A method for state estimation of an uncertain robot system

Abstract

This invention belongs to the field of robot motion control technology and discloses a method for estimating the state of an uncertain robot system. The method includes: constructing a corresponding state-space model based on parameters including the state variables of the uncertain robot system; wherein the state variables include at least one known state variable and a state variable to be estimated; approximating the nonlinear state function and nonlinear measurement function in the state-space model using a multidimensional Taylor network; employing a latent variable processing mechanism to merge and summarize robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network, constructing a linear polynomial equivalent system; and estimating the state of the linear polynomial equivalent system in real time to obtain the state variable to be estimated. This invention does not rely on specific sensors, does not require a precise mechanistic model, and has advantages such as high computational accuracy and low hardware cost, making it applicable to various uncertain robot systems.

Description

Technical Field

[0001] This invention belongs to the field of robot motion control technology, and more specifically, relates to a method for estimating the state of an uncertain robot system. Background Technology

[0002] The rapid development of robotics technology has made it an indispensable tool in automated and intelligent operations. Especially in dangerous or unsuitable environments for human work, such as the deep sea, outer space, high temperatures, and high radiation, robots can replace humans in performing tasks, reducing work risks and improving work efficiency.

[0003] To achieve the requirements of high adaptability and flexibility, a key technical challenge in robot motion control is the accurate acquisition of joint pose information for the robotic arm. However, due to limitations such as compact structure, limited joint space, and harsh working conditions, it is difficult to deploy multiple sensors to measure all pose information. Furthermore, data obtained from a single sensor may be subject to noise interference, resulting in poor data reliability. Therefore, existing technologies generally employ system state estimation techniques to acquire complete motion state information of the robot system with fewer sensor inputs.

[0004] State observers, as a simple and reliable system state estimation technique, are widely used in various control systems. However, existing state observer-based system state estimation techniques heavily rely on the mechanistic model of the observed system. Robotic systems, however, exhibit significant uncertainties and nonlinearities, especially under extreme conditions such as on the lunar surface, in space, and in nuclear facilities. Their system parameters exhibit time-varying uncertainties and are subject to unknown disturbances, making it difficult to construct accurate mechanistic models and resulting in low accuracy in existing robot system state estimations. How to effectively estimate the motion state information of uncertain robotic systems remains a challenge. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method for state estimation of uncertain robot systems, the purpose of which is to improve the accuracy of state estimation of uncertain robot systems.

[0006] To achieve the above objectives, according to a first aspect of the present invention, a method for estimating the state of an uncertain robot system is provided, comprising:

[0007] S1. Construct a state-space model of the uncertain robot system based on parameters including state variables of the uncertain robot system; wherein, the state variables include at least one known state variable and a state variable to be estimated;

[0008] S2. Use a multidimensional Taylor network to approximate the nonlinear state function and nonlinear measurement function in the state-space model;

[0009] S3. Using a latent variable processing mechanism, robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network are merged and summarized. Based on the merged and summarized nonlinear state function and nonlinear measurement function, the state space model is converted into a linear polynomial equivalent system.

[0010] S4. Estimate the state of the linear polynomial equivalent system in real time to obtain the state variables to be estimated.

[0011] Furthermore, in S2, the nonlinear state function f after approximation by the multidimensional Taylor network... i (q(k)) and nonlinear measurement function h i (q(k+1)) are respectively:

[0012]

[0013]

[0014] Where q(k) and q(k+1) are the N-dimensional state vectors of the uncertain robot system at time k and time k+1, respectively, and N is the number of state variables; n i and m i Let be the order of any state variable, and be a non-negative positive integer; and Let Δf be the weight corresponding to each order tensor in the multidimensional Taylor network model. i (k) and Δh i (k+1) represents the approximation error, and r represents the dimension of the set Taylor net.

[0015] Furthermore, in S3, a latent variable processing mechanism is employed to merge and summarize robot state terms with similar characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network, including:

[0016] Define hidden variables and the latent variable x (n) The weight of (k) is Where, x (1) (k)=q(k);

[0017] Based on the latent variable x (n) (k) and the corresponding weights are used to merge and summarize the robot state terms with the same characteristics in the nonlinear state function after the multidimensional Taylor network approximation.

