A robot dynamic balance control method and system

Abstract

The embodiment of the present application relates to the technical field of machine control, and discloses a robot dynamic balance control method, comprising: obtaining current state information of the robot, wherein the current state information comprises body posture information and environment perception information; predicting expected disturbance that the robot will be subjected to based on the environment perception information, and generating a feedforward control instruction based on the expected disturbance; based on the current state information and the feedforward control instruction, performing rolling optimization on a unified dynamic model through a model predictive control algorithm, and solving joint motion instructions of a bionic spine unit and an active tail wing unit in a future time period to minimize a predefined cost function, wherein the cost function at least comprises a robot posture deviation term; and outputting the joint motion instructions to actuators of the bionic spine unit and the active tail wing unit to control their cooperative motion. The method of the embodiment of the present application can ensure the balance of the robot through rolling optimization and real-time adjustment, and improves the robustness of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine control, in particular to a robot dynamic balance control method and system. BACKGROUND

[0002] At present, with the development of robot technology, the application demand of robots in complex and unstructured environments (such as disaster rescue and field exploration) is increasing. These environments require robots to have high-speed motion capability and strong dynamic stability. Traditional robot balance technology mostly relies on adjusting gait (for legged robots) or lowering the center of gravity (for wheeled robots). These methods often have insufficient response and limited effect when moving at high speed or being subjected to severe impact. Therefore, designing a robot balance control method with rapid response and multiple degrees of freedom has become a technical problem to be solved by those skilled in the art. SUMMARY

[0003] To solve the above problems, the embodiment of the present application discloses a robot dynamic balance control method, which can realize dynamic balance adjustment during robot motion.

[0004] The first aspect of the embodiment of the present application discloses a robot dynamic balance control method, comprising:

[0005] obtaining current state information of the robot, the current state information including body posture information and environment perception information;

[0006] predicting an expected disturbance that the robot will be subjected to based on the environment perception information, and generating a feedforward control instruction based on the expected disturbance;

[0007] establishing a unified dynamic model of a unified balance actuator by taking the bionic spine unit and the active tail wing unit of the robot as the unified balance actuator;

[0008] solving a joint motion instruction of the bionic spine unit and the active tail wing unit within a future time period by rolling optimization of the unified dynamic model through a model predictive control algorithm based on the current state information and the feedforward control instruction, so as to minimize a predefined cost function, wherein the cost function at least includes a robot posture deviation term;

[0009] outputting the joint motion instruction to the actuators of the bionic spine unit and the active tail wing unit to control their cooperative motion.

[0010] As an optional implementation, in the first aspect of the embodiment of the present application, the step of predicting an expected disturbance that the robot will be subjected to based on the environment perception information comprises:

[0011] Obtain elevation data of the terrain in front of the robot through a visual sensor and a laser radar, and fuse to generate a real-time terrain height map;

[0012] According to the current motion speed of the robot and the gait planning, predict an expected touch point sequence of the robot foot or wheel end in N control periods in the future;

[0013] Match the expected touch point sequence with the real-time terrain height map, and calculate the terrain normal vector and the expected height difference of each expected touch point;

[0014] Based on the terrain normal vector and the expected height difference, calculate the expected disturbance force and the expected disturbance torque generated by each expected touch point on the robot body through a robot inverse dynamics model;

[0015] Combine all the calculated expected disturbance forces and the expected disturbance torques in time sequence to form the expected disturbance.

[0016] As an optional implementation, in the first aspect of the embodiment of the present application, the cost function further includes at least one of the following:

[0017] The energy consumption term of the bionic spine unit and the active tail wing joint;

[0018] The motion smoothness term of the bionic spine unit and the active tail wing joint.

[0019] As an optional implementation, in the first aspect of the embodiment of the present application, the cost function is defined as:

[0020] J = ∑ [W θ *(θ-θ ref ) 2 +W ω *(ω-ω ref ) 2 +W τ *τ 2 +W Δ τ*Δτ 2 ]

[0021] Wherein, θ and ω are the body attitude angle and angular velocity of the robot, θ ref and ω ref are the reference values of the attitude angle and the angular velocity; τ is the output torque of each joint; Δτ is the change rate of the joint torque; W θ , W ω , W τ , W Δτ are the corresponding weight coefficients;

[0022] The weight coefficients W θ and W ωDetermined in the following manner:

[0023] Determine the current machine stability parameters based on the robot's current state and the stability state formula;

[0024] When the machine stability parameter exceeds the first stability threshold, increase W. θ and W ω The weights;

[0025] When the machine stability parameter reaches the second stability threshold, W θ and W ω The weight is increased to the set maximum value.

