Humanoid robot navigation method based on fuzzy neural network and adaptive PID hybrid control algorithm

Abstract

The invention discloses a humanoid robot navigation method based on a fuzzy neural network and an adaptive PID hybrid control algorithm, and the method comprises the following steps: S1, measuring distances: measuring the distances between a humanoid robot and a forward obstacle, a right obstacle and a left obstacle by using an infrared sensor; s2, carrying out extended Kalman filtering to remove noise; s3, defining a fuzzy control variable and determining a value range; s4, generating a fuzzy membership function; s5, constructing a fuzzy neural network; according to the method, the extended Kalman filtering technology is creatively used for giving out optimal estimation in the maximum likelihood sense for data obtained through measurement of the infrared sensor, the influence of noise on the system is reduced, the fuzzy neural network is introduced to replace a fuzzy control algorithm, and the fuzzy control algorithm is optimized. The input of the neural network is changed into a fuzzy input signal and a fuzzy weight, and the output of the neural network is defuzzified to obtain a visual effective value.

Description

Technical Field

[0001] The present invention belongs to the technical field of robots, and particularly relates to a navigation method for humanoid robots based on a hybrid control algorithm of a fuzzy neural network and an adaptive PID. Background Art

[0002] In recent years, with the progress of technology and the development of society, the intelligence level of robots has been gradually improved. Because of its similar human body shape and the ability to imitate humans in many aspects, humanoid robots have gradually become a research hotspot in recent years. Among them, the navigation problem is one of the research focuses of humanoid robots. In a dynamic and complex environment, how to achieve high-precision and high-robustness autonomous navigation has become the core challenge restricting the commercial application of humanoid robots.

[0003] At present, the Global Positioning System (GPS) is a commonly used technical means for humanoid robot navigation. It determines the absolute position of the robot through satellite signals and combines map data for path planning. This method is technically mature and low-cost, but it is only applicable to outdoor open scenarios. In GPS-denied environments such as indoors, underground, or in bad weather, satellite signals are easily blocked or interfered, resulting in a significant decrease in positioning accuracy and unable to meet the centimeter-level positioning requirements in scenarios such as medical assistance and warehousing logistics.

[0004] In view of the limitations of GPS, various alternative solutions have been proposed in the prior art. Among them, Patent CN119596675A proposes a navigation method for humanoid robots based on a hybrid fuzzy embedded PID control algorithm, which realizes the autonomous navigation of humanoid robots through fuzzy reasoning and a PID controller. However, this solution has defects such as being sensitive to infrared sensor noise, poor adaptability in high-fuzziness situations, slow response due to fixed PID controller parameters, and large steady-state errors. Therefore, it is necessary to propose a navigation method that can suppress infrared sensor noise, has strong adaptability in high-fuzziness situations, fast response speed, and smaller steady-state errors to solve the above problems, so that humanoid robots can better adapt to complex and dynamic navigation environments. Summary of the Invention

[0005] The purpose of the present invention is to provide a navigation method for humanoid robots based on a hybrid control algorithm of a fuzzy neural network and an adaptive PID, and innovatively uses the extended Kalman filter technology to give the best estimate in the sense of maximum likelihood for the data measured by the infrared sensor, reducing the influence of noise on the system.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a navigation method for humanoid robots based on a hybrid control algorithm of a fuzzy neural network and an adaptive PID, including the following steps:

[0007] Step S1: Measure distances. Use an infrared sensor to measure the distances between the humanoid robot and the forward obstacle, the right obstacle, and the left obstacle.

[0008] Step S2: Remove noise using the Extended Kalman Filter. Use the Extended Kalman Filter algorithm to process the measurement data of the infrared sensor to obtain the optimal state estimates of the forward obstacle distance, the right obstacle distance, and the left obstacle distance.

[0009] Step S3: Define fuzzy control variables and determine the value ranges. Define the fuzzy control variables as the optimal state estimates of the forward obstacle distance, the right obstacle distance, and the left obstacle distance, and the optimized turning angle, and determine the value ranges. Specifically: the optimal state estimate of the forward obstacle distance is 30 - 84, the optimal state estimate of the right obstacle distance is 30 - 84, the optimal state estimate of the left obstacle distance is 30 - 84, and the optimized turning angle is -35 - 35.

[0010] Step S4: Generate fuzzy membership functions. Select Gaussian membership functions, define the membership ranges, and divide the optimal state estimates of the input variables of the forward obstacle distance, the right obstacle distance, and the left obstacle distance into eight levels respectively, which are 30 - 42, 36 - 48, 42 - 54, 48 - 60, 54 - 66, 60 - 72, 66 - 78, 72 - 84. The eight levels are correspondingly represented as {extremely close, very close, quite close, close, far, very far, extremely far, extremely distant}.

[0011] Step S5: Construct and train a fuzzy neural network. Construct a fuzzy neural network, take the optimal state estimates of the forward obstacle distance, the right obstacle distance, and the left obstacle distance corresponding to the same moment as the input quantities, and the optimized turning angle at this moment as the output quantity, and train the fuzzy neural network. Among them, the fuzzy neural network includes: an input layer, a fuzzification layer, a fuzzy rule calculation layer, a normalization layer, and an output layer. The input layer is the first layer of the fuzzy neural network, which is used to input the optimal state estimate x1 of the forward obstacle distance, the optimal state estimate x2 of the right obstacle distance, and the optimal state estimate x3 of the left obstacle distance. Each node in the first layer is connected to each component of the input vector, and the input value is transmitted to the next layer. The number of nodes in this layer is 3.

[0012] Step S6: Control the humanoid robot using an adaptive PID controller. Input the optimal state estimates of the forward obstacle distance, the right obstacle distance, and the left obstacle distance at the current moment into the trained fuzzy neural network to obtain the optimized turning angle at the current moment. Input the obtained optimized turning angle at the current moment into the adaptive PID controller to generate a control signal to adjust the angles and torques of the robot joints, and the navigation control of the humanoid robot can be achieved.

[0013] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are as follows:

[0014] 1. The present invention innovatively uses the extended Kalman filtering technology to give the best estimate in the sense of maximum likelihood for the data measured by the infrared sensor, reducing the influence of noise on the system.

[0015] 2. The present invention innovatively introduces a fuzzy neural network to replace the fuzzy control algorithm. The input of the neural network is processed by the fuzzy system to become fuzzy input signals and fuzzy weights, and the output of the neural network is defuzzified to obtain an intuitive effective value, that is, the membership function and fuzzy rules of the fuzzy system are added to the hidden nodes of the neural network, giving full play to the parallel processing ability of the neural network and the reasoning ability of the fuzzy system, and greatly enhancing the adaptive ability in the case of high fuzziness.

[0016] 3. The present invention innovatively uses an adaptive PID controller to replace the traditional PID controller. By adaptively adjusting the three parameters of the proportional gain, integral gain, and derivative gain, the response speed of the humanoid robot is increased and the steady-state error is greatly reduced, enabling the humanoid robot to better adapt to complex and dynamic navigation environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 It is a flowchart of the humanoid robot navigation method based on the hybrid fuzzy embedded PID control algorithm provided by the present invention.

[0019] Figure 2 It is a schematic diagram of the fuzzy neural network structure provided by the present invention.

[0020] Figure 3 It is a fuzzy membership function graph of the best state estimate value of the forward obstacle distance provided by the present invention.

[0021] Figure 4 It is a fuzzy membership function graph of the best state estimate value of the right obstacle distance provided by the present invention.

[0022] Figure 5 It is a fuzzy membership function graph of the best state estimate value of the left obstacle distance provided by the present invention.

[0023] Figure 6 It is a schematic diagram of the principle of the extended Kalman filter provided by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0024] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts in the embodiments of the present invention belong to the scope of protection of the present invention.

[0025] The following specifically introduces a humanoid robot navigation method provided by the present invention based on a fuzzy neural network and an adaptive PID hybrid control algorithm, including the following steps:

[0026] Step S1: Measure the distance, and use an infrared sensor to measure the distances between the humanoid robot and the forward obstacle, the right obstacle, and the left obstacle.

[0027] Step S2: Remove noise by extended Kalman filtering, and use the extended Kalman filtering algorithm to process the measurement data of the infrared sensor to obtain the optimal state estimation values of the forward obstacle distance, the right obstacle distance, and the left obstacle distance.

[0028] Step S3: Define fuzzy control variables and determine the value range. Define the fuzzy control variables as the optimal state estimation values of the forward obstacle distance, the right obstacle distance, and the left obstacle distance, and the optimized turning angle, and determine the value range. Specifically: the optimal state estimation value of the forward obstacle distance is 30 - 84, the optimal state estimation value of the right obstacle distance is 30 - 84, the optimal state estimation value of the left obstacle distance is 30 - 84, and the optimized turning angle is -35 - 35.

[0029] Step S4: Generate fuzzy membership functions. Select Gaussian membership functions, define the membership range, and divide the optimal state estimation values of the input variables of the forward obstacle distance, the right obstacle distance, and the left obstacle distance into eight levels, which are 30 - 42, 36 - 48, 42 - 54, 48 - 60, 54 - 66, 60 - 72, 66 - 78, 72 - 84 respectively. The eight levels are correspondingly represented as {extremely close, very close, quite close, close, far, very far, extremely far, extremely distant}.

[0030] Step S5: Construct a fuzzy neural network and train it. Construct a fuzzy neural network, using the optimal state estimates of the forward obstacle distance, right obstacle distance, and left obstacle distance corresponding to the same moment as input variables, and the optimized turning angle at this moment as the output variable, and train the fuzzy neural network. Among them, the fuzzy neural network includes: an input layer, a fuzzification layer, a fuzzy rule calculation layer, a normalization layer, and an output layer. The input layer is the first layer of the fuzzy neural network, which is used to input the optimal state estimate x1 of the forward obstacle distance, the optimal state estimate x2 of the right obstacle distance, and the optimal state estimate x3 of the left obstacle distance. Each node in the first layer is connected to each component of the input vector, and the input value is transmitted to the next layer. The number of nodes in this layer is 3

[0031] As Figure 2 shown, the fuzzification layer is the second layer of the fuzzy neural network, which is used to calculate the optimal state estimates x1 of the forward obstacle distance value, x2 of the right obstacle distance value, and x3 of the left obstacle distance value input by the input layer according to the fuzzy membership function generated in step S4. The specific expression is:

[0032]

[0033] where c ij is the center of the membership function of the i-th input variable belonging to the j-th fuzzy subset of its fuzzy rule, (i = 1, 2, 3, j = 1, 2...8);

[0034] σ ij is the width of the membership function of the i-th input variable belonging to the j-th fuzzy subset of its fuzzy rule, (i = 1, 2, 3, j = 1, 2...8);

[0035] x i is the input value of the i-th input variable (i = 1, 2, 3);

[0036] is the fuzzy membership function value of the i-th input variable belonging to the j-th fuzzy subset of its fuzzy rule (i = 1, 2, 3, j = 1, 2...8).

[0037] Using the above expressions for calculation, for each fuzzy subset, the optimal state estimation values of the forward obstacle distance, right obstacle distance, and left obstacle distance input at the input layer each obtain a fuzzy membership function value, which is a node in the fuzzification layer; for this embodiment, the first eight nodes form a group of nodes, which are the fuzzy membership function values of the optimal state estimation value of the forward obstacle distance belonging to eight hierarchical fuzzy subsets; the middle eight nodes form a group of nodes, which are the fuzzy membership function values of the optimal state estimation value of the right obstacle distance belonging to eight hierarchical fuzzy subsets; the last eight nodes form a group of nodes, which are the fuzzy membership function values of the optimal state estimation value of the left obstacle distance belonging to eight hierarchical fuzzy subsets.