[0018] Define hidden variables and the latent variable x (m) The weight of (k+1) is Where, x(1) (k+1)=q(k+1);

[0019] Based on the latent variable x (m) (k+1) and the corresponding weights are used to merge and summarize the robot state terms with the same characteristics in the nonlinear measurement function approximated by the multidimensional Taylor network.

[0020] Furthermore, the linear polynomial equivalent system is:

[0021]

[0022] Where X(k+1) and X(k) are the state matrices of the linear polynomial equivalent system at time k+1 and time k, respectively, and they have the same matrix structure X(·)=[(x (1) (·)) T ,(x (2) (·)) T ,....,(x (r) (·)) T ,(Δf(·)) T ] T ψ and Γ are the state transition matrices of the equivalent system; Y(k+1) is the output vector of the linear polynomial equivalent system; W(k), V(k+1), and U(k) are vectors composed of the first-order to r-order terms of w(k), v(k+1), and u(k), respectively, w(k) is the state noise at time k, and v(k+1) is the measurement noise at time k+1; u(k) = [u1, ..., u N ] T The control input vector is known.

[0023] Further, in S4, an equivalent extended Kalman observer is used to estimate the state of the linear polynomial equivalent system in real time, obtaining the state variables to be estimated, including:

[0024] S41. Solve for the one-step prediction estimate of the state X(k+1) of the linear polynomial equivalent system. in,

[0025] S42. Using a maximum correlation entropy Kalman filter strategy, solve for the optimal state estimate of the linear polynomial equivalent system at time k+1, thereby obtaining the state variables to be estimated; wherein, the optimal state estimate of the linear polynomial equivalent system at time k+1 is... for:

[0026]

[0027] In the formula, K(k+1) is the Kalman filter gain.

[0028] Furthermore, in S1, the state-space model of the uncertain robot system is:

[0029]

[0030] Where q(k) and q(k+1) are the N-dimensional state vectors of the uncertain robot system at time k and time k+1, respectively, and N is the number of state variables; y(k+1) is the M-dimensional measurement vector of the uncertain robot system at time k+1, and M is the number of known state variables; F(q(k)) = [f1(q(k)), ..., f i (q(k)), ..., f N (q(k))] T Let f be a nonlinear state function vector. i (q(k)) represents the i-th nonlinear state function; H(q(k+1))=[h1(k+1),...,h i (k+1),...,h M (k+1)] T h is a vector of nonlinear measurement functions. i (k+1) represents the i-th nonlinear measurement function; u(k) = [u1, ..., u N ] T Given a known control input vector, w(k) is the state noise at time k, and v(k+1) is the measurement noise at time k+1. w(k) and v(k+1) are uncorrelated.

[0031] According to a second aspect of the present invention, an uncertain robot system state estimation device is provided, comprising: at least one processor, a data storage device connected to the processor, and a sensor data reading device;

[0032] The sensor data reading device is used to read the known state quantities of the uncertain robot system and transmit them to the data storage.

[0033] The processor executes the uncertainty robot system state estimation method described in any of the first aspects by reading the program in the data memory.

[0034] According to a third aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the uncertainty robot system state estimation method as described in any of the first aspects.

[0035] According to a fourth aspect of the invention, a computer program product is provided that, when the computer program product is run on a computer, causes the computer to execute the uncertainty robot system state estimation method described in any one of the first aspects.

[0036] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0037] (1) The uncertainty robot system state estimation method of the present invention does not rely on specific sensors and does not require the establishment of an accurate mechanism model. It only requires the sensor to collect at least one known state variable. By combining the known state data with the multidimensional Taylor network model of the nonlinear state function and the nonlinear measurement function, and using the latent variable processing mechanism, a virtual linear polynomial equivalent system of the measured uncertainty robot system is constructed. The introduction of the multidimensional Taylor network to approximate the nonlinear state function and the nonlinear measurement function also makes the constructed virtual linear polynomial equivalent system more than the original state space model with higher-order terms. Thus, the constructed virtual linear polynomial equivalent system can reflect the complex operation process of the actual uncertainty robot system with high fidelity. By extracting the state variables of the virtual linear polynomial equivalent system, the state variables to be estimated with high accuracy can be obtained. It has the advantages of high computational accuracy and low hardware cost. It can be applied to various uncertainty robot systems and solves the problem of low accuracy of motion state estimation when the mechanism model of the robot system is uncertain.

