Fast distributed humanoid robot state estimation method, system, equipment and medium

Abstract

The invention relates to a fast distributed humanoid robot state estimation method, system and device and a medium. The method comprises the following steps: acquiring IMU sensor data and vision measurement data; non-linear attitude estimation is carried out based on IMU sensor data and visual measurement data, and floating base attitude estimation information at the next moment is obtained; contact information and leg joint information are obtained, after physical constraint is carried out based on the contact information, linear movement window estimation is carried out according to the leg joint information and the attitude estimation information of the floating base at the next moment, and speed and position estimation information of the floating base at the next moment is obtained; and when linear movement window estimation is carried out, speed estimation information and position estimation information are optimized by adopting a marginalization method. According to the method, the calculation complexity can be reduced, and meanwhile, the accuracy of state estimation and the resistance to noise and model uncertainty are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot control, and particularly to a fast decentralized humanoid robot state estimation method, system, device, and medium. Background Art

[0002] When a humanoid robot walks in a complex and dynamic environment, accurate state estimation is crucial for achieving stable and efficient motion control. Currently, traditional centralized state estimation methods have many drawbacks. Their computational complexity is relatively high. When dealing with a large amount of data, they consume a large amount of computing resources and time, making it difficult for the robot to quickly make accurate state estimations in the face of a complex environment. At the same time, this centralized method has poor real-time performance and cannot meet the requirements of the robot for rapid response in a dynamic environment.

[0003] To overcome the deficiencies of traditional centralized state estimation methods, decentralized state estimation methods have emerged. This method can distribute computational tasks to multiple processing units, thereby effectively reducing the overall computational burden. This decentralized architecture enables the system to still maintain certain functions when facing partial unit failures, greatly improving the robustness of the system. Moreover, due to the distribution of computational tasks, each processing unit can process data in parallel, significantly improving the real-time response ability of the system, which better meets the application requirements of humanoid robots in complex dynamic environments.

[0004] However, in practical applications, how to achieve accurate and efficient state estimation under a decentralized architecture remains a challenge. On the one hand, the information interaction and cooperation mechanisms between processing units in a decentralized system are relatively complex. How to optimize these mechanisms to ensure the accurate transmission and effective fusion of information, so as to achieve accurate state estimation, is a key issue. On the other hand, while ensuring the accuracy of estimation, it is also necessary to consider computational efficiency and avoid excessive consumption of computing resources and degradation of real-time performance due to complex algorithms and excessive information interaction, which poses higher requirements for algorithm design and system architecture. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide a fast decentralized humanoid robot state estimation method, system, device, and medium that can reduce computational complexity and improve the accuracy of state estimation.

[0006] A fast decentralized humanoid robot state estimation method, the method includes: Obtain IMU sensor data and visual measurement data; perform non-linear attitude estimation based on the IMU sensor data and the visual measurement data to obtain the floating base attitude estimation information at the next moment; Obtain contact information and leg joint information. After performing physical constraints based on the contact information, perform a linear moving window estimation according to the leg joint information and the estimated information of the floating base attitude at the next moment to obtain the estimated information of the speed and position of the floating base at the next moment; When performing the linear moving window estimation, use the marginalization method to optimize the estimated information of the speed and position.

[0007] A fast decentralized humanoid robot state estimation system, the system includes: An attitude estimation module, configured to obtain IMU sensor data and visual measurement data; perform non-linear attitude estimation based on the IMU sensor data and the visual measurement data to obtain the estimated information of the floating base attitude at the next moment; A speed and position estimation module, configured to obtain contact information and leg joint information. After performing physical constraints based on the contact information, perform a linear moving window estimation according to the leg joint information and the estimated information of the floating base attitude at the next moment to obtain the estimated information of the speed and position of the floating base at the next moment; An optimization module, configured to use the marginalization method to optimize the estimated information of the speed and position when performing the linear moving window estimation.

