptability in task execution. At the same time, trajectory tracking of redundant manipulators is a difficult task. Motion planning of redundant manipulators mainly involves tracking the position of the end effector, which is crucial in many application scenarios. In the past few decades, many motion planning schemes have been constructed to address the motion control challenges of redundant manipulators, including joint angle-based control, joint velocity-based control, and joint acceleration-based control. Advanced techniques such as nonlinear model predictive control have also been used to improve trajectory tracking accuracy. However, the complexity of redundant manipulators, such as the existence of workspace constraints, brings additional challenges to trajectory tracking.

[0004] Sparse methods have been used in robotic control and other fields. In the field of motion planning, sparse control signals activate a limited number of joints to achieve a specific task, and the sparsity of the signal is usually quantified using a norm. By exploiting sparsity, the complexity of the action space can be reduced, the efficiency of motion planning can be improved, and the robot can perform actions more accurately and efficiently, minimize unnecessary movements, and improve overall performance. In some early works, such as the introduction of the concept of sparsity in human motion control and a mathematical model designed to explain the coordination of limb movements, the motion planning task was formulated as a constrained L1 norm optimization problem. The above studies focus on the application of sparsity in robot motion control in convex settings. However, the requirement for strict zero values ​​in convex sparse approximations imposes limitations on achieving high levels of sparsity.

[0005] With the continuous development of technology, neural dynamics methods and particle swarm optimization algorithms have also been applied to the sparse motion planning of robotic arms.

[0006] The brain-inspired neurodynamics method has the hardware implementation capability and biological interpretability to solve complex nonlinear optimization problems. The neurodynamics method has strong processing capabilities in solving robot motion planning problems. However, due to the introduction of sparsity in the robotic arm task planning problem, a single brain-inspired neurodynamics method is prone to fall into the local optimum, making it difficult to achieve the optimal solution for the sparse planning task of the robotic arm task.

[0007] The particle swarm optimization algorithm searches for the optimal solution by simulating the behavior of a flock of birds or a school of fish. Each particle represents a potential solution and updates its state based on its own experience and the collective experience of the group. At present, researchers have applied the particle swarm optimization algorithm to sparse motion planning of robotic arms. Although the bird-like particle swarm algorithm has global search capabilities, its computational complexity increases exponentially as the solution space increases.

[0008] In general, in the sparse motion planning of robotic arms, how to search for the global optimal solution at a faster speed is still a difficult problem that needs to be solved urgently. Summary of the invention

[0009] In view of the defects of the prior art and the need for improvement, the present invention provides a sparse motion planning method and system for a robotic arm based on cross-species optimization, whose purpose is to search for the global optimal solution at a faster speed in the sparse motion planning of the robotic arm, thereby improving the solution quality of the sparse motion planning of the robotic arm.

[0010] To achieve the above object, according to one aspect of the present invention, a sparse motion planning method for a robotic arm based on cross-species optimization is provided, comprising:

[0011] Step S1: Set the number of iterations k Initialize to 0; take the joint angle angular velocity space energy of the robot arm as the objective function, build the constrained sparse motion planning model of the robot arm, and transform it into an unconstrained sparse motion planning model based on the Lagrange multiplier method;

[0012] Step S2: In the current iteration, an error momentum neural network for solving the unconstrained sparse motion planning model is constructed, which includes a hidden layer and an output layer, and the expression is as follows:

[0013] ;

[0014] in, , , All are positive numbers, indicating the network weight coefficient; is the output layer variable, represents the joint angular velocity, λ ( k) represents the Lagrange multiplier, and the superscript T represents the transposition operation; y ( k ) represents the hidden layer variables; A ( k ) represents the mass matrix of the Lagrangian condition, b ( k ) represents the coupling vector of the Lagrangian condition;

[0015] Step S3: solving the output layer variables of the error momentum neural network model at different initial points, and using each output layer variable as a particle to search for the optimal output layer variable using a particle swarm algorithm;

[0016] Step S4: If the maximum number of iterations has not been reached, proceed to step S2; otherwise, implement the sparse motion planning of the robot arm according to the optimal output layer variables in each iteration.

