A control method and system of a wheel-legged balancing robot

Abstract

A control method of a wheel-legged balancing robot, comprising the following steps: acquiring pose information of the robot when moving, extracting attitude changes between adjacent time points, and taking the attitude changes as first features; inputting the first features into a pre-trained machine learning model to predict attitude change trends of the robot in the future, wherein the attitude change trends include natural attitude changes and large-amplitude attitude changes; judging whether the attitude change trends are large-amplitude attitude changes, and if so, adjusting a next-time advancing speed of the robot according to the predicted attitude change trends and a real-time advancing speed of the robot. Overall, the performance of the robot in a dynamic environment is optimized, the risk of instability is reduced, the reliability of task execution and user experience are improved, and a solid foundation is laid for the wide application of the wheel-legged balancing robot.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and in particular to a control method and system for a wheel-legged balancing robot. Background Technology

[0002] Wheel-legged balancing robots combine wheeled and legged drive structures, offering good mobility and stability. Their design typically includes four or more wheels and multiple adjustable legs for support. The wheels provide efficient linear motion and steering capabilities, while the legs help the robot maintain balance on uneven terrain. In terms of control, the key to wheel-legged balancing robots is coordinating the movements of the wheels and legs to maintain stability, especially during dynamic motion. Control systems typically acquire real-time information such as robot posture and velocity using sensors (e.g., gyroscopes, accelerometers, and force sensors), and use feedback control algorithms (e.g., PID control, LQR control) to adjust the coordination between the wheels and legs, ensuring the robot can adjust its posture promptly according to changes in the ground during movement, avoiding imbalance. Furthermore, advanced control strategies may incorporate machine learning algorithms to address uncertainties and dynamic changes in complex environments.

[0003] Wheeled-legged balancing robots typically move according to the user's speed commands, with the control system adjusting its movement by receiving instructions from handles, remote controls, or other human-machine interface devices. However, when the robot encounters significant changes in posture, current technology often fails to intelligently adjust its speed. If it continues to move at the user-set speed, the robot may experience violent fluctuations under conditions of large posture changes. If the speed is not adjusted in time, the robot may not be able to quickly regain its balance, leading to loss of control and falls. This could not only damage the robot itself but also cause damage to the surrounding environment and even harm people, especially in high-risk scenarios such as military reconnaissance or disaster relief, where the potential hazards of imbalance are even more severe. Summary of the Invention

[0004] To address the aforementioned shortcomings, the present invention aims to propose a control method and system for a wheel-legged balancing robot, which addresses the deficiencies of existing technologies in timely sensing and responding to posture changes and inability to adjust the robot's travel speed in a timely manner.

[0005] To achieve this objective, the present invention adopts the following technical solution: a control method for a wheel-legged balancing robot, comprising the following steps:

[0006] Step S1: Obtain the robot's pose information during movement, extract the pose changes between adjacent time moments, and use the pose changes as the first feature;

[0007] Step S2: Input the first feature into the pre-trained machine learning model to predict the future posture change trend of the robot, wherein the posture change trend includes natural posture change and large posture change.

[0008] Step S3: Determine whether the posture change trend is a large-scale posture change. If so, adjust the robot's next movement speed according to the predicted posture change trend and the robot's real-time movement speed.

[0009] Preferably, the attitude change in step S1 includes displacement data and angle data;

[0010] The displacement data refers to the rate of change of the robot's center of gravity in the horizontal and vertical directions.

[0011] The angle data refers to the rate of change of the robot's tilt angle over time.

[0012] Preferably, before executing step S2, the first feature needs to be processed, wherein the processing includes: strengthening the displacement data in a power form and strengthening the angle data by exponential operation;

[0013] The formula for enhancing the displacement data in power form is as follows:

[0014]

[0015] Where x i,i+1 y i,i+1 , respectively, represent the rate of displacement of the center of gravity in the horizontal and vertical directions, β is an exponential factor greater than 1, and i represents time;

[0016] The enhanced formula for exponential calculation of angle data is as follows:

[0017] S θi =exp(α·|θ) i | δ );

[0018] Where α and δ represent adjustment parameters, and θ i Rate of change of tilt angle.

[0019] Preferably, the step of predicting the future posture change trend of the robot in step S2 is as follows:

[0020] The first feature is input into a pre-trained machine learning model, which generates a pose change index.