[0008] A computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the fast decentralized humanoid robot state estimation method are implemented.

[0009] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the fast decentralized humanoid robot state estimation method are implemented.

[0010] For the above-mentioned fast decentralized humanoid robot state estimation method, system, device and medium, by obtaining IMU sensor data and visual measurement data; performing non-linear attitude estimation based on the IMU sensor data and the visual measurement data to obtain the estimated information of the floating base attitude at the next moment; obtaining contact information and leg joint information, performing physical constraints based on the contact information, and performing a linear moving window estimation according to the leg joint information and the estimated information of the floating base attitude at the next moment to obtain the estimated information of the speed and position of the floating base at the next moment; when performing the linear moving window estimation, use the marginalization method to optimize the estimated information of the speed and position.

[0011] The present invention decomposes the entire state estimation task of a humanoid robot into two sub-modules. Each module processes specific sensor data locally, achieving decentralized computing and improving the real-time performance of computing. This method can improve the accuracy of state estimation and the resistance to noise and model uncertainty while reducing the computational complexity through decentralized non-linear attitude estimation and linear moving window estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the structures shown in these drawings without creative efforts.

[0013] Figure 1 It is a schematic flowchart of the fast decentralized humanoid robot state estimation method provided in Embodiment 1; Figure 2 It is a block diagram of the structure of the fast decentralized humanoid robot state estimation system provided in Embodiment 2; Figure 3 It is an internal structure diagram of the computer device provided in Embodiment 3.

[0014] The implementation, functional features, and advantages of the present invention will be further described in conjunction with the embodiments and with reference to the drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0016] It can be understood that the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0017] The following will describe the embodiments of the present invention in detail with reference to the drawings in the embodiments of the present invention.

[0018] Embodiment 1 This embodiment discloses a fast decentralized humanoid robot state estimation method. By decomposing the entire state estimation task of the humanoid robot into two sub-modules, each module processes specific sensor data locally, achieving the decentralization of calculations and improving the real-time performance of calculations. This method can improve the accuracy of state estimation and the resistance to noise and model uncertainty while reducing the computational complexity through decentralized non-linear attitude estimation and linear moving window estimation.

[0019] As Figure 1 shown, the fast decentralized humanoid robot state estimation method provided in this embodiment includes the following steps: Step 201: Obtain IMU sensor data and visual measurement data; perform non-linear attitude estimation based on the IMU sensor data and visual measurement data to obtain the floating base attitude estimation information at the next moment.

[0020] Step 202: Obtain contact information and leg joint information. After performing physical constraints based on the contact information, perform linear moving window estimation according to the leg joint information and the floating base attitude estimation information at the next moment to obtain the speed and position estimation information of the floating base at the next moment.

[0021] Step 203: When performing linear moving window estimation, use the marginalization method to optimize the speed and position estimation information.

[0022] In the specific implementation process of step 201, the IMU sensor data is obtained through a nine-axis IMU, and the IMU sensor data can provide a reasonable attitude measurement data . In addition, considering that the rapid change of acceleration will cause the attitude measurement data to have estimation drift, therefore, visual measurement data is introduced as additional correction data. The visual measurement data is obtained through image acquisition devices such as cameras and is mainly image data. In this step, based on the projection of gravitational acceleration for correction and by fusing additional visual measurement data, the accuracy and stability of attitude estimation can be improved, the environmental adaptability and interaction ability can be enhanced, and the accuracy and robustness of motion control can be improved.

[0023] In addition, the motion of a humanoid robot is often non-linear. Therefore, using non-linear attitude estimation for a humanoid robot can improve the estimation accuracy, enhance stability and robustness. In this embodiment, the non-linear attitude estimation is mainly completed through the Extended Kalman Filter (EKF).