[0017] Furthermore, the expression of the constrained sparse motion planning model is as follows:

[0018] ;

[0019] in, Q represents the weight matrix used to adjust the contribution of each joint angular velocity to the joint angular velocity space energy; Represents the angular velocity of the joint L p Norm, 0< p <1; Indicates the ideal speed of the end effector of the robot arm; J ( k ) represents the Jacobian matrix of the kinematic transfer function of the robot arm; and are the minimum angular velocity and the maximum angular velocity respectively.

[0020] Furthermore, the expression of the unconstrained sparse motion planning model is as follows:

[0021] ;

[0022] in, Represents an unconstrained sparse motion planning model.

[0023] Further,

[0024] ;

[0025] ;

[0026] in, Represents the angular velocity of the joint L p Norm For joint angular velocity The derivative of .

[0027] Further, A ( k )and b ( k ) can be determined by:

[0028] The first-order gradients of the unconstrained sparse motion planning model with respect to the joint angular velocity and the Lagrange multiplier are solved as follows:

[0029] ;

[0030] set up , and construct the following conditional equation:

[0031] ;

[0032] Solve the conditional equations to obtain the mass matrix of the Lagrangian condition A ( k ) and the coupling vector of the Lagrangian condition b ( k ).

[0033] Furthermore, in step S4, the sparse motion planning of the robot arm is implemented according to the optimal output results in each iteration, including:

[0034] The joint angular velocity sequence obtained based on the optimal output results of each fall is as follows:

[0035] ;

[0036] according to The velocity sequence of the end effector of the robot arm is as follows:

[0037] ;

[0038] The velocity sequence of the end effector of the robot arm is converted into the position sequence of the end effector as follows:

[0039] ;

[0040] Control the motion of the robot arm according to the joint angular velocity sequence, the velocity sequence of the end effector of the robot arm and the position sequence of the end effector to realize the sparse motion planning of the robot arm;

[0041] in, k m Indicates the maximum number of iterations.

[0042] According to another aspect of the present invention, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method for sparse motion planning of a robotic arm based on cross-species optimization provided by the present invention is implemented.

[0043] According to another aspect of the present invention, a computer-readable storage medium is provided, comprising a stored computer program. When the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned robot arm sparse motion planning method based on cross-species optimization provided by the present invention.

[0044] According to yet another aspect of the present invention, there is provided a sparse motion planning controller for a robotic arm based on cross-species optimization, comprising: an initialization module, a neural dynamics module, a cross-species optimization module, and a planning control module;

[0045] Initialization module, used to set the number of iterations k Initialize to 0; take the joint angle angular velocity space energy of the robot arm as the objective function, build the constrained sparse motion planning model of the robot arm, and transform it into an unconstrained sparse motion planning model based on the Lagrange multiplier method, and then trigger the neural dynamics module;

[0046] The neural dynamics module is used to build an error momentum neural network for solving the unconstrained sparse motion planning model in the current iteration, and then trigger the cross-species optimization module;

[0047] The cross-species optimization module is used to solve the output layer variables of the error momentum neural network model at different initial points, and use each output layer variable as a particle to search for the optimal output layer variable using the particle swarm algorithm, and then trigger the planning control module;

[0048] The planning control module is used to trigger the neural dynamics module to start the next iteration when the maximum number of iterations has not been reached, and to implement the sparse motion planning of the robot arm according to the optimal output layer variables in each iteration when the maximum number of iterations is reached;

[0049] The error momentum neural network consists of a hidden layer and an output layer, and the expression is as follows:

[0050] ;

[0051] in, , , All are positive numbers, indicating the network weight coefficient; is the output layer variable, represents the joint angular velocity, λ ( k ) represents the Lagrange multiplier, and the superscript T represents the transposition operation; y (k ) represents the hidden layer variables; A ( k ) represents the mass matrix of the Lagrangian condition, b ( k ) represents the coupling vector of the Lagrangian condition.