[0038] (2) Furthermore, for the expansion form of the multidimensional Taylor network model of nonlinear state function and nonlinear measurement function, a specific hidden variable and corresponding weight are designed to merge and summarize the robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function after the multidimensional Taylor network approximation. This specific merging object and merging method facilitates the linearization of nonlinear state terms and simplifies the process of constructing a linear polynomial equivalent system. Attached Figure Description

[0039] Figure 1 This is a flowchart of the uncertainty robot system state estimation method in an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of the rope-driven robotic arm structure used in an embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0042] Example 1

[0043] like Figure 1 The uncertainty robot system state estimation method described in this embodiment of the invention mainly includes:

[0044] S1. Construct a state-space model of the uncertain robot system based on parameters including the state variables of the uncertain robot system; wherein, the state variables of the uncertain robot system include at least one known state variable and a state variable to be estimated;

[0045] S2. Use a multidimensional Taylor network to approximate the nonlinear state function and nonlinear measurement function in the state-space model;

[0046] S3. Using a latent variable processing mechanism, robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network are merged and summarized. Based on the merged and summarized nonlinear state function and nonlinear measurement function, the state space model of the uncertain robot system is transformed into a linear polynomial equivalent system.

[0047] S4. Estimate the state of the equivalent linear polynomial system in real time to obtain the state variables to be estimated for the uncertain robot system.

[0048] Specifically, the state-space model of the uncertain robot system constructed in this embodiment of the invention is as follows:

[0049]

[0050] Where q(k+1) is the N-dimensional state vector of the uncertain robot system at time k+1, where N is the number of state variables of the uncertain robot system; y(k+1) is the M-dimensional measurement vector of the uncertain robot system at time k+1, where M is the number of known state variables; F(q(k)) = [f1(q(k)), ..., f i (q(k)), ..., f N (q(k))] T Let f be an unknown nonlinear state function vector. i (q(k)) represents the i-th nonlinear state function; H(q(k+1))=[h1(k+1),...,h i(k+1),...,h M (k+1)] T Let u(k) be an unknown nonlinear measurement function vector, where u(k) = [u1, ..., u2]. N ] T Given a known control input vector, w(k) and v(k) are uncorrelated noises in the system, where w(k) is the state noise at time k and v(k+1) is the measurement noise at time k+1.

[0051] Preferably, the nonlinear state function f in S2 i (q(k)) and nonlinear measurement function h i The multidimensional Taylor network model of (q(k+1)) is:

[0052]

[0053]

[0054] in, The order n of any state variable i and m i All are non-negative positive integers. and In this embodiment of the invention, the weights corresponding to each order tensor in a multidimensional Taylor network model are calculated using the differential evolution algorithm commonly used in neural networks, Δf. i (k) and Δh i (k+1) represents the approximation error, and r represents the dimension of the set Taylor net.

[0055] Preferably, in embodiment S3 of the present invention, for the nonlinear state function, a hidden variable is defined. For nonlinear measurement functions, latent variables are defined. Simultaneously define the hidden variable x (n) The weight of (k) is Define a hidden variable x (m) The weight of (k+1) is Where, x (1) (k)=q(k) (the N-dimensional state vector of the robot system at time k), x (1) (k+1) = q(k+1).

[0056] Based on the hidden variables and weights designed above, the robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network are merged and summarized. Based on the merged and summarized nonlinear state function and nonlinear measurement function, the transformed linear polynomial equivalent system is as follows:

[0057]

[0058] Where X(k+1) and X(k) are the state matrices of the equivalent system at time k+1 and time k, respectively, and they have the same matrix structure X(·)=[(x (1) (·)) T ,(x (2) (·)) T ,....,(x (r) (·)) T ,(Δf(·)) T ] T ψ and Γ are the state transition matrices of the equivalent system, ψ = [ψ1, ψ2, ..., ψ] r ,ψ r+1 ] T , Γ=[Γ1,Γ2,......,Γ r ,Γ r+1 ] T ;Y(k+1)=[(y (1) (k+1)) T ,(y (2) (k+1)) T ,....,(y (r) (k+1)) T ] T Let W(k), V(k+1), and U(k) be the output vectors of the equivalent system; W(k), V(k+1), and U(k) are vectors composed of the first-order to r-order terms of w(k), v(k+1), and u(k), respectively, and are expressed as follows:

[0059] W(k)=[w (1) (k)…w (n) (k)…w (r) (k)] T ;

[0060] V(k+1)=[v (1) (k+1)…v (n) (k+1)…v (r) (k+1)] T ;

[0061] U(k)=[u (1) (k)…u (n) (k)u (r) (k)] T .

[0062] Preferably, in S4, an equivalent extended Kalman observer is used to estimate the state of the linear polynomial equivalent system in real time, including:

[0063] S41. Solve for the one-step prediction estimate of the state X(k+1) of the equivalent linear polynomial system. The formula is

[0064] S42. Using the maximum correlation entropy Kalman filter strategy, the optimal state estimate of the linear polynomial equivalent system is solved, and then the optimal state estimate of the uncertain robot system is obtained.

[0065] Wherein, the optimal state estimate at time k+1 as follows:

[0066]

[0067] In the formula, K(k+1) is the Kalman filter gain.

[0068] The uncertainty robot system state estimation method of the present invention uses a multidimensional Taylor network to approximate the nonlinear state function and nonlinear measurement function in the state space model, and combines a latent variable processing mechanism to merge and summarize robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network, thereby constructing a linear polynomial equivalent system of the uncertainty robot system. The equivalent virtual system is used to simulate the complex operation process of the actual uncertainty robot, and the state of the equivalent virtual linear polynomial system can be estimated in real time to obtain the state variables of the uncertainty robot system to be estimated.

[0069] The method of this invention does not rely on specific sensors and does not require the establishment of an accurate mechanistic model. It only requires sensors to collect at least one known state variable. Through the known state data, combined with a multidimensional Taylor network model of nonlinear state function and nonlinear measurement function, and employing a latent variable processing mechanism, a virtual linear polynomial equivalent system of the measured uncertain robot system is constructed. Furthermore, the introduction of a multidimensional Taylor network to approximate the nonlinear state function and nonlinear measurement function increases the information of higher-order terms in the constructed virtual linear polynomial equivalent system compared to the original state space model. This enables the constructed virtual linear polynomial equivalent system to reflect the complex operation process of the actual uncertain robot system with high fidelity. Extracting the state variables of the virtual linear polynomial equivalent system yields the estimated state variables with high accuracy. It has advantages such as high computational accuracy and low hardware cost, and can be applied to various uncertain robot systems, solving the problem of low accuracy in motion state estimation when the mechanistic model of the robot system is uncertain.

[0070] Furthermore, for the expansion form of the multidimensional Taylor network model of nonlinear state function and nonlinear measurement function, a specific hidden variable and corresponding weight are designed to merge and summarize robot state terms with the same characteristics in the nonlinear state function and nonlinear measurement function after multidimensional Taylor network approximation. This specific merging object and merging method facilitates the linearization of nonlinear state terms and simplifies the process of constructing a linear polynomial equivalent system.

[0071] In summary, the state estimation method of this invention does not rely on the precise mechanistic model of the observed system, nor on specific sensors. It obtains all the state information of the uncertain robot system simply by constructing a linear polynomial equivalent system based on a multidimensional Taylor network and performing Kalman filtering on the measurable data according to the maximum correlation entropy principle. This method can be adapted to different types of uncertain robot systems, increasing the practicality of the estimation algorithm.