[0021] Determine whether the attitude change index is greater than a preset attitude change index reference threshold. If the attitude change index is greater than the attitude change index reference threshold, the current attitude change is a large attitude change. If the attitude change index is less than or equal to the attitude change index reference threshold, the current attitude change is a natural attitude change.

[0022] Preferably, the specific steps of step S3 are as follows:

[0023] The evaluation coefficients are obtained based on the robot's current speed and attitude change index.

[0024] The robot's current speed is adjusted by evaluating the coefficients and the attitude change index to obtain the robot's speed at the next moment.

[0025] The formula for obtaining the evaluation coefficients is as follows:

[0026] Δθ speed-impact =g(V current Posture Change );

[0027] Where g() is a functional relation, V current The current speed of the robot, Posture Change The attitude change index;

[0028] The formula for obtaining the robot's next moving speed is as follows:

[0029] Where k′ is the proportional adjustment coefficient, Δθ ref This is the reference threshold for the attitude change index.

[0030] A control system for a wheeled-legged balancing robot, using the control method for the wheeled-legged balancing robot, includes a feature acquisition module, a judgment module, and an adjustment module;

[0031] The feature acquisition module is used to acquire the robot's pose information during movement, extract the pose changes between adjacent moments, and use the pose changes as the first feature.

[0032] The judgment module is used to input the first feature into a pre-trained machine learning model to predict the future posture change trend of the robot.

[0033] The adjustment module is used to determine whether the posture change trend is a large-scale posture change. If so, it adjusts the robot's next movement speed according to the predicted posture change trend and the robot's real-time movement speed.

[0034] Preferably, the feature acquisition module includes a first extraction submodule and a second extraction submodule;

[0035] The first extraction submodule is used to obtain the displacement rate of the robot's center of gravity in the horizontal and vertical directions;

[0036] The second extraction submodule is used to obtain the rate of change of the robot's tilt angle over time.

[0037] Preferably, it also includes an optimization module, which is used to enhance the displacement data in a power form and enhance the angle data by performing exponential operations.

[0038] Preferably, the judgment module includes an input submodule and a prediction submodule;

[0039] The input submodule is used to input the first feature into a pre-trained machine learning model, and the machine learning model generates a pose change index.

[0040] The prediction submodule is used to determine whether the attitude change index is greater than a preset attitude change index reference threshold. If the attitude change index is greater than the attitude change index reference threshold, the current attitude change is a large attitude change. If the attitude change index is less than or equal to the attitude change index reference threshold, the current attitude change is a natural attitude change.

[0041] Preferably, the adjustment module includes an evaluation submodule and an adjustment submodule;

[0042] The estimation module is used to obtain evaluation coefficients based on the robot's current travel speed and attitude change index;

[0043] The adjustment submodule is used to adjust the current travel speed of the robot by evaluating the coefficients and the attitude change index, so as to obtain the travel speed of the robot at the next moment.

[0044] One of the above technical solutions has the following advantages or beneficial effects: This invention acquires robot posture data in real time and extracts key features, then combines this with a pre-trained machine learning model to dynamically predict future posture change trends. This allows for the early identification of potential balance risks and the implementation of targeted strategies. When posture changes are within a natural range, the user's control intention is accurately preserved, ensuring the smoothness and stability of the robot's movement. When posture changes are significant, the robot intelligently adjusts its speed, giving it sufficient time and control margin to correct its posture, preventing loss of control and falls. This not only effectively solves the shortcomings of existing technologies in timely perception and response to posture changes but also significantly enhances the robot's adaptability in high-risk scenarios, such as military reconnaissance, disaster relief, or operations on uneven terrain. Overall, it optimizes the robot's performance in dynamic environments, reduces the risk of instability, improves the reliability of task execution and user experience, and lays a solid foundation for the widespread application of wheeled legged balancing robots. Attached Figure Description

[0045] Figure 1 This is a flowchart of one embodiment of the method of the present invention.

[0046] Figure 2 This is a schematic diagram of the structure of one embodiment of the system of the present invention. Detailed Implementation

[0047] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0048] In the description of embodiments of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0049] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0050] like Figures 1-2 As shown, a control method for a wheel-legged balancing robot includes the following steps:

[0051] Step S1: Obtain the robot's pose information during movement, extract the pose changes between adjacent time moments, and use the pose changes as the first feature;

[0052] Step S2: Input the first feature into the pre-trained machine learning model to predict the future posture change trend of the robot, wherein the posture change trend includes natural posture change and large posture change.