[0052] According to another aspect of the present invention, an industrial robot system is provided, comprising: a robot arm and the above-mentioned robot arm sparse motion planning controller based on cross-species optimization provided by the present invention.

[0053] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0054] This paper constructs an unconstrained robotic arm motion planning model containing sparse terms, which can reduce the number of joints required for the robotic arm to complete the trajectory tracking task and reduce energy consumption; on this basis, an error momentum neural network is designed, which fully considers the error information and momentum information in the iterative process and can accurately solve the unconstrained robotic arm motion planning model; further, a cross-species fusion framework of error momentum neural network and particle swarm algorithm is proposed, and the output layer variables of multiple error momentum neural networks are fused into the particle swarm algorithm as different ions, and multiple particles are used to collaboratively search for the global optimal solution, thereby integrating the respective biological advantages of the neural dynamics algorithm and the swarm particle algorithm, so that the solution process of the robotic arm sparse motion planning has both the fast optimization capability of the neural dynamics and the global optimization capability of the particle swarm algorithm, effectively improving the solution quality, thereby improving the efficiency, stability and accuracy of the robotic arm sparse motion planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flow chart of a method for sparse motion planning of a robotic arm based on cross-species optimization provided in an embodiment of the present invention.

[0056] Figure 2 The motion trajectory of the robot end effector generated by applying the existing neural network method.

[0057] Figure 3 A motion trajectory of a robotic arm end effector generated by a robotic arm sparse motion planning method based on cross-species optimization provided by an embodiment of the present invention.

[0058] Figure 4 Schematic diagram of the robotic arm motion planning process generated by applying existing neural network methods.

[0059] Figure 5 A schematic diagram of a robotic arm motion planning process generated by a robotic arm sparse motion planning method based on cross-species optimization provided by an embodiment of the present invention.

[0060] Figure 6 Schematic diagram of the error state between the three-dimensional spatial position of the robot end effector and the expected trajectory generated by applying the existing neural network method.

[0061] Figure 7 A schematic diagram of the error state between the three-dimensional spatial position of the end effector of a robotic arm and the expected trajectory generated by the robotic arm sparse motion planning method based on cross-species optimization provided in an embodiment of the present invention.

[0062] Figure 8 A concentrated schematic diagram of the angular velocities of each joint of the robotic arm generated by applying the existing neural network method.

[0063] Fig. 9 Individual schematic diagrams of the angular velocities of each joint of the robotic arm generated by applying the existing neural network method.

[0064] Fig.10 A concentrated schematic diagram of the angular velocities of each joint of a robotic arm generated by a sparse motion planning method for a robotic arm based on cross-species optimization provided by an embodiment of the present invention.

[0065] Fig.11 A separate schematic diagram of the angular velocities of each joint of a robotic arm generated by the robotic arm sparse motion planning method based on cross-species optimization provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0067] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the drawings are used to distinguish similar objects but not necessarily to describe a specific order or sequence.

[0068] In order to solve the technical problems that in the existing sparse motion planning methods for robotic arms, the neural dynamics algorithm can quickly converge to the local optimal solution of the optimization problem, but it is difficult to get rid of the local optimal trap, and the particle swarm algorithm has the global search capability, but the optimality of the solution is difficult to ensure and the time complexity is high, the present invention provides a sparse motion planning method and system for robotic arms based on cross-species optimization. The overall concept is that based on the solution requirements of the sparse motion planning of the robotic arm, an error-momentum neural network containing both error information and momentum information is designed to solve the motion planning model of the robotic arm. On this basis, a particle swarm algorithm is introduced, and a cross-species optimization fusion framework is proposed. The output results of multiple error-momentum neural networks are fused into the particle swarm algorithm as different particles, and multiple particles collaborate to search for the global optimal solution, thereby realizing the cross-species fusion of the error-momentum neural network and the particle swarm algorithm, so that the sparse motion planning process of the robotic arm has the fast optimization capability of the neural dynamics and the global optimization capability of the bird-like particle swarm algorithm, and effectively improves the solution quality of the sparse motion planning of the robotic arm.