[0038] As Figure 2 shown, the fuzzy rule calculation layer is the third layer of the fuzzy neural network, which is used to complete the fuzzy inference operation. The fuzzy operator used for the fuzzy inference operation is the multiplication operator. The nodes of the third layer are the values obtained by combining a non-repeating fuzzy membership function value from each group of nodes in the second layer, that is, the direct product result is obtained;

[0039] Among them, the expression for the direct product result of the fuzzy rule calculation layer in the fuzzy neural network is:

[0040]

[0041] Among them, α i is the fuzzy direct product calculation value corresponding to the i-th node in the fuzzy rule calculation layer of the fuzzy neural network; (i represents the i-th node in the third layer, i = 1, 2... 512);

[0042] is the fuzzy membership function value of the optimal state estimation value of the forward obstacle distance as the input variable belonging to the j1-th fuzzy subset of its fuzzy rule (j1 = 1, 2... 8);

[0043] is the fuzzy membership function value of the optimal state estimation value of the right obstacle distance as the input variable belonging to the j2-th fuzzy subset of its fuzzy rule (j2 = 1, 2... 8);

[0044] is the fuzzy membership function value of the optimal state estimation value of the left obstacle distance as the input variable belonging to the j3-th fuzzy subset of its fuzzy rule (j3 = 1, 2... 8);

[0045] It should be noted that for a given input, only near the input point is there a relatively large value of the fuzzy membership function, and the value of the fuzzy membership function far from the input point is very small; when the value of the fuzzy membership function is very small (in this embodiment, less than 0.05 is considered very small), it is approximately taken as 0. Therefore, in the third layer, only a small number of nodes have a non-zero output, while the output of most nodes is 0.

[0046] As Figure 2 shown, the normalization layer is the fourth layer of the fuzzy neural network, which is used to normalize the calculation result of the fuzzy rule calculation layer; the number of nodes in the fourth layer is the same as that in the third layer, and the calculation formula for the node value of this layer is:

[0047]

[0048] is the value obtained after normalizing the fuzzy direct product calculation value corresponding to the j-th node of the fuzzy rule calculation layer (j = 1, 2... m);

[0049] α i is the fuzzy direct product calculation value corresponding to the i-th node of the fuzzy rule calculation layer (i = 1, 2... m);

[0050] α j is the fuzzy direct product calculation value corresponding to the j-th node of the fuzzy rule calculation layer (j = 1, 2... m); (both i and j represent the nodes in the third layer);

[0051] m is the total number of nodes in the third layer of the fuzzy neural network (m = 512).

[0052] As Figure 2 shown, the output layer is the fifth layer of the fuzzy neural network, which is used to defuzzify the normalization result and output the optimized turning angle. The calculation formula is:

[0053]

[0054] In the formula, w 1j is the weight coefficient of the j-th node in the fourth layer (j = 1, 2... 512);

[0055] is the value obtained after normalizing the fuzzy direct product calculation value corresponding to the j-th node of the fuzzy rule calculation layer;

[0056] y1 is the output value of the fuzzy neural network; that is, the optimized turning angle.

[0057] Step S6: Use an adaptive PID controller to control the humanoid robot;

[0058] The optimal state estimation values of the forward obstacle distance, right obstacle distance, and left obstacle distance at the current moment are input into the trained fuzzy neural network to obtain the optimized turning angle at the current moment. The optimized turning angle at the current moment obtained is input into the adaptive PID controller to generate a control signal to adjust the angles and torques of the robot joints, enabling the navigation control of the humanoid robot.