[0026] As an optional implementation, in the first aspect of the present invention, the balance control method further includes:

[0027] Based on the robot's current state information and environmental perception information, predict the expected set of disturbances for the robot's next stage, where each disturbance type corresponds to an expected probability and intensity of occurrence;

[0028] The multiple degrees of freedom of motion of the bionic spine unit and active tail fin unit of the robot are defined as multiple basic balance action primitives, each of which includes balance performance parameters and energy consumption parameters.

[0029] The corresponding hedging weight coefficient is determined based on the expected disturbance set, wherein the hedging weight coefficient corresponds to the contribution of the action primitive in responding to the expected disturbance;

[0030] Based on the hedging weight coefficient, a combined motion command for the bionic spine unit and the active tail fin unit is synthesized; and the combined motion command is output to the corresponding actuator.

[0031] As an optional implementation, in a first aspect of the present invention, after predicting the expected disturbance that the robot will experience based on the environmental perception information, the method further includes:

[0032] Obtain a pre-stored balance motion library, which stores optimized spine-tail coordinated motion sequences corresponding to various typical disturbance scenarios;

[0033] The predicted expected disturbance is matched with the typical disturbance patterns in the balance action library, and the similarity is calculated; if the highest similarity exceeds a preset threshold, the corresponding motion sequence is called from the balance action library.

[0034] Based on the strength of the expected disturbance, the parameters in the invoked motion sequence are updated to generate feedforward control commands.

[0035] As an optional implementation, in the first aspect of the present invention, the balance control method further includes:

[0036] Real-time monitoring of changes in robot system parameters, including total mass, center of mass position, and moment of inertia about the main axis;

[0037] The system parameters are identified online using a Bayesian filtering algorithm. The specific steps are as follows:

[0038] The system parameters are constructed as state variables;

[0039] In each control cycle, the robot motion state predicted by the model predictive control algorithm is compared with the actual motion state measured by the sensor components, and the observation residual is calculated.

[0040] The posterior probability distribution of the state variables is updated based on the observed residuals;

[0041] The expected value of the posterior distribution is used as the latest identified system parameter to update the unified dynamic model.

[0042] A second aspect of this invention discloses a robot dynamic balance control system, comprising:

[0043] Acquisition module: used to acquire the robot's current state information, which includes the robot's posture information and environmental perception information;

[0044] Command generation module: used to predict the expected disturbances that the robot will experience based on the environmental perception information, and to generate feedforward control commands based on the expected disturbances;

[0045] Model building module: used to establish a unified dynamic model of the robot's bionic spine unit and active tail fin unit as a unified balancing actuator;

[0046] Optimization calculation module: Based on the current state information and the feedforward control command, it performs rolling optimization on the unified dynamics model through model predictive control algorithm to solve the joint motion command of the bionic spine unit and the active tail unit in a future time period, so as to minimize the predefined cost function, wherein the cost function includes at least the robot posture deviation term;

[0047] Cooperative control module: used to output the joint motion command to the actuators of the bionic spine unit and the active tail fin unit to control their cooperative motion.

[0048] A third aspect of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to execute the robot dynamic balance control method disclosed in the first aspect of the present invention.

[0049] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the robot dynamic balance control method disclosed in the first aspect of the present invention.

[0050] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0051] The control method in this embodiment of the invention comprehensively considers multiple factors such as the robot's posture information, environmental perception information, and dynamic model. By combining feedforward control and model predictive control, the system can maintain a good balance when facing different working environments and complex task requirements. This method can also ensure the robot's balance through rolling optimization and real-time adjustment, thereby improving the system's robustness. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart illustrating the robot dynamic balance control method disclosed in an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the expected disturbance prediction process disclosed in the embodiments of the present invention;

[0055] Figure 3 This is a schematic diagram of the feedforward control instruction generation process disclosed in an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram showing the dynamic adjustment of the robot as disclosed in an embodiment of the present invention;

[0057] Figure 5 This is a schematic diagram of the structure of a robot dynamic balance control system provided in an embodiment of the present invention;

[0058] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] It should be noted that the terms first, second, third, fourth, etc., in the specification and claims of this invention are used to distinguish different objects, not to describe a specific order. The terms used in the embodiments of this invention include and have, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0061] Example 1