[0069] The following is Example 1.

[0070] A sparse motion planning method for a robotic arm based on cross-species optimization, such as Figure 1 As shown, it includes: step S1 to step S4, and the specific implementation method of each step is explained below.

[0071] like Figure 1 In the embodiment, step S1 includes: k Initialized to 0; taking the joint angle angular velocity space energy of the robot as the objective function, a constrained sparse motion planning model of the robot is constructed, and then converted into an unconstrained sparse motion planning model based on the Lagrange multiplier method.

[0072] It is easy to understand that the constrained sparse motion planning model of the robot arm can be constructed based on the kinematic information and structural information of the robot arm. When performing the sparse motion planning task of the robot arm, the total motion planning time of the robot arm is given as T , the sampling interval is τ , the ideal trajectory of the robot motion planning is r d ( k ), the ideal speed of the robot motion planning is , the kinematic transfer function of the robot arm is The parameters to be solved include the position of the end effector of the robot arm r ( k ), the speed of the end effector of the robot , robot arm joint angle θ ( k ) and the angular velocity of the robot joint .

[0073] Based on the above information, the maximum number of iterations can be calculated as .

[0074] Further, a nonlinear transformation model of the robot's forward kinematics from the joint space to the Cartesian workspace can be established as follows:

[0075] ;

[0076] Since the robot arm kinematic transfer function Nonlinearity, can calculate the mechanical arm kinematic transfer function The Jacobian matrix of , and establish the robot arm motion planning model in the joint angular velocity space as shown below:

[0077] ;

[0078] Without loss of generality, the joint angular velocity space energy is used as the objective function , where the superscript T represents the vector transpose operation, Q Represents the weight matrix used to adjust the contribution of each joint angular velocity to the joint angular velocity space energy. In practical applications, the weight matrix Q It can be set based on experience; in order to enhance sparsity, this embodiment further adds the joint angular velocity to the robot arm motion planning model L p Norm (0< p <1), and considering the actual physical limitations, the angular velocity of each joint is limited to a certain bounded range , and In summary, the constrained robot arm sparse motion planning model established in this embodiment can be expressed as follows:

[0079] ;

[0080] The above constrained manipulator sparse motion planning model also includes the joint angular velocity space energy and sparse items , which can effectively reduce the number of joints required for the robot arm to complete trajectory tracking tasks and reduce energy consumption.

[0081] In order to reduce the difficulty of solving, this embodiment introduces the Lagrange multiplier λ ( k ), the constrained robot arm sparse motion planning model is transformed into an unconstrained robot arm sparse motion planning model as shown below:

[0082] ;

[0083] in, Represents a sparse motion planning model for an unconstrained robotic arm.

[0084] Using existing technology, we can obtain a sparse motion planning model for unconstrained robotic arms. About joint angular velocity and Lagrange auxiliary variables λ ( k ) are as follows:

[0085] ;

[0086] in, Represents the angular velocity of the joint L p Norm For joint angular velocity The derivative of can be further solved to obtain The expression is as follows:

[0087] ;

[0088] in, i Indicates the variable number, n represents the total number of variables; express No. i components; sign(·) represents the signum function, that is, the sign function, which is a logical function used to determine the positive and negative signs of real numbers.

[0089] like Figure 1 As shown, in this embodiment, step S2 includes: constructing an error momentum neural network for solving an unconstrained sparse motion planning model in the current iteration.

[0090] The error momentum neural network constructed in this embodiment is constructed according to the optimized KKT (Karush Kuhn Tucker Conditions). In order to improve the accuracy of the solution, the error momentum neural network established in this embodiment specifically includes a hidden layer and an output layer, and an error term and a momentum term are introduced into the model. Specifically, the expressions of the hidden layer variables and the output layer variables are as follows:

[0091] ;

[0092] in, , , All are positive numbers, indicating the network weight coefficient; is the output layer variable; y (k ) represents hidden layer variables; A ( k ) represents the mass matrix of the Lagrangian condition, b ( k ) represents the coupling vector of the Lagrangian condition.