[0053] Step S3: Determine whether the posture change trend is a large-scale posture change. If so, adjust the robot's next movement speed according to the predicted posture change trend and the robot's real-time movement speed.

[0054] This invention acquires robot posture data in real time and extracts key features, then combines this with a pre-trained machine learning model to dynamically predict future posture change trends. This allows for the early identification of potential balance risks and the implementation of targeted strategies. When posture changes are within a natural range, the user's control intention is accurately preserved, ensuring the smoothness and stability of the robot's movement. However, when posture changes are significant, the robot intelligently adjusts its speed, giving it sufficient time and control margin to correct its posture, preventing loss of control and falls. This not only effectively addresses the shortcomings of existing technologies in timely perception and response to posture changes but also significantly enhances the robot's adaptability in high-risk scenarios, such as military reconnaissance, disaster relief, or operations on uneven terrain. Overall, it optimizes the robot's performance in dynamic environments, reduces the risk of instability, improves task execution reliability and user experience, and lays a solid foundation for the widespread application of wheeled legged balancing robots.

[0055] Preferably, the attitude change in step S1 includes displacement data and angle data;

[0056] The displacement data refers to the rate of change of the robot's center of gravity in the horizontal and vertical directions.

[0057] The angle data refers to the rate of change of the robot's tilt angle over time.

[0058] Displacement data, quantified by center of gravity offset quantization, measures the rate of change of the robot's center of gravity in the horizontal and vertical directions due to posture changes or external disturbances during movement, reflecting the robot's posture stability. Angle data, quantified by tilt rate quantization, measures the rate of change of the robot's tilt angle over time, reflecting the speed of tilt changes during movement and further indicating the robot's posture changes.

[0059] Preferably, before executing step S2, the first feature needs to be processed, wherein the processing includes: strengthening the displacement data in a power form and strengthening the angle data by exponential operation;

[0060] The formula for enhancing the displacement data in power form is as follows:

[0061]

[0062] Where x i,i+1 y i,i+1 , respectively, represent the rate of displacement of the center of gravity in the horizontal and vertical directions, β is an exponential factor greater than 1, and i represents time;

[0063] After obtaining the displacement rates in the horizontal and vertical directions, the displacement rates in the horizontal and vertical directions are combined to generate a comprehensive center of gravity offset quantization value, which reflects the displacement rate of the center of gravity in the horizontal and vertical directions caused by posture changes or external disturbances during the robot's movement. In order to further enhance the sensitivity to large displacement changes, a power-law form is used to strengthen the response to velocity changes.

[0064] The enhanced formula for exponential calculation of angle data is as follows:

[0065] S θi =exp(α·|θ) i | δ );

[0066] Where α and δ represent adjustment parameters, and θ i Rate of change of tilt angle.

[0067] After obtaining the rate of change of tilt angle between each time point, the rate of change of tilt angle is converted into a quantized value of tilt rate, and the sensitivity of the system to drastic attitude changes is enhanced. Specifically, this invention uses exponential operation to enhance the rate of change.

[0068] Preferably, the step of predicting the future posture change trend of the robot in step S2 is as follows:

[0069] The first feature is input into a pre-trained machine learning model, which generates a pose change index.

[0070] Determine whether the attitude change index is greater than a preset attitude change index reference threshold. If the attitude change index is greater than the attitude change index reference threshold, the current attitude change is a large attitude change. If the attitude change index is less than or equal to the attitude change index reference threshold, the current attitude change is a natural attitude change.

[0071] These pre-trained machine learning models are typically based on supervised learning methods such as regression analysis, support vector machines (SVM), random forests, or deep learning networks (e.g., LSTM networks). These algorithms can extract valuable information from input features, such as quantized values ​​of center of gravity offset (displacement data) and tilt rate (angle data), establishing a mapping relationship between input and output. After training, the model can not only predict the robot's instantaneous posture changes but also infer future posture changes based on current state trends and generate a posture change index (a probability measure of posture changes). This predictive model is built on previously known training data. Through a large amount of data in different scenarios, the model can learn the robot's behavioral patterns and responses in different states, thereby effectively assessing potential risks and helping the robot adjust its posture control strategy in real time. In practical applications, when input features change, the model can immediately reflect the potential risk level, take measures in advance to ensure the robot maintains a stable posture, and avoid instability or falls due to failure to detect risks in time.