[0062] Please see Figure 1 , Figure 1 This is a flowchart illustrating the robot dynamic balance control method disclosed in this embodiment of the invention. The execution entity of the method described in this embodiment is an execution entity composed of software and / or hardware. This execution entity can receive relevant information via wired or / or wireless means and can send certain instructions. It may also have certain processing and storage functions. This execution entity can control multiple devices, such as remote physical servers or cloud servers and related software, or local hosts or servers and related software that perform related operations on devices located in a certain location. In some scenarios, multiple storage devices can also be controlled; these storage devices may be placed in the same location as the devices or in different locations. Figure 1 As shown, the robot dynamic balance control method includes the following steps:

[0063] S101: Obtain the robot's current state information, which includes the robot's posture information and environmental perception information;

[0064] S102: Based on the environmental perception information, predict the expected disturbance that the robot will experience, and based on the expected disturbance, generate feedforward control commands;

[0065] S103: Using the bionic spine unit and active tail fin unit of the robot as a unified balancing actuator, establish a unified dynamic model of the balancing actuator;

[0066] S104: Based on the current state information and the feedforward control command, the unified dynamic model is rolled by the model predictive control algorithm to solve the joint motion command of the bionic spine unit and the active tail unit in a future time period, so as to minimize the predefined cost function, wherein the cost function includes at least the robot posture deviation term.

[0067] S105: Output the combined motion command to the actuators of the bionic spine unit and the active tail fin unit to control their coordinated motion.

[0068] Specifically, by using environmental perception information to predict the expected disturbances the robot will experience and generating feedforward control commands, it can compensate for potential disturbances in advance. For example, when the robot walks on uneven ground or is disturbed by external wind, the feedforward control can adjust the movements of the bionic spine unit and active tail fin unit in advance based on the predicted disturbances, thereby effectively counteracting the impact of the disturbances. This allows the robot to better maintain balance when facing various external disturbances, improving the robot's anti-interference capability.

[0069] In this embodiment of the invention, a unified dynamic model of the balancing actuator is established, and a model predictive control algorithm is used to perform rolling optimization on it to solve for joint motion commands, thereby minimizing a predefined cost function, which includes a robot posture deviation term. This enables the robot to accurately calculate the optimal motion mode of the bionic spine unit and the active tail unit based on the current state and predicted disturbances, thus controlling the robot's posture more precisely, reducing posture deviations, and improving the accuracy of balance control. The model predictive control algorithm can optimize the motion of the balancing actuator over a future time period, making decisions quickly based on the current state and expected disturbances, and generating corresponding control commands. This rolling optimization method allows the robot to adapt to changes in the environment and its own state in real time, achieving rapid response and dynamic adjustment, ensuring that the robot maintains balance throughout its movement.

[0070] Specifically, the rolling optimization calculation logic based on the Model Predictive Control (MPC) algorithm first transforms the unified dynamic model, which includes the bionic spine unit and the active tail unit, into a discrete-time state-space equation. This equation encompasses variables related to the robot's posture, spine and tail motion states, drive control variables for the spine and tail, and expected disturbance variables predicted based on environmental perception. Then, within each control cycle, starting from the current moment, an optimization problem for a specific number of future steps (i.e., the prediction time domain) is constructed. Optimization is achieved by minimizing the objective function, which includes core posture deviation penalty terms (to penalize the deviation between the robot's actual posture and the target posture), control energy consumption terms, and terminal stability penalty terms. It also needs to satisfy physical constraints (such as the range of spine bending angles and tail deflection speed limits), safety constraints (such as the robot's tilt angle not exceeding an imbalance threshold), and dynamic constraints (conforming to the motion relationships of the unified dynamic model). The process then proceeds to the rolling optimization solution stage. First, the current state is updated based on real-time sensor data. Then, the expected disturbance is substituted into the state equation as a known quantity, forming a feedforward-feedback composite control framework. Next, a quadratic programming algorithm is used to solve the optimization problem with the aforementioned constraints, obtaining the optimal control sequence. In actual execution, only the first control variable of this sequence is applied to the actuator. The process is repeated in the next control cycle, achieving a rolling forward of the optimization window. Finally, at the collaborative control level, the motion allocation between the spine and tail fin is completed through unified optimization. Low-frequency attitude adjustment (such as adapting to slope) mainly relies on spine bending to achieve center of gravity transfer, while high-frequency disturbance suppression (such as dealing with sudden gusts) preferentially uses rapid tail fin deflection to generate compensating torque. Simultaneously, cross terms are added to the objective function to avoid conflicts between the spine and tail fin movements. Finally, the joint motion command of the bionic spine unit and the active tail fin unit for a future time period is solved.