[0093] In the hidden layer variables, A ( k ) x ( k )- b ( k ) represents the tracking error of the previous iteration, Represents the cumulative tracking error of previous iterations; in the output layer variables, y ( k +1)- y ( k ) is the difference of the hidden layer variables in two adjacent iterations. The hidden layer variables contain speed information, so this term constitutes the momentum term.

[0094] In this embodiment, the mass matrix of the Lagrangian condition is A ( k ) and the coupling vector of the Lagrangian condition b ( k ) can be determined by:

[0095] According to the output layer variables of the error momentum neural network and the sparse motion planning model of the unconstrained robot arm About joint angular velocity and Lagrange auxiliary variables λ ( k ) constructs the following conditional equation:

[0096] ;

[0097] Solve the conditional equations to obtain the mass matrix of the Lagrangian condition A ( k ) and the coupling vector of the Lagrangian condition b ( k ).

[0098] A ( k )and b ( k ) is expressed as: , .

[0099] The error-momentum neural network constructed in this embodiment can fully consider the impact of the tracking error in the previous iteration and the momentum information on the optimal solution in the current iteration, effectively improving the solution quality.

[0100] like Figure 1 As shown, in this embodiment, step S3 includes: solving the output layer variables of the error momentum neural network model at different initial points, and using each output layer variable as a particle to search for the optimal output layer variable using a particle swarm algorithm.

[0101] Specifically, in the particle swarm algorithm, multiple particles collaborate to search for the global optimal solution as follows:

[0102] .

[0103] in, x i ( k ) indicates the k In the iteration i Error momentum neural network output layer variables, v i ( k ) indicates the i The speed of particle search, p i ( k ) represents the local optimal particle, g i ( k ) represents the globally optimal particle, c and d Represents the random weight of the particle swarm.

[0104] like Figure 1 As shown, in this embodiment, step S4 includes: if the maximum number of iterations is not reached, proceed to step S2 to perform the next iteration; otherwise, the sparse motion planning of the robot arm is implemented according to the optimal output layer variables in each iteration.

[0105] Specifically, when the maximum number of iterations is reached, the joint angular velocity sequence can be extracted from the optimal output layer variables in each iteration as follows:

[0106] ;

[0107] According to the joint angular velocity sequence, the joint angle sequence can be converted as follows:

[0108] ;

[0109] according to The velocity sequence of the end effector of the robot arm is as follows:

[0110] ;

[0111] The velocity sequence of the end effector of the robot arm is converted into the position sequence of the end effector as follows:

[0112] ;

[0113] The motion of the robot arm is controlled according to the joint angular velocity sequence, the joint angle sequence, the velocity sequence of the end effector of the robot arm and the position sequence of the end effector to realize the sparse motion planning of the robot arm.

[0114] The following is Example 2.

[0115] A computer program product includes a computer program. When the computer program is executed by a processor, the method for sparse motion planning of a robotic arm based on cross-species optimization provided in the above-mentioned embodiment 1 is implemented.

[0116] The following is Example 3.

[0117] A computer-readable storage medium includes a stored computer program. When the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to execute the robot arm sparse motion planning method based on cross-species optimization provided in the above-mentioned embodiment 1.

[0118] The following is Example 4.

