A gait control method for a quadruped robot and a quadruped robot thereof

Abstract

The present invention relates to the field of robotics, and more particularly to a quadruped robot gait control method and the quadruped robot thereof. The quadruped robot gait control method comprises the following steps: S1: acquiring source domain and target domain data, performing fuzzy processing on the source domain data using a fuzzy controller, and training an error back propagation neural network model using the fuzzy processed source domain data and target domain data; S2: deploying the optimized error back propagation fuzzy neural network model to a simulation model of the quadruped robot for verification; S3: deploying the optimized error back propagation neural network model to an algorithm platform for verification; and S4: deploying the optimized error back propagation neural network model to an algorithm verification machine for verification. The error back propagation neural network model is trained using the fuzzy processed source domain data and target domain data to obtain an optimized error back propagation neural network model, so that the gait parameter adjustment function in the virtual model controller has self-learning and self-adaptive capabilities.

Description

Technical Field

[0001] The present invention relates to the field of robots, and in particular to a gait control method of a quadruped robot and the quadruped robot. Background Art

[0002] In related technologies, the deployment of quadruped robot control systems often requires researchers to manually fine-tune gait and parameter design to adapt to the robot's dynamic characteristics, gradually converging the parameters to a reasonable range. This requires a high level of theoretical knowledge in dynamics and control science, is time-consuming, and requires extensive field training. Existing control methods typically require repeated attempts at parameter tuning, which hinders their widespread deployment. Summary of the Invention

[0003] Based on this, it is necessary to provide a quadruped robot gait control method to address the problems of manual participation in parameter adjustment and tedious parameter adjustment, including the following steps: S1: obtaining source domain data and using a fuzzy controller to fuzzy process the source domain data; establishing a simulation model for the quadruped robot to obtain target domain data, and using the fuzzy processed source domain data and target domain data to train the error back propagation neural network model to obtain an optimized error back propagation neural network model; S2: deploying the optimized error back propagation fuzzy neural network model to the simulation model of the quadruped robot, adjusting the parameters of the virtual model controller, and using the attitude angle measured by the virtual GPS speed measurement node and the inertial unit as a quantitative indicator to observe the root of the error back propagation fuzzy neural network model. The accuracy of the parameter adjustment decisions made according to the target domain data is observed; S3: the optimized error back propagation neural network model is deployed to the algorithm platform to verify the prototype, the parameters of the virtual model controller are adjusted, and the attitude angles measured by the virtual GPS speed measurement nodes and the inertial unit are used as quantitative indicators to observe the accuracy of the parameter adjustment decisions made by the error back propagation fuzzy neural network according to the target domain data; S4: an algorithm verification machine is built, and the optimized error back propagation neural network model is deployed to the algorithm verification machine to verify the algorithm verification machine, the parameters of the virtual model controller are adjusted, and the attitude angles measured by the virtual GPS speed measurement nodes and the inertial unit are used as quantitative indicators to observe the accuracy of the parameter adjustment decisions made by the fuzzy neural network according to the target domain data.

[0004] As a further improvement of the present invention, the step of obtaining target domain data includes: S11: building a dynamic model of a quadruped robot in the Webots simulation software; S12: debugging the parameters of the virtual model controller of the quadruped robot in the Webots simulation software to obtain posture feedback data of the quadruped robot dynamic model; S13: storing the posture feedback data of the quadruped robot dynamic model obtained above into the Matlab workspace, and retaining the correct adjustment operation based on the gait performance before and after the adjustment to obtain an input-decision output data set.

[0005] As a further improvement of the present invention, debugging the parameters of the virtual model controller includes: debugging the stiffness of the virtual model controller, debugging the initial height of the legs, debugging the swing phase amplitude of the legs, debugging the flight phase amplitude of the legs, and debugging the step frequency of the legs; the posture feedback data of the quadruped robot dynamic model includes: motor torque feedback data based on the current loop, the body posture of the quadruped robot, the body angular velocity of the quadruped robot, and the three-axis acceleration of the quadruped robot.

[0006] As a further improvement of the present invention, the error back propagation neural network includes: an input layer for preprocessing the input target domain data; an output layer for outputting the control parameters of the virtual model controller; and two hidden layers for establishing a characteristic relationship between the input and the output.

[0007] As a further improvement of the present invention, the preprocessing method adopted in the input layer is setting a bias.

[0008] As a further improvement of the present invention, output cutoff and anti-saturation measures are set at the output layer.