[0071] This control method comprehensively considers multiple factors, including the robot's posture information, environmental perception information, and dynamic model. By combining feedforward control and model predictive control, the system can maintain good control performance when facing different working environments and complex task requirements. Even if there is some uncertainty in the robot's model parameters or significant changes in the external environment, this method can ensure the robot's balance through rolling optimization and real-time adjustments, thus improving the system's robustness.

[0072] By treating the robot's biomimetic spine unit and active tail fin unit as a unified balancing actuator, establishing a unified dynamic model, and implementing coordinated control, the advantages of biomimetic structures can be fully utilized. The biomimetic spine unit and active tail fin unit can cooperate with each other to coordinately adjust the robot's posture and center of gravity, achieving more flexible and stable movement. This makes the robot's movement more in line with biomechanical principles, improving the robot's motion performance and adaptability.

[0073] The core innovation of this invention lies in proposing a three-in-one system architecture consisting of a biomimetic flexible spine unit, a multi-degree-of-freedom active tail fin unit, and a feedforward-feedback collaborative controller.

[0074] The system mainly consists of three parts:

[0075] Bionic flexible spinal unit: Located in the middle of the robot's torso, connecting the upper and lower body.

[0076] Multi-degree-of-freedom active tail fin unit: connected to the end of the bionic spine unit or the rear of the robot's torso.

[0077] Intelligent sensing and control system: including sensor group, central processing unit and feedforward-feedback cooperative control algorithm.

[0078] Detailed composition of each component: a) Bionic flexible spinal unit: Structure: Composed of multiple series-connected universal joint modules. Each module has a built-in micro motor, high-precision encoder, and torque sensor. The modules are movably connected, allowing for two degrees of freedom of movement: pitch and yaw.

[0079] The spine in this invention is not a simple rigid connection or a passive flexible component, but a distributed active posture control system. It can not only bend in an S-shape like an animal spine, assisting the robot in achieving more flexible steering and gait adjustments, but its built-in sensors can also perceive changes in trunk posture and stress distribution in real time, providing rich proprioceptive data for the control system.

[0080] b) Multi-DOF Active Tail Unit: Structure: This unit connects to the end of the spine via a high-speed responsive rotary joint, which provides yaw motion for the tail. The tail body consists of a skeleton and adjustable counterweights. The skeleton is composed of multiple segments and can incorporate linear actuators to achieve active and rapid extension and retraction of the tail length. The counterweights can move along tailbone rails or be quickly replaced, allowing dynamic adjustment of the tail's base axis rotational inertia according to different robot loads and tasks, optimizing balance performance.

[0081] Three-degree-of-freedom tail: Traditional tails typically only swing left and right. The tail of this invention has three degrees of freedom: yaw, pitch, and extension. The extension degree of freedom can greatly change the moment of inertia of the tail, thereby generating a larger balancing torque with a smaller swing amplitude, resulting in higher efficiency and lower energy consumption.

[0082] Intelligent perception and control system: Sensor group: including IMU (inertial measurement unit) in various parts of the robot body, force / torque sensors (in the feet or wheel hubs), encoders and torque sensors in the spine unit, and vision sensors (such as cameras and lidar) for environmental perception.

[0083] Feedforward-feedback cooperative controller:

[0084] Feedforward control: This is the key to this scheme. The controller predicts upcoming posture disturbances (e.g., when taking a step, jumping from a height, or about to cross an obstacle) based on motion planning algorithms and visual / terrain pre-sensing information. Before the imbalance actually occurs, instructions are generated in advance to drive the spine and tail to move to the optimal position to counteract the disturbance.

[0085] Feedback control: Rapid compensation is performed based on real-time measurements of current attitude deviations (such as tilt angle and angular velocity) from sensors such as IMUs. The system employs stability detection or model predictive control algorithms to collaboratively calculate multiple degrees of freedom of the spine and tail, solving for the optimal joint motion scheme to minimize attitude deviations.

[0086] Synergistic Strategy: The controller treats the spine and tail as a unified balancing organ. For example, for a small disturbance, only a slight bend in the spine may be needed to adjust the center of gravity; for a large, violent impact (such as being hit), a combined mode of spine bending and high-speed, large-amplitude tail swinging is activated.

[0087] