Height-controllable robot continuous jumping control method

Abstract

The invention discloses a height-controllable robot continuous jumping control method, and relates to the technical field of robot motion control. According to the height-controllable robot continuous jumping control method, the robot is simulated through the mass point spring model in the virtual control model, the single jumping period is divided in detail, continuous jumping control is facilitated, meanwhile, the motion state of the robot is conveniently judged through analysis of the foot end acting force and the foot end speed, and the robot continuous jumping control accuracy is improved. And the jump height is estimated by analyzing the jump time, the input variable of the mass spring model is controlled by matching with the establishment of a mapping function and combining with the calculation of the expected jump height, that is, the height of the mass spring model in the continuous jump state is controlled by adjusting the spring rigidity, and the control precision of the mass spring model in the continuous jump state is improved. And accurate control is provided for the jumping height of the robot in a continuous adjustment state.

Description

Technical Field

[0001] The invention relates to the technical field of robot motion control, in particular to a highly controllable robot continuous jumping control method. Background Art

[0002] Current robots can achieve single or multiple jumps through explosive movements. There are two main methods for continuous jumping of robots. One is based on mechanical mechanisms, which use energy storage devices such as springs to absorb the energy generated during the jump and landing process, thereby providing enough energy for the next jump. This type of method does not involve the control method of continuous jumping; the other is based on control methods to achieve continuous jumping. The main arrangement is to carry out detailed control of the take-off, flight, landing and other stages of the jump process. The main purpose is to ensure that there is enough energy to achieve a second jump after the jump and landing. In general, the main purpose of the current method is to enable the robot to It can provide enough energy to achieve multiple consecutive jumps, such as a continuous dynamic stable jumping control method for a humanoid robot described in application number 202110083373.X, which obtains the driving torque of each joint of the lower limbs of the humanoid robot according to the whole-body dynamics equation of the humanoid robot, applies the driving torque to the lower limbs of the humanoid robot, and the humanoid robot enters the take-off stage. When the upper body position of the humanoid robot reaches the take-off height, it enters the air stage, and controls each joint of the humanoid robot to reach the desired posture when landing. When the humanoid robot falls from the air and contacts the ground, it enters the landing stage, completes a jumping cycle, and returns to the initial state of the robot jumping.

[0003] Another example is a double-joint driven bionic self-balancing continuous jumping robot with application number 202410356504.0, which includes a support plate, a transmission unit and a double-joint jumping unit. It utilizes the periodic engagement and disconnection between the toothless gear and the pinion, and controls the energy accumulation and release of the double-joint jumping unit by pulling a rope to achieve continuous jumping of the jumping robot.

[0004] Based on the search of the above information, it can be seen that the existing robot continuous jumping control has no method to achieve accurate control of the jumping height during continuous jumping due to the instability of the jumping process, especially the changes in the robot's state during take-off and landing during multiple jumps. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a highly controllable robot continuous jumping control method, which solves the problem that the height cannot be accurately controlled in the existing robot continuous jumping.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a highly controllable robot continuous jumping control method, specifically comprising the following steps:

[0007] S1. Build a virtual control model, simulate the robot with a mass-spring model, and divide a jumping cycle of the robot into a continuous flight phase, a landing phase, and a take-off phase;

[0008] S2. Establishing a state machine in the virtual control model, monitoring the ground contact of the mass-spring model, and using the state machine to determine the motion state of the mass-spring model according to the ground contact monitoring result;

[0009] S3, according to the result of motion state judgment, controlling the mass spring model to perform continuous jumping by changing the virtual stiffness;

[0010] S4, record the flight phase time and calculate the highest point height of the mass spring model;

[0011] S5. Introduce the pseudo control law to calculate the expected jump height and perform continuous jump height control of the particle-spring model.

[0012] The present invention is further configured as follows: the flight phase is a stage in which the force at the foot end of the particle spring model is equal to zero, and the foot end speed is greater than zero to the foot end speed is less than zero;

[0013] The landing phase is a stage from when the force on the foot end of the particle spring model is zero and the speed on the foot end is less than zero to when the force on the foot end is greater than zero and the speed on the foot end is zero;

[0014] The take-off phase is a stage from when the force at the foot end of the particle spring model is greater than zero and the foot end velocity is equal to zero to when the force at the foot end is equal to zero and the foot end velocity is greater than zero.

[0015] The present invention is further configured as follows: the motion state in S2 includes a take-off state and a buffer state;

[0016] The take-off state includes a take-off phase. When the motion state is determined to be the take-off state in S3, before the mass spring model takes off from the ground, the stiffness of the virtual spring in the mass spring model increases and the original length increases, and the mass spring model is pushed off the ground by the foot end force.

[0017] The buffer state includes a flying phase and a landing phase. When the motion state is determined to be the buffer state in S3, when the mass-spring model falls and touches the ground, the stiffness of the virtual spring in the mass-spring model decreases and the original length becomes shorter.

[0018] The present invention is further configured as follows: the ground contact detection in S2 includes:

[0019] The robot's motor encoder is used to measure the foot end speed. When the foot end speed is less than zero, the robot is judged to be in the landing phase. When the foot end speed is equal to zero, the robot is judged to be at the lowest point, which is the trigger point of the take-off phase. When the foot end speed is greater than zero, the robot is judged to be in the take-off phase.

[0020] The present invention is further configured as follows: the method for calculating the height of the highest point of the mass-spring model in S4 includes:

[0021]

[0022] Where ΔT is the time difference from this jump to the next landing, Δh LT h is the height of the mass center of the mass spring model at this jump LO The height h of the center of mass of the mass-spring model when it touches the ground next time TD The height difference, g′ is the acceleration due to gravity, h smt is the highest point height of the mass-spring model.

[0023] The present invention is further configured as follows: the method of introducing the pseudo control law in S5 to calculate the expected jump height includes:

[0024]

[0025] k vl =S(h d ,h B )

[0026] In the formula, is the estimated value of the next jump height, h B is the lowest point height of the last landing, k vl is the virtual stiffness used for this take-off, f is the mapping function, S is the solution of the mapping function, h d is the expected jump height.

[0027] The present invention is further configured as follows: the method of performing continuous jump height control of the mass spring model in S5 includes:

[0028] According to the expected jump height, the virtual stiffness used for this jump is obtained, and the virtual stiffness is used as an input variable and imported into the virtual control model to control the spring stiffness of the mass-spring model to achieve continuous jump height control of the mass-spring model.

[0029] The present invention is further configured as follows: the establishment of the mapping function includes:

[0030] The robot is simulated with similar parameters, the numerical solution of the system dynamics is calculated, the vertex jumping height under different input parameters is obtained, and the calculation results are fitted using machine learning methods to obtain the mapping function that best conforms to the actual relationship.

[0031] The present invention provides a highly controllable robot continuous jumping control method, which has the following beneficial effects:

[0032] The present invention simulates the robot with a particle-spring model in a virtual control model and divides the single jump cycle in detail. While facilitating continuous jump control, the robot's motion state is conveniently judged by analyzing the foot-end force and foot-end speed, and the jump height is estimated by analyzing the jump time. In conjunction with the construction of a mapping function and the calculation of the expected jump height, the control of the input variables of the particle-spring model is achieved, that is, the height of the particle-spring model in the continuous jumping state is controlled by adjusting the spring stiffness, thereby providing precise control of the robot's jumping height in a continuously adjustable state. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of three stages of single-cycle jump of the mass-spring model in the present invention;

[0034] Figure 2 It is a schematic diagram of the principle of the state machine in the present invention;

[0035] Figure 3 It is a schematic diagram of the touchdown monitoring logic in the present invention;

[0036] Figure 4 It is a schematic diagram of a single take-off and landing of the mass spring model in the present invention. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0038] See also Figure 1-4 The embodiment of the present invention provides the following technical solution: a highly controllable robot continuous jumping control method, specifically comprising the following steps:

[0039] S1. Build a virtual control model, simulate the robot with a mass spring model, and divide a jump cycle of the robot into a continuous flight phase, landing phase and take-off phase, as shown in the attached figure. Figure 1 As shown, where F GRF is the foot end force, V is the foot end velocity, and the flight phase is the stage from the point-spring model foot end force equal to zero, and the foot end velocity is greater than zero to the foot end velocity is less than zero;

[0040] The landing phase is a stage from when the force on the foot end of the particle spring model is zero and the speed on the foot end is less than zero to when the force on the foot end is greater than zero and the speed on the foot end is zero;

[0041] The take-off phase is a stage from when the force at the foot end of the particle spring model is greater than zero and the foot end velocity is equal to zero to when the force at the foot end is equal to zero and the foot end velocity is greater than zero.

[0042] S2. Establish a state machine in the virtual control model to monitor the ground contact of the mass spring model. The ground contact monitoring is as shown in the attached figure. Figure 3 As shown, including:

[0043] The robot's motor encoder is used to measure the foot end speed. When the foot end speed is less than zero, the robot is judged to be in the landing phase. When the foot end speed is equal to zero, the robot is judged to be at the lowest point, which is the trigger point of the take-off phase. When the foot end speed is greater than zero, the robot is judged to be in the take-off phase. Then, when the foot end speed is equal to zero, the robot is judged to be at the highest point.