[0009] As a further improvement of the present invention, the accuracy of the parameter adjustment decision is based on calculating the unit transportation cost CoT of the robot movement before and after the decision. The expression of the unit transportation cost CoT is as follows:

[0010]

[0011] Among them, P inpu is the power consumption; mgv is the actual effective power; mg is the weight of the quadruped robot; v is the speed of the quadruped robot.

[0012] As a further improvement of the present invention, if the unit transportation cost after the decision is greater than the unit transportation cost before the decision, the decision is correct; if the unit transportation cost after the decision is less than the unit transportation cost before the decision, the correct parameter adjustment decision data is saved, and the saved parameter adjustment decision data is used to train the error back propagation fuzzy neural network model until the unit transportation cost after the decision is greater than the unit transportation cost before the decision.

[0013] The present invention also proposes a quadruped robot, which is applied to the above-mentioned quadruped robot gait control method. The quadruped robot gait control method is used to control the gait parameters of the quadruped robot to adjust the movement posture of the quadruped robot, including: a fuselage; four legs, the four legs are symmetrically arranged on both sides of the fuselage to drive the movement of the fuselage, and the structure of the legs is a parallel opposed five-link structure.

[0014] As a further improvement of the present invention, a foot end is provided at the end of the leg that contacts the ground, and the cross-section of the foot end is a semicircular structure.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] 1. The present invention obtains source domain data through sensors and uses a fuzzy controller to perform fuzzy processing on the source domain data; establishes a simulation model for a quadruped robot to obtain target domain data, and uses the fuzzy-processed source domain data and target domain data to train an error back propagation neural network model to obtain an optimized error back propagation neural network model, so that the gait parameter adjustment function in the virtual model controller has the ability of self-learning and self-adaptation, which can be used for training multiple gaits and multiple movements of quadruped robots with various leg and foot configurations.

[0017] 2. The legs of the robot in the present invention adopt a parallel opposed design. The parallel opposed design can increase the load capacity of the robot and expand the range of motion of the legs. In this way, the quadruped robot can perform somersaults and jumps by adjusting the relative position between the legs and the body, thereby achieving self-rescue. Moreover, the semicircular shape of the foot end can ensure good contact between the quadruped robot and the ground in different postures, especially good contact with the ground when jumping, so that the foot end is not easily damaged, thereby extending the service life of the robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a quadruped robot control method according to an embodiment of the present invention;

[0019] Figure 2 The figure is a schematic structural diagram of a quadruped robot according to an embodiment of the present invention.

[0020] Description of main component symbols

[0021] 1. Body; 2. Legs; 21. First rod group; 22. Second rod group; 23. Motor; 24. First connecting rod; 25. Second connecting rod; 26. Foot end.

[0022] The above description of the main component symbols is combined with the accompanying drawings and specific embodiments to further illustrate the present invention in detail. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component. When a component is considered to be "fixed to" another component, it may be directly fixed to the other component or there may be a central component.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0026] like Figure 2 As shown in FIG: An embodiment of the present invention provides a quadruped robot, which includes a body 1 and four legs 2. It should be noted that two motors 23 are provided on each leg.

[0027] The four legs 2 are symmetrically arranged on either side of the body 1 to drive the body 1. Each leg 2 is constructed as a parallel-opposed five-bar linkage. Designing the legs 2 as a parallel-opposed five-bar linkage allows for more flexible movement, as each five-bar linkage has two degrees of freedom.

[0028] As a further improvement of the present invention, a foot end 26 is provided at the end of the leg 2 that contacts the ground. Foot end 26 has a semicircular cross-section. Compared to existing flat structures, this semicircular foot end 26 ensures good ground contact between the quadruped robot and the ground in various postures, particularly during jumping. Furthermore, the semicircular foot end 26 is less susceptible to damage when traversing rough surfaces due to its small, smooth contact surface, thereby extending the lifespan of the quadruped robot.

[0029] In addition, the legs 2 adopt a parallel opposed design, which can increase the load capacity of the robot and expand the range of motion of the legs 2. In this way, the quadruped robot can perform somersaults and jumps by adjusting the relative position between the legs 2 and the body 1, thereby achieving self-rescue.

[0030] The parallel, opposed design divides the leg 2 into a first rod group 21 and a second rod group 22. The first and second rod groups 21, 22 are each controlled by a separate motor 23. The movements of the first and second rod groups 21, 22 are simultaneously input to the foot end 26, resulting in a combined output motion. This allows both the first and second rod groups 21, 22 to bear the load, thereby increasing the robot's load capacity.