[0072] The state estimation method of the present invention will be further described below with specific embodiments.

[0073] The state estimation method described in the above embodiments is used to... Figure 2 The state estimation of the cable-driven robotic arm system shown is performed.

[0074] S1. In this embodiment of the invention, the actuator angle q of the robot system m and the connecting rod angle q l The link speed can be directly measured by an encoder and is a known state quantity. Linkage acceleration Actuator speed Actuator acceleration Information such as these cannot be directly obtained through sensors and are state variables to be estimated.

[0075] For the robot system in this embodiment, its robot dynamics and actuator dynamics are given by the following Lagrange formulas:

[0076]

[0077] Where, q l , q represents the link angle, link velocity, and link acceleration, respectively. m , Let M(q) represent the angle, velocity, and acceleration of the actuator, respectively. G(q) and G(q) are the inertia matrix, the Coriolis force matrix, and the gravity vector, respectively. For unknown disturbance terms, K is the elastic modulus, J and B are the actuator inertia and damping matrix, respectively, and u is the input torque of the motor.

[0078] In this embodiment, the M(q) of the above-mentioned robot system G(q) The terms K, J, and B are all unknown and exhibit significant nonlinearity and uncertainty.

[0079] Therefore, the state-space model of the uncertain robot system is constructed as follows:

[0080]

[0081] Where q(k+1) = [q1(k+1), ..., q6(k+1)] T Let q1, q2, q3, q4, q5, and q6 be the state vector at time k+1, and let q1, q2, q3, q4, q5, and q6 be the link angles q1, q2, q3, q4, q5, and q6, respectively. l Linkage speed Linkage acceleration Actuator angle q m Actuator speed Actuator acceleration y(k+1) = [y1(k+1), y2(k+1)] T Let F(q(k)) be the measurement vector at time k+1, and let F(q(k)) = [f1(q(k)), ..., f6(q(k))]. T And H(q(k+1))=[h1(k+1),h2(k+1)] T Let u(k) = [u1, ..., u6] be the unknown nonlinear state function and the nonlinear measurement function, respectively. T Given the control input vector, w(k) = [w1, ..., w6] T And v(k) = [v1, v2] T These are uncorrelated noises in the system.

[0082] S2. Use a multidimensional Taylor network to process each nonlinear state function f i (q(k))(i=1,2,...,6) and nonlinear measurement function h i (q(k+1))(i=1,2), are modeled. In this embodiment, the nonlinear state function f i (q(k)) and nonlinear measurement function h i The specific definition of the multidimensional Taylor network model of (q(k+1)) is as follows:

[0083]

[0084]

[0085] in, The order n of any state variable i and m i All are non-negative positive integers. and Let Δf be the weight corresponding to each order tensor in the multidimensional Taylor network model. Since the network structure of the multidimensional Taylor network is similar to that of a neural network, the differential evolution algorithm commonly used in neural networks can be used to calculate the weights. i (k) and Δh i (k+1) represents the approximation error.

[0086] S3. Combining the hidden variable processing mechanism, construct a linear polynomial equivalent system for the uncertain robot system.

[0087] Define the hidden variable as Simultaneously define the hidden variable x (n) The weight of (k) is Where, x (1) (k+1)=[q1(k+1),...,q6(k+1)] T .

[0088] Based on the latent variable representation, the state equation of the uncertain robot system is reconstructed, that is, a linear polynomial equivalent state equation is established, as shown in the following formula:

[0089]

[0090] The above state equations can be rearranged into the following pseudo-linearized form:

[0091] x (1) (k+1)=ψ1x(k)+Δf (1) (k)+u (1) (k)+w (1) (k);

[0092] in, x(k)=[x (1) (k),x (2) (k),......,x (r) (k)] T , Δf (1) (k)=Δf(k)=[Δf1(k),Δf2(k),...,Δf6(k)] T u (1) (k) and w (1) (k) represents u(k) and w(k) raised to the first power, respectively.

[0093] Similarly, define the latent variable as Simultaneously define the hidden variable x (m) The weight of (k+1) is

[0094] Among them, y (1) (k+1)=[y1(k+1),...,y6(k+1)] T .