[0044] The state machine is used to determine the motion state of the mass-spring model according to the ground contact monitoring result. The motion state includes a take-off state and a buffer state. The take-off state includes a take-off phase, and the buffer state includes a flight phase and a landing phase.

[0045] S3. According to the result of the motion state judgment, the particle spring model is controlled to perform continuous jumping by changing the virtual stiffness. Specifically, when the motion state is judged to be a take-off state, before the particle spring model takes off the ground, the stiffness of the virtual spring in the particle spring model increases and the original length increases, and the particle spring model is pushed off the ground by the force of the foot end. When the motion state is judged to be a buffer state, when the particle spring model falls and touches the ground, the stiffness of the virtual spring in the particle spring model decreases and the original length shortens.

[0046] When taking off, the virtual spring is compressed to the shortest, and the virtual spring stiffness parameters are changed instantly to achieve the take-off; after leaving the ground, it instantly switches to the buffer state as preparation for the next landing.

[0047] S4. Record the flight phase time and calculate the highest point height of the mass spring model. The simulation method is as shown in the attached Figure 4 As shown, the calculation method includes:

[0048]

[0049] Where ΔT is the time difference from this jump to the next landing, Δh LT h is the height of the mass center of the mass spring model at this jump LO The height h of the center of mass of the mass-spring model when it touches the ground next time TD The height difference, g′ is the acceleration due to gravity, h smt is the highest point height of the mass-spring model.

[0050] S5. The pseudo control law is introduced to calculate the expected jump height and to control the continuous jump height of the mass spring model. Specifically, the continuous jump process of the mass spring model is approximately a Markov process. The expected height of the next jump is only related to the lowest point height after the landing compression and the characteristics of the applied virtual spring, but has nothing to do with the previous jump process. Its height is:

[0051]

[0052] In the formula, is the estimated value of the next jump height, h B is the lowest point height of the last landing, k vl is the virtual stiffness used for this jump, f is the mapping function, that is, the function of the lowest point height of the last landing and the virtual stiffness used for this jump. At this time, the virtual damping is not controlled. The establishment of the mapping function includes: simulating the robot with similar parameters, calculating the numerical solution of the system dynamics, obtaining the vertex jump height under different input parameters, fitting the calculation results using machine learning methods, obtaining the mapping function that best conforms to the actual relationship, and then proposing a pseudo control law based on the mapping function:

[0053] k vl =S(h d ,h B )

[0054] In the formula, S is the solution of the mapping function, h d is the expected jump height;

[0055] According to the expected jump height, the virtual stiffness used for this jump is obtained, and the virtual stiffness is used as an input variable and imported into the virtual control model to control the spring stiffness of the mass-spring model to achieve continuous jump height control of the mass-spring model.

[0056] As a detailed description, the linear regression method is used to fit the binary mapping relationship between the jump height, the lowest point height and the virtual stiffness, and the model of the multiple regression problem is established as follows:

[0057] y=β0+β1x1+β2x2+…+β p x p +ε

[0058] Among them, y is the dependent variable, x i is the independent variable, β i is the weight coefficient, ε is the random error term;

[0059] Since there are multiple regression variables and multiple relationships in multiple regression, there is a concept of multicollinearity, which means that the explanatory variables in the linear regression model have precise or highly correlated relationships, which makes the model estimation distorted or inaccurate. When multicollinearity exists, the accuracy of model parameter estimation will be reduced, resulting in unreasonable parameter estimation and the model fitting cannot achieve good results. Ridge regression method and Lasso regression method are introduced on the basis of least squares, and a more refined and realistic regression model is established by losing part of the information, reducing the accuracy and constructing a penalty function to compress the regression coefficient.

[0060] Among them, the optimization function expression of ridge regression is:

[0061] mmin||Xβ-y|| 2 +z||β|| 2

[0062] Where z is a non-negative positive parameter, X refers to the independent variable vector;

[0063] Like ridge regression, Lasso regression also adds a bias term to the regression optimization function to reduce the impact of collinearity. However, the difference between the two is that the absolute value deviation is used as the regularization term in Lasso regression instead of the square term. The optimization function expression of Lasso regression is:

[0064] min||Xβ-y|| 2 +Z||β|| 2

[0065] Where Z is a non-negative regular parameter;

[0066] Lasso regression uses the L1 norm as a penalty term. One of its important properties is built-in feature selection, which is not available in the L2 norm of ridge regression. Since the L1 norm can achieve sparseness, that is, Lasso regression implements a parameter selection, the unselected feature variables have a weight of 0 for the whole, which makes the calculation less complex and more efficient.

[0067] Furthermore, elastic network regression combines Lasso regression and ridge regression techniques. It uses L1 and L2 regularization and obtains the common features of the two methods. The expression of elastic regression network is as follows:

[0068] min||Xβ-y|| 2 +z1||β||+z2||β|| 2

[0069] Among them, z1 and z2 are both non-negative regular parameters;

[0070] Elastic net regression supports group effects in the case of highly correlated variables, rather than setting some parameters to 0 like Lasso regression. This method is very useful when multiple features are correlated with another feature.

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A highly controllable robot continuous jumping control method, characterized in that: The specific steps include: S1. Build a virtual control model, simulate the robot with a mass-spring model, and divide a jumping cycle of the robot into a continuous flight phase, a landing phase, and a take-off phase; S2. Establishing a state machine in the virtual control model, monitoring the ground contact of the mass-spring model, and using the state machine to determine the motion state of the mass-spring model according to the ground contact monitoring result; S3, according to the result of motion state judgment, controlling the mass spring model to perform continuous jumping by changing the virtual stiffness; S4, record the flight phase time and calculate the highest point height of the mass spring model; S5. Introduce the pseudo control law to calculate the expected jump height and perform continuous jump height control of the particle-spring model.

2. A highly controllable robot continuous jumping control method according to claim 1, characterized in that: The flight phase is a stage in which the force acting on the foot end of the particle spring model is zero, and the foot end velocity is greater than zero to less than zero; The landing phase is a stage from when the force on the foot end of the particle spring model is zero and the speed on the foot end is less than zero to when the force on the foot end is greater than zero and the speed on the foot end is zero; The take-off phase is a stage from when the force at the foot end of the particle spring model is greater than zero and the foot end velocity is equal to zero to when the force at the foot end is equal to zero and the foot end velocity is greater than zero.

3. The method for controlling continuous jumping of a highly controllable robot according to claim 1, characterized in that: The motion state in S2 includes a take-off state and a buffer state; The take-off state includes a take-off phase. When the motion state is determined to be the take-off state in S3, before the mass spring model takes off from the ground, the stiffness of the virtual spring in the mass spring model increases and the original length increases, and the mass spring model is pushed off the ground by the foot end force. The buffer state includes a flying phase and a landing phase. When the motion state is determined to be the buffer state in S3, when the mass-spring model falls and touches the ground, the stiffness of the virtual spring in the mass-spring model decreases and the original length becomes shorter.

4. The method for controlling continuous jumping of a highly controllable robot according to claim 1, characterized in that: The touchdown detection in S2 includes: The robot's motor encoder is used to measure the foot end speed. When the foot end speed is less than zero, the robot is judged to be in the landing phase. When the foot end speed is equal to zero, the robot is judged to be at the lowest point, which is the trigger point of the take-off phase. When the foot end speed is greater than zero, the robot is judged to be in the take-off phase.

5. The method for controlling continuous jumping of a highly controllable robot according to claim 1, characterized in that: The method for calculating the highest point height of the mass-spring model in S4 includes: Where ΔT is the time difference from this jump to the next landing, Δh LT h is the height of the mass center of the mass spring model at this jump LO The height h of the center of mass of the mass-spring model when it touches the ground next time TD The height difference, g′ is the acceleration due to gravity, h smt is the highest point height of the mass-spring model.

6. A highly controllable robot continuous jumping control method according to claim 5, characterized in that: The method of introducing the pseudo control law in S5 to calculate the expected jump height includes: k vl =S(h d ,h B ) In the formula, is the estimated value of the next jump height, h B is the lowest point height of the last landing, k vl is the virtual stiffness used for this take-off, f is the mapping function, S is the solution of the mapping function, h d is the expected jump height.

7. A highly controllable robot continuous jumping control method according to claim 6, characterized in that: The method of performing continuous jump height control of the mass spring model in S5 includes: According to the expected jump height, the virtual stiffness used for this jump is obtained, and the virtual stiffness is used as an input variable and imported into the virtual control model to control the spring stiffness of the mass-spring model to achieve continuous jump height control of the mass-spring model.

8. The highly controllable robot continuous jumping control method according to claim 1, characterized in that: The establishment of the mapping function includes: The robot is simulated with similar parameters, the numerical solution of the system dynamics is calculated, the vertex jumping height under different input parameters is obtained, and the calculation results are fitted using machine learning methods to obtain the mapping function that best conforms to the actual relationship.

Images