[0031] Furthermore, both the first and second rod groups 21, 22 include a first connecting rod 24, 25. The motor 23 drives the first connecting rod 24 via a flange. Thin-walled bearings are installed on the outer sides of these flanges to enhance the structural strength of the first and second rod groups 21, 22. The first and second connecting rods 24, 25 are connected via flange bearings, thrust ball bearings, plug bolts, and gaskets. This not only reduces the resistance of the revolute joint but also ensures that the legs 2 are less susceptible to damage during prolonged movement. Furthermore, the fuselage 1 is primarily constructed of 18mm diameter carbon tubes and aluminum alloy, ensuring bending and torsional strength while reducing production costs and vehicle weight.

[0032] like Figure 1 As shown, the present invention also proposes a quadruped robot gait control method for controlling the gait parameters of the quadruped robot to adjust the motion posture of the quadruped robot, which includes the following steps:

[0033] S1: Acquire source domain data and perform fuzzy processing on the source domain data using a fuzzy controller; wherein the source domain data may be the root mean square of the horizontal force of the quadruped robot within several cycles, the mean square error of the horizontal force of the quadruped robot within several cycles, the root mean square of the vertical force of the quadruped robot within several cycles, the mean square error of the vertical force of the quadruped robot within several cycles, the root mean square of the pitch attitude angle of the quadruped robot within several gait cycles, the square root of the horizontal acceleration of the quadruped robot within several cycles, and the square root of the vertical acceleration of the quadruped robot within several cycles.

[0034] The significance of extracting this data lies in the fact that the dynamic characteristics reflected by this data can be used to infer the state of the control system through the cyber-physical system. The gait control method proposed in this paper is mainly aimed at adjusting the gait parameters of a quadruped robot moving rapidly on relatively flat ground. Therefore, during control, it is hoped that the vertical impact of the quadruped robot is as small as possible. For example, the output force of the foot on the ground is only expected to bear the weight of the robot body, and its variation is as small as possible. Therefore, measurement data such as the root mean square of the vertical force of the quadruped robot over several cycles, the mean square error of the vertical force of the quadruped robot over several cycles, and the square root of the vertical acceleration of the quadruped robot over several cycles are introduced.

[0035] The specific steps of the fuzzy controller for processing source domain data are as follows: based on expert experience, the fuzzy controller is first used to fuzzify the source domain data features extracted by the sensor, so that its features are reasonably identified as fuzzy languages ​​such as "too large", "too small" and "reasonable", and then these languages ​​are converted into discrete values ​​as input to the error back propagation neural network model.

[0036] Build a simulation model for the quadruped robot to obtain target domain data. The specific steps are:

[0037] S11: Build a dynamic model of a quadruped robot in the Webots simulation software; and design a control system based on a virtual force model.

[0038] S12: Debug the parameters of the virtual model controller of the quadruped robot in the Webots simulation software to obtain the posture feedback data of the quadruped robot dynamic model.

[0039] This debugging process is a manual debugging process, which mainly debugs the gait parameters of the quadruped robot in the virtual model controller. The debugging gait parameters include: debugging the stiffness KP of the virtual model controller, debugging the initial height of the legs, debugging the swing phase amplitude of the legs, debugging the flight phase amplitude of the legs, and debugging the step frequency of the legs.

[0040] The posture feedback data includes: the torque feedback data of the motor 23 based on the current loop, the body posture of the quadruped robot, the body angular velocity of the quadruped robot, and the three-axis acceleration of the quadruped robot.

[0041] S13: The obtained posture feedback data of the quadruped robot dynamics model is stored in the MATLAB workspace for subsequent training of the error back-propagation fuzzy neural network model. Based on the gait performance before and after adjustment, the correct adjustment operation is retained to obtain an input-decision-output dataset. The obtained input-decision-output dataset is the data of the target domain. This process is mainly performed by humans through continuous debugging, and the corresponding decision is obtained based on the number of debugging times and experience.

[0042] Then, the fuzzy processed source domain data and target domain data are used to train the error back propagation neural network model to obtain an optimized error back propagation neural network model.

[0043] The error back-propagation neural network model consists of an input layer, two hidden layers, and an output layer. The input layer preprocesses the input source and target domain data, facilitating subsequent training of the error back-propagation neural network model. The two hidden layers establish the relationship between input and output. The output layer outputs the control parameters of the virtual model controller: the virtual model controller's stiffness, the initial leg height, the leg swing phase amplitude, the leg flight phase amplitude, and the cadence of the debugged leg. This is equivalent to the error back-propagation fuzzy neural network model appropriately adjusting these five parameters based on the input.