[0024] Specifically, a pose dynamics model is constructed, and preliminary non-linear pose estimation is performed based on IMU sensor data to obtain the initial pose estimation information at the next moment; the drifts of the roll and pitch axes in the initial pose estimation information at the next moment are corrected by measuring the projection of the gravitational acceleration in the body coordinate system; and the pose measurement values related to vision in the initial pose estimation information at the next moment are corrected through visual measurement data to obtain the final floating base pose estimation information at the next moment.

[0025] Among them, the mathematical expression of the pose dynamics model is: ; In the formula, represents the initial pose estimation information at the next moment; represents the identity matrix, which is a 4×4 identity matrix; represents the angular rate; represents converting the rotational velocity vector into a cross product matrix; represents the time interval; represents the collected pose information, which is a pose quaternion; represents the first-order approximate Jacobian matrix of the exponential map; represents the angular velocity vector; represents the noise vector related to the angular velocity. Among them, is the first-order approximate Jacobian matrix of the exponential map with respect to the quaternion ; is the first-order approximate Jacobian matrix of the exponential map with respect to the angular rate .

[0026] The drifts of the roll and pitch axes in the initial pose estimation information at the next moment are corrected by measuring the projection of the gravitational acceleration in the body coordinate system, and the calculation expression is: ; ; In the formula, represents the gravitational acceleration projection information; represents the pose matrix; represents the gravitational acceleration; represents the ratio of the magnitude of the acceleration vector to the magnitude of the gravitational acceleration vector; represents the noise vector related to the acceleration; represents the vector transpose, among which, represents a replaceable variable.

[0027] The pose measurement values related to vision in the initial pose estimation information at the next moment are corrected through visual measurement data to obtain the final floating base pose estimation information at the next moment, and the calculation expression is: ; In the formula, represents the floating base attitude estimation information at the next moment; represents the attitude information collected currently; represents the measurement noise of the attitude; represents the attitude part in the estimation information; represents the estimation information including attitude and position.

[0028] It can be understood that is the attitude part of and is the noise of the attitude measurement in the visual measurement data. In the buffer of the stored IMU measurement values, when the new visual measurement data arrives at time

[0029] In the specific implementation process of step 202, the contact information is obtained through the foot-end force sensor, mainly including data such as contact force. The leg joint information is obtained through the joint encoder, mainly including data such as joint angle and angular velocity. When estimating the velocity and position, the constructed evolution process model, measurement model, and additional physical constraints are all represented by state constraints. In addition, in order to better conform to the motion characteristics of the humanoid robot, linear moving window estimation is adopted for the position and velocity, which can quickly update the estimation of the robot's position and velocity according to the latest observation data at each new time step, and has strong real-time performance and high efficiency. In this embodiment, the linear moving window estimation is mainly completed through moving horizon estimation (MHE).

[0030] Specifically, physical constraints are imposed based on the contact information, including: based on the contact information, a stable constraint model for the contact between the robot's foot-end and the ground is constructed, and the expression is: ; In the formula, represents the position of the foot-end at the next moment; represents the current position of the foot-end; represents the ground normal force acting on the foot.

[0031] It can be seen that when constructing the stable constraint model, the contact between the foot-end and the ground is modeled as a deterministic constraint, that is, the foot will not slide on the ground, so as to construct the stable constraint model when the foot is in stable contact.

[0032] In the above stable constraint model, when using a contact sensor, it is equivalent to a Boolean variable. Alternatively, the static constraint of the foot position can be replaced by the velocity constraint instead.

[0033] When performing linear moving window estimation, the linear state constraints generated during the constraint process are uniformly described as: ; In this embodiment, the physical constraints provide prior knowledge of the system. By considering both measurement constraints and physical constraints simultaneously, MHE can better handle sensor noise and outliers, improving the accuracy and robustness of state estimation. When controlling to enforce static foot contact, the setting of physical constraints enables the humanoid robot to complete tasks well and eliminates the need for extensive adjustment of the sliding covariance. When the humanoid robot walks on a slippery terrain, the stochastic contact model in EKF can still be enforced in MHE.