[0119] According to yet another aspect of the present invention, there is provided a sparse motion planning controller for a robotic arm based on cross-species optimization, comprising: an initialization module, a neural dynamics module, a cross-species optimization module, and a planning control module;

[0120] Initialization module, used to set the number of iterations k Initialize to 0; take the joint angle angular velocity space energy of the robot arm as the objective function, build the constrained sparse motion planning model of the robot arm, and transform it into an unconstrained sparse motion planning model based on the Lagrange multiplier method, and then trigger the neural dynamics module;

[0121] The neural dynamics module is used to build an error momentum neural network for solving the unconstrained sparse motion planning model in the current iteration, and then trigger the cross-species optimization module;

[0122] The cross-species optimization module is used to solve the output layer variables of the error momentum neural network model at different initial points, and use each output layer variable as a particle to search for the optimal output layer variable using the particle swarm algorithm, and then trigger the planning control module;

[0123] The planning control module is used to trigger the neural dynamics module to start the next iteration when the maximum number of iterations has not been reached, and to implement the sparse motion planning of the robot arm according to the optimal output layer variables in each iteration when the maximum number of iterations is reached;

[0124] The error momentum neural network consists of a hidden layer and an output layer, and the expression is as follows:

[0125] ;

[0126] in, , , All are positive numbers, indicating the network weight coefficient; is the output layer variable, represents the joint angular velocity, λ ( k ) represents the Lagrange multiplier, and the superscript T represents the transposition operation; y ( k ) represents the hidden layer variables; A ( k ) represents the mass matrix of the Lagrangian condition, b ( k ) represents the coupling vector of the Lagrangian condition.

[0127] In this embodiment, the specific implementation of each module can refer to the description in the above embodiment 1 and will not be repeated here.

[0128] The following is Example 5.

[0129] An industrial robot system includes: a robotic arm and the robotic arm sparse motion planning controller based on cross-species optimization provided in the above-mentioned embodiment 4.

[0130] The following is a further analysis and explanation of the beneficial effects that can be achieved by the present invention in conjunction with a specific example of executing a sparse motion planning task for a robot arm. The relevant parameters of the sparse motion planning task for a robot arm include: τ =0.001s, initial time t= 0, end time , the initial number of iterations k = 0, maximum number of iterations k m =10000; the error momentum neural network weight coefficient is , , ; Particle swarm algorithm random weight and ; Get the initial position of the redundant robot arm end effector r (0), initial speed , the initial joint angles θ (0), the angular velocity of each joint at the initial moment , the ideal trajectory position of the robot arm end effector at the initial moment r d (0), the ideal trajectory speed at the initial moment , the Jacobian matrix of the robot at the initial momentJ ( θ (0)), weight matrix Q is the unit matrix, the minimum angular velocity and maximum angular velocity of the joint are and .

[0131] The specific expressions of the above parameters are as follows:

[0132] ;

[0133] The ideal trajectory of the robot arm's end effector is a "bullfighting dog" shape.

[0134] Taking the existing neural network method as a comparison, the above-mentioned robotic arm motion planning method is executed using the existing neural network method and the robotic arm sparse motion planning method based on cross-species optimization provided by the present invention respectively.

[0135] Figure 2 The figure shows the motion trajectory of the end effector of the robot arm generated by the existing neural network method. The result shows that when the sparse norm is considered, the end effector of the robot arm driven by the existing neural network method cannot perform the motion generation of the specified ideal trajectory "dogfighting".

[0136] Figure 3 The figure shows the motion trajectory of the end effector of the robot arm generated by the sparse motion planning method of the robot arm based on cross-species optimization provided by the present invention. The result shows that, under the consideration of the sparse norm, the present invention can enable the end effector of the robot arm to successfully execute the motion generation of the specified ideal trajectory "dogfighting", which reflects the feasibility of the present invention.

[0137] Figure 4 Shown is a schematic diagram of the robotic arm motion planning process generated using the existing neural network method. The results show that the robotic arm driven by the existing neural network method cannot repeatedly generate the specified ideal trajectory "dogfighting".

[0138] Figure 5 Shown is a schematic diagram of the robot arm motion planning process generated by the robot arm sparse motion planning method based on cross-species optimization provided by the present invention. The result shows that the present invention can repeatedly generate the specified ideal trajectory "dog fighting", which reflects the superiority and stability of the present invention.