[0072] The machine learning model is not specifically limited here; any machine learning model capable of comprehensively analyzing displacement data and angles to generate an attitude change index is acceptable. To achieve the technical solution of this invention, a specific implementation method is provided; the expression for generating the attitude change index is:

[0073]

[0074] Where v x v y These are preset proportional coefficients. As can be seen from the attitude change index, the larger the value of the center of gravity offset quantization value generated after analyzing the rate of change of the robot's center of gravity displacement in the horizontal and vertical directions between adjacent moments, the larger the value of the tilt rate quantization value generated after analyzing the rate of change of the robot's tilt angle over time between adjacent moments, that is, the larger the value of the attitude change index generated when the robot's future attitude change trend is predicted by the pre-trained machine learning model, the greater the attitude change amplitude of the robot between adjacent moments, and vice versa.

[0075] Preferably, the specific steps of step S3 are as follows:

[0076] The evaluation coefficients are obtained based on the robot's current speed and attitude change index.

[0077] The robot's current speed is adjusted by evaluating the coefficients and the attitude change index to obtain the robot's speed at the next moment.

[0078] The formula for obtaining the evaluation coefficients is as follows:

[0079] Δθ speed-impact =g(V current Posture Change );

[0080] Where g() is a functional relation, V current The current speed of the robot, Posture Change The attitude change index;

[0081] The function g(V) current Posture Change The function g(V) is used to assess the degree of influence of the current velocity on the attitude change. current Posture Change The specific expression for the function g() can be obtained through experimental data, physical models, and empirical analysis. First, a large amount of experimental data is collected to collect the robot's motion performance under different traveling speeds and attitude change exponents, analyzing the impact of traveling speed on attitude change. This data helps derive the form of the function g(), which is typically a functional relationship, possibly linear, nonlinear, or piecewise. Through deep learning or regression analysis techniques, the expression of this function can be further fitted to accurately describe the impact of traveling speed on attitude change under different conditions.

[0082] The formula for obtaining the robot's next moving speed is as follows:

[0083] Where k′ is the proportional adjustment coefficient, Δθ ref This is the reference threshold for the attitude change index.

[0084] When a robot's posture change is identified as a significant one, the system adjusts its current speed using evaluation coefficients and posture change indices. This step primarily ensures the robot maintains balance under conditions of drastic posture changes or unstable external environments, preventing loss of control or falls. Specifically, significant posture changes typically indicate that the robot is facing substantial dynamic disturbances or complex terrain. In such situations, excessive speed may exacerbate the risk of imbalance, while insufficient speed may reduce task efficiency. Through intelligent speed adjustment, the system can dynamically slow down or accelerate the robot's movement, giving it sufficient time to correct its posture and maintain stability in complex scenarios. This process not only improves the robot's safety but also enhances its adaptability to dynamic environments, ensuring the robot can smoothly complete its mission objectives.

[0085] A control system for a wheeled-legged balancing robot, using the control method for the wheeled-legged balancing robot, includes a feature acquisition module, a judgment module, and an adjustment module;

[0086] The feature acquisition module is used to acquire the robot's pose information during movement, extract the pose changes between adjacent moments, and use the pose changes as the first feature.

[0087] The judgment module is used to input the first feature into a pre-trained machine learning model to predict the future posture change trend of the robot.

[0088] The adjustment module is used to determine whether the posture change trend is a large-scale posture change. If so, it adjusts the robot's next movement speed according to the predicted posture change trend and the robot's real-time movement speed.

[0089] Preferably, the feature acquisition module includes a first extraction submodule and a second extraction submodule;

[0090] The first extraction submodule is used to obtain the displacement rate of the robot's center of gravity in the horizontal and vertical directions;

[0091] The second extraction submodule is used to obtain the rate of change of the robot's tilt angle over time.

[0092] Preferably, it also includes an optimization module, which is used to enhance the displacement data in a power form and enhance the angle data by performing exponential operations.

[0093] Preferably, the judgment module includes an input submodule and a prediction submodule;

[0094] The input submodule is used to input the first feature into a pre-trained machine learning model, and the machine learning model generates a pose change index.

[0095] The prediction submodule is used to determine whether the attitude change index is greater than a preset attitude change index reference threshold. If the attitude change index is greater than the attitude change index reference threshold, the current attitude change is a large attitude change. If the attitude change index is less than or equal to the attitude change index reference threshold, the current attitude change is a natural attitude change.

[0096] Preferably, the adjustment module includes an evaluation submodule and an adjustment submodule;

[0097] The estimation module is used to obtain evaluation coefficients based on the robot's current travel speed and attitude change index;

[0098] The adjustment submodule is used to adjust the current travel speed of the robot by evaluating the coefficients and the attitude change index, so as to obtain the travel speed of the robot at the next moment.