[0034] Based on the leg joint information and the estimated information of the floating base attitude at the next moment, linear moving window estimation is performed to obtain the estimated information of the speed and position of the floating base at the next moment, including: Construct a state vector of speed and position, expressed as: ; At each discrete time, use the equivalent control input: ; Then, the state vector of speed and position undergoes incident linear dynamic evolution, and the expression of the evolution process model is: ; At each time index , when the above evolution process model performs linear moving window estimation, continuous state constraints are formed, expressed as: .

[0035] In addition, when the state vector of speed and position undergoes incident linear dynamic evolution, the measured output is taken as the foot end position of each leg in the leg joint information, and a measurement constraint model is constructed, with the expression: ; In the formula, represents the state vector composed of position, speed, foot end position, and additional deviation terms; represents the position of the humanoid robot in the world coordinate system; represents the speed of the humanoid robot in the world coordinate system; represents the foot end position of the humanoid robot in the world coordinate system; represents the acceleration offset represented in the body coordinate system; represents the attitude matrix corresponding to the estimated attitude value; represents the acceleration vector; represents the acceleration due to gravity; is the state vector representing the velocity and position of the floating base at the next moment; represents the state vector noise; represents the time interval; represents the estimate of the acceleration vector; represents the representation vector measurement value; vector measurement matrix; represents the vector at moment state; represents the measurement noise; represents N discrete moments included in the moving window, where represents the number of discrete moments included in the moving window, represents each discrete moment; represents the noise applied to the state vector; represents the state transition matrix; represents the state transition vector; represents noise applied to the state vector at the

[0036] It can be understood that the measurement constraint provides the immediate information observed by the leg joint sensors. In the linear moving window estimation, the measurement constraint and the physical constraint jointly define the feasible solution space. When performing the optimization calculation, the optimal solution is usually searched within this space to minimize the difference between the estimated state and the measurement data while satisfying the operating conditions of the physical system.

[0037] In the specific implementation process of step 203, when the window in the linear moving window estimation slides forward, the oldest measurement data in the window needs to be removed from the optimization problem, and the new measurement data needs to be added. The marginalization method removes the influence of the oldest measurement data from the current optimization problem by calculating the arrival cost, including the following steps: Step 301, construct an optimization problem including all measurement data, and the solution of this problem is the state estimation within the current window.

[0038] Step 302, solve the KKT (Karush-Kuhn-Tucker) conditions to calculate the marginalization result of the oldest data. In this step, the inverse of the covariance matrix of the oldest data is calculated, and the covariance matrix of the current state is updated.

[0039] Step 303, the marginalization result is used as the arrival cost to initialize a new optimization problem. In this way, the equivalence with the non-linear attitude estimation is ensured during the linear moving window estimation process when the window slides.

[0040] Through marginalization, the computational burden can be effectively reduced because there is no need to recalculate the data of the entire window. At the same time, the marginalization method ensures that the optimal state estimation can still be provided each time the window is updated during the linear moving window estimation process.

[0041] It should be noted that the method provided in this embodiment is applicable to humanoid robots with IMU, leg joint encoders, and visual information input. However, it can also be applied to other humanoid robots with similar data acquisition functions according to the situation. For example, the IMU can be replaced by a motion capture system; the leg joint encoder can be replaced by a linear displacement sensor; the visual information input can be replaced by a lidar.