[0095] Using latent variables, a model of the measurement function of an uncertain robot system is constructed, and its equivalent linear polynomial model is expressed as follows:

[0096] y (1)(k+1)=Γ1x(k+1)+Δh (1) (k+1)+v (1) (k+1);

[0097] in, x(k+1)=[x (1) (k+1),x (2) (k+1),......,x (r) (k+1)] T ,Δh (1) (k+1)=Δh(k+1)=[Δh1(k+1),Δh2(k+1),...,Δh6(k+1)] T v (1) (k+1) represent v(k+1) raised to the power of one.

[0098] Based on the dynamic relationships between latent variables of different orders, the linear polynomial equivalent system of the uncertain robot system is constructed as follows:

[0099]

[0100] Among them, the state matrix X(k+1) at time k+1 and the state matrix X(k) at time k have the same matrix structure X()=[(x (1) (·)) T ,(x (2) (·)) T ,....,(x (r) (·)) T ,(Δf(·)) T ] T ;

[0101] ψ=[ψ1,ψ2,......,ψ r ,ψ r+1 ] T Y(k+1)=[(y (1) (k)) T ,(y (2) (k)) T ,....,(y (r) (k)) T ] T ;

[0102] Γ=[Γ1,Γ2,......,Γ r ,Γ r+1 ] T ;

[0103] W(k)=[w (1) (k)…w (n) (k)…w (r) (k)]T U(k)=[u (1) (k)…u (n) (k)u (r) (k)] T ;

[0104] V(k)=[v (1) (k)…v (n) (k)v (r) (k)] T .

[0105] S4. Design an equivalent extended Kalman observer to estimate the state of the robot system in real time.

[0106] In this embodiment, the state variables of the linear polynomial type equivalent system are... first-order term This is the estimated value of the state of the uncertain robot system, namely:

[0107]

[0108] In the formula, The connecting rod angles q l Linkage speed Linkage acceleration Actuator angle q m Actuator speed Actuator acceleration The optimal estimate at time k+1.

[0109] As can be seen, this invention addresses the difficulty in measuring the joint pose information of a tethered flexible robotic arm by proposing a practical and reliable measurement method. It extracts information from only a limited number of sensors, constructs a linear polynomial equivalent system based on a multidimensional Taylor network, and applies Kalman filtering to the extracted measurement data according to the maximum correlation entropy principle to obtain the optimal estimate based on maximum correlation entropy, containing the required joint pose information. This invention does not rely on specific sensors and requires no precise mechanistic model. It accurately estimates the motion states of a robot system, such as joint velocity, joint acceleration, actuator velocity, and actuator acceleration, using only limited sensor information. It offers advantages such as high computational accuracy and low hardware cost, and can be applied to various uncertain robot systems.

[0110] Example 2

[0111] This invention provides an uncertain robot system state estimation device, comprising: at least one processor, a data storage device connected to the processor, and a sensor data reading device.

[0112] Among them, the sensor data reading device is used to read the known state quantities of the uncertain robot system and transmit them to the data storage;

[0113] The processor executes the robot system state estimation method in Example 1 by reading the program from the data memory. For a detailed implementation process, please refer to the specific scheme in Example 1, which will not be repeated here.

[0114] Example 3

[0115] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the uncertainty robot system state estimation method as described in Embodiment 1. For a detailed implementation process, please refer to the specific scheme in Embodiment 1, which will not be repeated here.

[0116] Example 4

[0117] This invention provides a computer program product that, when run on a computer, causes the computer to execute the uncertainty robot system state estimation method in Embodiment 1. For a detailed implementation process, please refer to the specific scheme in Embodiment 1, which will not be repeated here.

[0118] The robot system state estimation method of this invention addresses the issue of unknown mechanistic models and inherent uncertainties in robot systems. By extracting information from only a limited number of sensors, a linear polynomial equivalent system is constructed based on a multidimensional Taylor network. Combined with a latent variable processing mechanism, a virtual linear polynomial equivalent system of the uncertain robot system is built. Extracting the state variables from this virtual linear polynomial equivalent system yields the estimated state variables with high accuracy. This method is independent of specific sensors and requires no precise mechanistic model, offering advantages such as high computational accuracy and low hardware cost. It can be applied to various uncertain robot systems.