[0044] As an optional embodiment, the preprocessing method can be to set a bias. For example, the amplitude of the plantar force is generally large. If the plantar force data and the pitch angle data of the quadruped robot are processed in the same way to train the error back propagation neural network model, the training effect will be affected. In other words, different processing methods are used for different input data.

[0045] As an optional embodiment, output cutoff and anti-saturation measures are set at the output layer. Because the error back propagation neural network model may output some data with large deviations during output, resulting in overfitting and other phenomena, setting output cutoff and anti-saturation measures can limit the output to a range, thereby shielding data with large deviations.

[0046] S2: The optimized error back-propagation fuzzy neural network model is deployed in a simulation model of a quadruped robot after randomly generating gait parameters. The parameters of the virtual model controller are adjusted, and the attitude angles measured by the virtual GPS speed measurement node and the inertial unit (IMU) are used as quantitative indicators to observe the accuracy of the parameter adjustment decisions made by the error back-propagation fuzzy neural network model based on the target domain data.

[0047] Specifically, the accuracy of the parameter adjustment decision is based on calculating the unit transportation cost CoT of the robot's movement before and after the decision. The expression of the unit transportation cost CoT is as follows:

[0048]

[0049] Among them, P input is the power consumption; mgv is the actual effective power; mg is the weight of the quadruped robot; v is the speed of the quadruped robot.

[0050] If the unit transportation cost after the decision is greater than the unit transportation cost before the decision, the decision is correct;

[0051] If the unit transportation cost after the decision is less than the unit transportation cost before the decision, the correct parameter adjustment decision data is saved, and the saved parameter adjustment decision data is spliced ​​into the source domain data. The error back propagation fuzzy neural network model is trained again until the unit transportation cost after the decision is greater than the unit transportation cost before the decision, which meets the accuracy standard.

[0052] S3: Deploy the optimized error back propagation neural network model to the algorithm platform to verify the prototype and adjust the parameters of the virtual model controller. Use the attitude angles measured by the virtual GPS speed measurement node and inertial unit (IMU) as quantitative indicators to observe the accuracy of the parameter adjustment decisions made by the error back propagation fuzzy neural network based on the target domain data.

[0053] Specifically, the accuracy of the parameter adjustment decision is based on calculating the unit transportation cost CoT of the robot's movement before and after the decision. The expression of the unit transportation cost CoT is as follows:

[0054]

[0055] Among them, P input is the power consumption; mgv is the actual effective power; mg is the weight of the quadruped robot; v is the speed of the quadruped robot.

[0056] If the unit transportation cost after the decision is greater than the unit transportation cost before the decision, the decision is correct;

[0057] If the unit transportation cost after the decision is less than the unit transportation cost before the decision, the correct parameter adjustment decision data is saved, and the saved parameter adjustment decision data is spliced ​​into the source domain data. The error back propagation fuzzy neural network model is trained again until the unit transportation cost after the decision is greater than the unit transportation cost before the decision, which meets the accuracy standard.

[0058] It should be noted that since the speed in the real world is not easy to obtain, the CoT is calculated here by measuring the average speed of the quadruped robot traveling a certain distance.

[0059] S4: Build an algorithm verification machine and deploy the optimized error back propagation neural network model to the algorithm verification machine to verify the algorithm verification machine, adjust the parameters of the virtual model controller, and use the attitude angle measured by the virtual GPS speed measurement node and inertial unit as quantitative indicators to observe the accuracy of the parameter adjustment decision made by the fuzzy neural network based on the target domain data.

[0060] Specifically, the accuracy of the parameter adjustment decision is based on calculating the unit transportation cost CoT of the robot's movement before and after the decision. The expression of the unit transportation cost CoT is as follows:

[0061]

[0062] Among them, P input is the power consumption; mgv is the actual effective power; mg is the weight of the quadruped robot; v is the speed of the quadruped robot.

[0063] If the unit transportation cost after the decision is greater than the unit transportation cost before the decision, the decision is correct;

[0064] If the unit transportation cost after the decision is less than the unit transportation cost before the decision, the correct parameter adjustment decision data is saved and spliced ​​into the source domain data. The error back propagation fuzzy neural network model is trained again until the unit transportation cost after the decision is greater than the unit transportation cost before the decision, which meets the accuracy standard. This will result in the error back propagation neural network model we need.