[0139] Figure 6 The figure shows the error state between the three-dimensional spatial position of the end effector of the robot arm and the expected trajectory generated by the existing neural network method. Figure 6 It can be seen that using the existing neural network method, the error in the X-axis and Y-axis directions is approximately plus or minus 0.05 meters.

[0140] Figure 7 The figure shows the error state between the three-dimensional spatial position of the end effector of the manipulator and the expected trajectory generated by the sparse motion planning method of the manipulator based on cross-species optimization provided by the present invention. Figure 7 It can be seen that the error of the present invention in the X-axis direction and the Y-axis direction is approximately plus or minus 0.0008 meters.

[0141] contrast Figure 6 and Figure 7 It can be seen that the accuracy of the present invention is significantly improved compared with the existing neural network method.

[0142] Figure 8 and Fig. 9 The joint speeds of the robotic arm generated by the existing neural network method are normal, within the range of plus or minus 0.3 radians per second. The speed of each joint varies greatly, and there is no joint with zero speed. Only the seventh joint is slowly approaching zero, indicating that the existing neural network method has problems such as poor sparsity and high energy consumption when solving the sparse motion planning of the robotic arm.

[0143] Fig.10 and Fig.11 The joint speeds of the robotic arm generated by the robotic arm sparse motion planning method based on cross-species optimization provided by the present invention are normal, within the range of plus or minus 0.3 radians per second, of which two joint speeds (the fifth joint and the seventh joint) are close to zero during the running time, indicating that the present invention has the advantages of good sparsity and low energy consumption when solving the sparse motion planning of the robotic arm.

[0144] In general, the present invention provides a method and system for sparse motion planning of a robotic arm based on cross-species optimization. On the basis of constructing an unconstrained robotic arm motion planning model containing sparse terms, an error momentum neural network is designed. The error momentum neural network fully considers the error information and momentum information in the iterative process, and can accurately solve the unconstrained robotic arm motion planning model. A cross-species fusion framework of the error momentum neural network and the particle swarm algorithm is further proposed. The output layer variables of multiple error momentum neural networks are fused into the particle swarm algorithm as different ions, and multiple particles are used to collaboratively search for the global optimal solution. In this way, the respective biological advantages of the neural dynamics algorithm and the swarm particle algorithm can be integrated, so that the solution process of the sparse motion planning of the robotic arm has both the fast optimization capability of the neural dynamics and the global optimization capability of the particle swarm algorithm, which effectively improves the solution quality, thereby improving the efficiency, stability and accuracy of the sparse motion planning of the robotic arm.

[0145] It will be easily understood by those skilled in the art that the above description is only a preferred 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 in the protection scope of the present invention.

Claims

1. A sparse motion planning method for a robotic arm based on cross-species optimization, characterized in that: include: Step S1: Initialize the number of iterations k to 0; Taking the joint angle angular velocity space energy of the robot arm as the objective function, a constrained sparse motion planning model of the robot arm is constructed, and then transformed into an unconstrained sparse motion planning model based on the Lagrange multiplier method; Step S2: construct an error momentum neural network for solving the unconstrained sparse motion planning model in the current iteration, which includes a hidden layer and an output layer, and the expression is as follows: Among them, α, β, and ω are all positive numbers, representing the network weight coefficients; is the output layer variable, represents the joint angular velocity, λ(k) represents the Lagrangian multiplier, and the superscript T represents the transpose operation; y(k) represents the hidden layer variable; A(k) represents the mass matrix of the Lagrangian condition, and b(k) represents the coupling vector of the Lagrangian condition; Step S3: solving the output layer variables of the error momentum neural network at different initial points, and using each output layer variable as a particle to search for the optimal output layer variable using a particle swarm algorithm; Step S4: If the maximum number of iterations has not been reached, proceed to step S2 for the next iteration; otherwise, implement the sparse motion planning of the robot arm according to the optimal output layer variables in each iteration.