[0042] Although each step in this embodiment Figure 1 is shown in sequence according to the indication of the arrow, these steps do not necessarily need to be executed in sequence according to the indication of the arrow. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, Figure 1 at least a part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages do not necessarily need to be executed at the same moment, but can be executed at different moments. The execution order of these sub-steps or stages does not necessarily need to be sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0043] Embodiment 2 Based on the fast decentralized humanoid robot state estimation method in Embodiment 1, this embodiment discloses a fast decentralized humanoid robot state estimation system. As Figure 2 shown, the fast decentralized humanoid robot state estimation system includes: an attitude estimation module 401, a velocity and position estimation module 402, and an optimization module 403, where: The attitude estimation module 401 is used to obtain IMU sensor data and visual measurement data; perform non-linear attitude estimation based on the IMU sensor data and visual measurement data to obtain the floating base attitude estimation information at the next moment. That is to say, the attitude estimation module 401 can be used to process acceleration, angular velocity data, and visual measurement data.

[0044] The velocity and position estimation module 402 is used to obtain contact information and leg joint information. After performing physical constraints based on the contact information, a linear moving window estimation is performed according to the leg joint information and the floating base attitude estimation information at the next moment, so as to obtain the velocity and position estimation information of the floating base at the next moment. The velocity and position estimation module 402 can be used to process multi-rate sensor data and perform optimization within a fixed window.

[0045] The optimization module 403 is used to optimize the velocity and position estimation information by using the marginalization method when performing the linear moving window estimation. The optimization module 403 is based on the optimization structure of the full information filter and mainly calculates the arrival cost of the linear moving window estimation problem.

[0046] In this embodiment, the specific working processes and working principles of the attitude estimation module 401, the velocity and position estimation module 402, and the optimization module 403 are the same as those of the method in Embodiment 1, so they will not be elaborated herein. Each of these unit modules can be implemented in whole or in part by software, hardware, and their combinations. Each unit module can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of these unit modules.

[0047] Embodiment 3 As Figure 3 shown, a terminal device disclosed in this embodiment includes a transmitter, a receiver, a memory, and a processor. Among them, the transmitter is used to send instructions and data, the receiver is used to receive instructions and data, the memory is used to store computer execution instructions, and the processor is used to execute the computer execution instructions stored in the memory to implement the method in Embodiment 1 above.

[0048] It should be noted that the above memory can be either independent or integrated with the processor. When the memory is set independently, the terminal device further includes a bus for connecting the memory and the processor.

[0049] Embodiment 4 This embodiment discloses a computer-readable storage medium in which computer execution instructions are stored. When the processor executes the computer execution instructions, the method in Embodiment 1 above is implemented.

[0050] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0051] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0052] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the appended claims.

Claims

1. A fast decentralized humanoid robot state estimation method, characterized in that: The method comprises: Acquire IMU sensor data and visual measurement data; perform nonlinear attitude estimation based on the IMU sensor data and the visual measurement data to obtain floating base attitude estimation information at the next moment; Acquire contact information and leg joint information, perform physical constraints based on the contact information, perform linear moving window estimation based on the leg joint information and the next moment floating base posture estimation information, and obtain the next moment floating base speed and position estimation information; When performing linear moving window estimation, a marginalization method is used to optimize the speed and position estimation information.

2. The fast decentralized humanoid robot state estimation method according to claim 1, characterized in that: Nonlinear attitude estimation is performed based on the IMU sensor data and the visual measurement data to obtain the floating base attitude estimation information at the next moment, including: Constructing a posture dynamics model, performing preliminary nonlinear posture estimation based on the IMU sensor data, and obtaining initial posture estimation information at the next moment; The drift of the roll and pitch axes in the initial attitude estimation information at the next moment is corrected by measuring the projection of the gravitational acceleration in the body coordinate system; and the attitude measurement values ​​related to vision in the initial attitude estimation information at the next moment are corrected by the visual measurement data to obtain the final floating base attitude estimation information at the next moment.

3. The fast decentralized humanoid robot state estimation method according to claim 2, characterized in that: The mathematical expression of the posture dynamics model is: ； In the formula, Indicates the initial posture estimation information for the next moment; represents the identity matrix; represents the angular rate; Represents the conversion of the rotation velocity vector into a cross product matrix; Indicates the interval time; Indicates the currently collected posture information; The first-order approximate Jacobian matrix representing the exponential map; represents the angular velocity vector; represents the noise vector associated with the angular velocity.