[0119] Those skilled in the art will readily understand that the above description is merely 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 within the scope of protection of the present invention.

Claims

1. A method for estimating the state of an uncertain robot system, characterized in that, include: S1. Construct a state-space model of the uncertain robot system based on parameters including state variables of the uncertain robot system; wherein, the state variables include at least one known state variable and a state variable to be estimated; the state-space model of the uncertain robot system is as follows: ； in, and They are respectively k time, k+ The uncertain robot system at time 1 N dimensional state vector, N The number of the state variables; for k+ The M-dimensional measurement vector of the uncertain robot system at time 1, where M is the number of known state variables; It is a nonlinear state function vector. Represents the i-th nonlinear state function; For nonlinear measurement function vectors, Represents the i-th nonlinear measurement function; Given the control input vector, for k Real-time noise, for k+ Measurement noise at time 1 and Unrelated; S2. The nonlinear state function and nonlinear measurement function in the state-space model are approximated using a multidimensional Taylor network; wherein, the nonlinear state function... and nonlinear measurement function They are respectively: ； ； , , and Let be the order of any state variable, and be a non-negative positive integer; and These are the weights corresponding to each order tensor in the multidimensional Taylor network model. and To approximate the error, r Indicates the defined dimensions of the Taylor net; S3. Employing a latent variable processing mechanism, robot state terms with similar characteristics in the nonlinear state function and nonlinear measurement function approximated by the multidimensional Taylor network are merged and summarized. Based on the merged and summarized nonlinear state function and nonlinear measurement function, the state space model is converted into a linear polynomial equivalent system. The latent variable processing mechanism includes: Define hidden variables and latent variables The weight is ;in, ; for k The i-th dimension of the uncertain robot system at time t; Based on the latent variables With the corresponding weights, robot state terms with the same characteristics in the nonlinear state function approximated by the multidimensional Taylor network are merged and summarized. Define hidden variables and latent variables The weight is ;in, ; Based on the latent variables With the corresponding weights, robot state terms with the same characteristics in the nonlinear measurement function approximated by the multidimensional Taylor network are merged and summarized; The linear polynomial equivalent system is: ； in, and The linear polynomial equivalent system is respectively Time and The state matrices at each time step have the same matrix structure. ; and This is the state transition matrix of the equivalent system; This is the output vector of the linear polynomial equivalent system; , and They are , and first-order terms to r A vector composed of terms of order, for k Real-time noise, for k+ Measurement noise at time 1; The control input vector is known. S4. Estimate the state of the linear polynomial equivalent system in real time to obtain the state variables to be estimated.

2. The method for estimating the state of an uncertain robot system according to claim 1, characterized in that, In S4, an equivalent extended Kalman observer is used to estimate the state of the linear polynomial equivalent system in real time, obtaining the state variables to be estimated, including: S41. Solve for the state of the linear polynomial equivalent system. One-step prediction estimate ,in, ; S42. Using the maximum correlation entropy Kalman filter strategy, solve the linear polynomial equivalent system. The optimal state estimate at time t is obtained, thereby yielding the state variables to be estimated; wherein, the linear polynomial equivalent system Optimal state estimate at time t for: ； In the formula, This represents the Kalman filter gain.

3. A state estimation device for an uncertain robot system, characterized in that, include: At least one processor, a data storage device connected to the processor, and a sensor data reading device; The sensor data reading device is used to read the known state quantities of the uncertain robot system and transmit them to the data storage. The processor executes the uncertainty robot system state estimation method according to claim 1 or 2 by reading the program in the data memory.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the uncertainty robot system state estimation method as described in claim 1 or 2.

5. A computer program product, characterized in that, When the computer program product is run on a computer, it causes the computer to execute the uncertainty robot system state estimation method according to any one of claims 1 or 2.

Images