[0099] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0100] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A control method for a wheel-legged balancing robot, characterized in that, Includes the following steps: Step S1: Obtain the robot's pose information during movement, extract the pose changes between adjacent time moments, and use the pose changes as the first feature; Step S2: Input the first feature into the pre-trained machine learning model to predict the future posture change trend of the robot, wherein the posture change trend includes natural posture change and large posture change. Step S3: Determine whether the posture change trend is a large-scale posture change. If so, adjust the robot's next movement speed according to the predicted posture change trend and the robot's real-time movement speed. The attitude change in step S1 includes displacement data and angle data; The displacement data refers to the rate of change of the robot's center of gravity in the horizontal and vertical directions. The angle data is the rate of change of the robot's tilt angle over time; The steps in step S2 to predict the future posture change trend of the robot are as follows: The first feature is input into a pre-trained machine learning model, which generates a pose change index. Determine whether the attitude change index is greater than a preset attitude change index reference threshold. If the attitude change index is greater than the attitude change index reference threshold, the current attitude change is a large attitude change. If the attitude change index is less than or equal to the attitude change index reference threshold, the current attitude change is a natural attitude change. The specific steps of step S3 are as follows: The evaluation coefficients are obtained based on the robot's current speed and attitude change index. The robot's current speed is adjusted by evaluating the coefficients and the attitude change index to obtain the robot's speed at the next moment. The formula for obtaining the evaluation coefficients is as follows: ； Where g() is a functional relationship, This represents the robot's current speed. The attitude change index; The formula for obtaining the robot's next moving speed is as follows: ,in This is the proportional adjustment coefficient. This is the reference threshold for the attitude change index.

2. The control method for a wheel-legged balancing robot according to claim 1, characterized in that, Before executing step S2, the first feature needs to be processed, including: strengthening the displacement data in a power form and strengthening the angle data by exponential operation. The formula for enhancing the displacement data in power form is as follows: ； in These represent the rates of displacement of the center of gravity in the horizontal and vertical directions, respectively. Let i be an exponential factor greater than 1, and let i represent time. The enhanced formula for exponential calculation of angle data is as follows: ； in Indicates the adjustment parameter. Rate of change of tilt angle.

3. A control system for a wheel-legged balancing robot, characterized in that, The control method for a wheeled legged balancing robot according to any one of claims 1 to 2 includes a feature acquisition module, a judgment module, and an adjustment module; The feature acquisition module is used to acquire the robot's pose information during movement, extract the pose changes between adjacent moments, and use the pose changes as the first feature. The judgment module is used to input the first feature into a pre-trained machine learning model to predict the future posture change trend of the robot. The adjustment module is used to determine whether the posture change trend is a large-scale posture change. If so, it adjusts the robot's next movement speed according to the predicted posture change trend and the robot's real-time movement speed.

4. The control system for a wheel-legged balancing robot according to claim 3, characterized in that, The feature acquisition module includes a first extraction submodule and a second extraction submodule; The first extraction submodule is used to obtain the displacement rate of the robot's center of gravity in the horizontal and vertical directions; The second extraction submodule is used to obtain the rate of change of the robot's tilt angle over time.

5. The control system for a wheel-legged balancing robot according to claim 3, characterized in that, It also includes an optimization module, which is used to enhance the displacement data in a power form and enhance the angle data by performing exponential operations.

6. The control system for a wheel-legged balancing robot according to claim 3, characterized in that, The judgment module includes an input submodule and a prediction submodule; The input submodule is used to input the first feature into a pre-trained machine learning model, and the machine learning model generates a pose change index. The prediction submodule is used to determine whether the attitude change index is greater than a preset attitude change index reference threshold. If the attitude change index is greater than the attitude change index reference threshold, the current attitude change is a large attitude change. If the attitude change index is less than or equal to the attitude change index reference threshold, the current attitude change is a natural attitude change.

7. The control system for a wheel-legged balancing robot according to claim 3, characterized in that, The adjustment module includes an evaluation submodule and an adjustment submodule; The estimation module is used to obtain evaluation coefficients based on the robot's current travel speed and attitude change index; The adjustment submodule is used to adjust the current travel speed of the robot by evaluating the coefficients and the attitude change index, so as to obtain the travel speed of the robot at the next moment.

Images