4. The fast decentralized humanoid robot state estimation method according to claim 2, characterized in that: The drift of the roll and pitch axes in the initial attitude estimation information at the next moment is corrected by measuring the projection of the gravity acceleration in the body coordinate system. The calculation expression is: ； In the formula, Represents the gravitational acceleration projection information; represents the posture matrix; represents the acceleration due to gravity; It represents the ratio of the magnitude of the acceleration vector to the magnitude of the gravity acceleration vector; represents the noise vector associated with acceleration; represents the vector transpose, where Indicates a replaceable variable.

5. The fast decentralized humanoid robot state estimation method according to claim 2, characterized in that: The visual measurement data is used to correct the visual-related attitude measurement value in the initial attitude estimation information at the next moment, and the final attitude estimation information of the floating base at the next moment is obtained. The calculation expression is: ； In the formula, Indicates the estimated attitude information of the floating base at the next moment; Indicates the currently collected posture information; The measurement noise representing the attitude; Represents the posture part of the estimated information; The representation includes estimated information of pose and position.

6. The fast decentralized humanoid robot state estimation method according to any one of claims 1 to 5, characterized in that: Physical constraints are performed based on the contact information, including: Based on the contact information, a stable constraint model of the robot foot end in contact with the ground is constructed, and the expression is: ； In the formula, Indicates the position of the foot at the next moment; Indicates the current position of the foot end; is the normal force due to the ground acting on the foot.

7. The fast decentralized humanoid robot state estimation method according to claim 6, characterized in that: According to the leg joint information and the floating base posture estimation information at the next moment, a linear moving window estimation is performed to obtain the speed and position estimation information of the floating base at the next moment, including: Construct the state vector of velocity and position, expressed as: ； At each discrete time, use the equivalent control input: ； Then, the state vectors of velocity and position are Carry out linear dynamic evolution of events, and the model expression of the evolution process is: ； The evolution process model forms a continuous state constraint, which is: ； The state vectors for velocity and position When performing linear dynamic evolution of events, the leg joint information is used as the measurement output to construct a measurement constraint model, which is expressed as follows: ； In the formula, represents the state vector consisting of position, velocity, foot end position and additional deviation term; represents the position of the humanoid robot in the world coordinate system; represents the velocity of the humanoid robot in the world coordinate system; Indicates the foot position of the humanoid robot in the world coordinate system; represents the acceleration offset expressed in the body coordinate system; The attitude matrix representing the corresponding attitude estimate; represents the acceleration vector; represents the acceleration due to gravity; Represents the velocity estimate and position estimate state vector of the floating base at the next moment; Indicates the interval time; represents an estimate of the acceleration vector; Representation vector The measured value of Representation vector The measurement matrix of Representation vector exist The state of the moment; represents the measurement noise; Represents the N discrete moments contained in the moving window; represents the noise applied to the state vector; represents the state transfer matrix; represents the state transition vector; express The noise applied to the state vector at all times.

8. A fast decentralized humanoid robot state estimation system, characterized in that: The system comprises: The attitude estimation module is used to obtain IMU sensor data and visual measurement data; perform nonlinear attitude estimation based on the IMU sensor data and the visual measurement data to obtain the floating base attitude estimation information at the next moment; A speed and position estimation module is used to obtain contact information and leg joint information, and after physical constraints are performed based on the contact information, a linear moving window estimation is performed based on the leg joint information and the floating base posture estimation information at the next moment, to obtain the speed and position estimation information of the floating base at the next moment; The optimization module is used to optimize the speed and position estimation information by using a marginalization method when performing linear moving window estimation.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for fast decentralized humanoid robot state estimation according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for fast decentralized humanoid robot state estimation according to any one of claims 1 to 7 are implemented.

Images