[0065] It should be noted that since real-world speed is difficult to obtain, the CoT is calculated here by measuring the average speed of the quadruped robot over a certain distance. Furthermore, to ensure the universality and versatility of the gait control method, motors 23 with different parameters can be used, and the algorithm verification machine can be built to a specific size. The operating characteristic parameters of motor 23 mainly include torque constant and rated speed.

[0066] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A robot gait control method, characterized in that: The process includes the following steps: S1: Acquire source domain data and perform fuzzy processing on the source domain data using a fuzzy controller; A simulation model of the quadruped robot is established to obtain target domain data. The fuzzy processed source domain data and target domain data are used to train the error back propagation neural network model to obtain an optimized error back propagation neural network model. S2: Deploy the optimized error back propagation fuzzy neural network model to a simulation model of a quadruped robot. Tune the parameters of the virtual model controller, using the attitude angles measured by the virtual GPS speed measurement node and the inertial unit as quantitative indicators, and observe the accuracy of the parameter adjustment decisions made by the error back propagation fuzzy neural network model based on the target domain data. S3: Build an algorithm platform and deploy the optimized error back propagation neural network model to the algorithm platform to verify the prototype. Tune the parameters of the virtual model controller and use the attitude angles measured by the virtual GPS speed measurement node and inertial unit as quantitative indicators to observe the accuracy of the parameter adjustment decisions made by the error back propagation fuzzy neural network based on the target domain data. S4: Build an algorithm verification machine and deploy the optimized error back propagation neural network model into the algorithm verification machine to verify the algorithm verification machine, adjust the parameters of the virtual model controller, and use the attitude angle measured by the virtual GPS speed measurement node and the inertial unit as quantitative indicators to observe the accuracy of the parameter adjustment decision made by the fuzzy neural network based on the target domain data.

2. A robot gait control method according to claim 1, characterized in that: The steps to obtain target domain data include: S11: Build a dynamic model of a quadruped robot in the Webots simulation software; S12: Debug the parameters of the virtual model controller of the quadruped robot in the Webots simulation software to obtain the posture feedback data of the quadruped robot dynamic model; S13: The posture feedback data of the quadruped robot dynamics model obtained above is stored in the MATLAB workspace, and the correct adjustment operation is retained according to the gait performance before and after adjustment to obtain an input-decision output data set.

3. A robot gait control method according to claim 2, characterized in that: Debugging the parameters of the virtual model controller includes: debugging the stiffness of the virtual model controller, debugging the initial height of the legs, debugging the swing phase amplitude of the legs, debugging the flight phase amplitude of the legs, and debugging the step frequency of the legs; The posture feedback data of the quadruped robot dynamics model include: motor torque feedback data based on the current loop, the body posture of the quadruped robot, the body angular velocity of the quadruped robot, and the three-axis acceleration of the quadruped robot.

4. A robot gait control method according to claim 1, characterized in that: Error back propagation neural network includes: Input layer, used to preprocess the input target domain data; The output layer is used to output the control parameters of the virtual model controller; Two hidden layers are used to establish the feature relationship between input and output.

5. A robot gait control method according to claim 4, characterized in that: The preprocessing method used in the input layer is to set the bias.

6. A robot gait control method according to claim 5, characterized in that: Set output cutoff and anti-saturation measures at the output layer.

7. A robot gait control method according to claim 1, characterized in that: The accuracy of the parameter adjustment decision is based on calculating the unit transportation cost CoT of the robot movement before and after the decision. The expression of the unit transportation cost CoT is as follows: Among them, P input is the power consumption; mgv is the actual effective power; mg is the weight of the quadruped robot; v is the speed of the quadruped robot.

8. A quadruped robot gait control method according to claim 7, characterized in that: If the unit transportation cost after the decision is greater than the unit transportation cost before the decision, the decision is correct; If the unit transportation cost after the decision is less than the unit transportation cost before the decision, the correct parameter adjustment decision data is saved, and the error back propagation fuzzy neural network model is trained with the saved parameter adjustment decision data until the unit transportation cost after the decision is greater than the unit transportation cost before the decision.

9. A quadruped robot, applied to a quadruped robot gait control method according to any one of claims 1 to 8, the quadruped robot gait control method being used to control gait parameters of the quadruped robot to adjust the motion posture of the quadruped robot, characterized in that: include: fuselage (1); Four legs (2) are symmetrically arranged on both sides of the fuselage (1) to drive the fuselage (1) to move, and the structure of the legs (2) is a parallel opposed five-link structure.

10. The quadruped robot according to claim 9, characterized in that: A foot end (26) is provided at one end of the leg (2) that contacts the ground, and the cross section of the foot end (26) is a semicircular structure.

Images