2. The sparse motion planning method for a robotic arm based on cross-species optimization according to claim 1, characterized in that: The expression of the constrained sparse motion planning model is as follows: Wherein, Q represents the weight matrix used to adjust the contribution of each joint angular velocity to the joint angular velocity space energy; L represents the angular velocity of the joint p Norm, 0 <p<1; represents the ideal velocity of the end effector of the robot arm; J(k) represents the Jacobian matrix of the kinematic transfer function of the robot arm; and are the minimum angular velocity and the maximum angular velocity respectively.

3. The sparse motion planning method for a robotic arm based on cross-species optimization according to claim 2, characterized in that: The expression of the unconstrained sparse motion planning model is as follows: in, represents the unconstrained sparse motion planning model.

4. The sparse motion planning method for a robotic arm based on cross-species optimization according to claim 3, characterized in that: in, L represents the angular velocity of the joint p Norm For joint angular velocity The derivative of .

5. The sparse motion planning method for a robotic arm based on cross-species optimization according to claim 4, characterized in that: A(k) and b(k) are determined by: The first-order gradients of the unconstrained sparse motion planning model with respect to the joint angular velocity and the Lagrange multiplier are solved as follows: set up And construct the following conditional equation: The conditional equation is solved to obtain the mass matrix A(k) of the Lagrangian condition and the coupling vector b(k) of the Lagrangian condition.

6. The method for sparse motion planning of a robotic arm based on cross-species optimization according to any one of claims 1 to 5, characterized in that: In step S4, the sparse motion planning of the robot arm is implemented according to the optimal output result in each iteration, including: According to the optimal output results of each iteration, the joint angular velocity sequence is as follows: according to The velocity sequence of the end effector of the robot arm is as follows: The velocity sequence of the end effector of the robot arm is converted into the position sequence of the end effector as follows: {r(0),r(1),…,r(k m -1)} Control the motion of the robot arm according to the joint angular velocity sequence, the velocity sequence of the end effector of the robot arm and the position sequence of the end effector to realize the sparse motion planning of the robot arm; Among them, k m Indicates the maximum number of iterations.

7. A computer program product, characterized in that It comprises a computer program, which, when executed by a processor, implements the sparse motion planning method for a robotic arm based on cross-species optimization as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: It includes a stored computer program, which, when executed by a processor, controls the device where the computer-readable storage medium is located to execute the sparse motion planning method for a robotic arm based on cross-species optimization as described in any one of claims 1 to 6.

9. A sparse motion planning controller for a robotic arm based on cross-species optimization, characterized in that: include: Initialization module, neural dynamics module, cross-species optimization module and planning control module; The initialization module is used to initialize the number of iterations k to 0; take the joint angle angular velocity space energy of the robot arm as the objective function, construct a constrained sparse motion planning model of the robot arm, and convert it into an unconstrained sparse motion planning model based on the Lagrange multiplier method, and then trigger the neural dynamics module; The neural dynamics module is used to construct an error momentum neural network for solving the unconstrained sparse motion planning model in a current iteration, and then trigger the cross-species optimization module; The cross-species optimization module is used to solve the output layer variables of the error momentum neural network at different initial points, and use each output layer variable as a particle to search for the optimal output layer variable using a particle swarm algorithm, and then trigger the planning control module; The planning control module is used to trigger the neural dynamics module to start the next iteration when the maximum number of iterations has not been reached, and to implement the sparse motion planning of the robot arm according to the optimal output layer variables in each iteration when the maximum number of iterations is reached; The error momentum neural network includes a hidden layer and an output layer, and the expression is as follows: Among them, α, β, and ω are all positive numbers, representing the network weight coefficients; is the output layer variable, represents the joint angular velocity, λ(k) represents the Lagrangian multiplier, and the superscript T represents the transpose operation; y(k) represents the hidden layer variable; A(k) represents the mass matrix of the Lagrangian condition, and b(k) represents the coupling vector of the Lagrangian condition.

10. An industrial robot system, characterized in that: include: A robotic arm and a robotic arm sparse motion planning controller based on cross-species optimization as described in claim 9.

Images