[0059] Among them, the adaptive PID controller adjusts the system output to reach the desired set value by combining three basic control strategies - proportional control, integral control, and derivative control. And the three parameters of the proportional gain, integral gain, and derivative gain adaptively change with the change of the environment, which can be represented by the following equations:

[0060]

[0061] Where:

[0062] e(t) is the error between the set point and the process variable;

[0063] U(t) is the controller output;

[0064] K p (t) is the proportional gain;

[0065] K i (t) is the integral gain;

[0066] K d (t) is the derivative gain;

[0067] The adjustment formula for the proportional gain is:

[0068]

[0069] Where:

[0070] K p (t) is the proportional gain;

[0071] K p0 is the initial proportional gain;

[0072] α p is the adjustment coefficient;

[0073] e(t) is the error between the set point and the process variable; e th is the error threshold;

[0074] σ p is the smoothing factor;

[0075] The adjustment formula for the integral gain is:

[0076]

[0077] Wherein:

[0078] K i (t) is the integral gain;

[0079] K i0 is the initial integral gain;

[0080] α i is the adjustment coefficient;

[0081] e(t) is the error between the setpoint and the process variable; the differential gain adjustment formula is:

[0082]

[0083] K d (t) is the differential gain;

[0084] K d0 is the initial differential gain;

[0085] α d is the adjustment coefficient;

[0086] e(t) is the error between the setpoint and the process variable; σ d is the sensitivity factor.

[0087] In this paper, the optimized turning angle obtained at the current moment is input into the adaptive PID controller to generate a control signal to adjust the angles and torques of the robot joints, and the navigation control of the humanoid robot can be realized.

[0088] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific embodiments. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A navigation method for a humanoid robot based on a hybrid control algorithm of a fuzzy neural network and an adaptive PID, characterized in that, It includes the following steps: Step S1: Measure distances. Use an infrared sensor to measure the distances between the humanoid robot and the forward obstacle, the right obstacle, and the left obstacle. Step S2: Remove noise using extended Kalman filtering. Use the extended Kalman filtering algorithm to process the measurement data of the infrared sensor to obtain the optimal state estimation values of the forward obstacle distance, the right obstacle distance, and the left obstacle distance. Step S3: Define fuzzy control variables and determine the value ranges. Define the fuzzy control variables as the optimal state estimation values of the forward obstacle distance, the right obstacle distance, and the left obstacle distance, and the optimized turning angle, and determine the value ranges. Specifically, the optimal state estimation value of the forward obstacle distance is 30 - 84, the optimal state estimation value of the right obstacle distance is 30 - 84, the optimal state estimation value of the left obstacle distance is 30 - 84, and the optimized turning angle is -35 - 35. Step S4: Generate fuzzy membership functions. Select Gaussian membership functions, define the membership ranges, and divide the optimal state estimation values of the input variables of the forward obstacle distance, the right obstacle distance, and the left obstacle distance into eight levels, which are 30 - 42, 36 - 48, 42 - 54, 48 - 60, 54 - 66, 60 - 72, 66 - 78, 72 - 84 respectively. The eight levels are correspondingly represented as {extremely close, very close, quite close, close, far, very far, extremely far, extremely distant}. Step S5: Construct and train a fuzzy neural network. Construct a fuzzy neural network. Use the optimal state estimation values of the forward obstacle distance, the right obstacle distance, and the left obstacle distance corresponding to the same moment as the input quantities, and the optimized turning angle at this moment as the output quantity to train the fuzzy neural network. Among them, the fuzzy neural network includes: an input layer, a fuzzification layer, a fuzzy rule calculation layer, a normalization layer, and an output layer. The input layer is the first layer of the fuzzy neural network, which is used to input the optimal state estimation value x1 of the forward obstacle distance, the optimal state estimation value x2 of the right obstacle distance, and the optimal state estimation value x3 of the left obstacle distance. Each node in the first layer is connected to each component of the input vector to transmit the input value to the next layer. The number of nodes in this layer is 3. Step S6: Control the humanoid robot using an adaptive PID controller. Input the optimal state estimation values of the forward obstacle distance, the right obstacle distance, and the left obstacle distance at the current moment into the trained fuzzy neural network to obtain the optimized turning angle at the current moment. Input the obtained optimized turning angle at the current moment into the adaptive PID controller to generate a control signal to adjust the angles and torques of the robot joints, which can achieve the navigation control of the humanoid robot.

Images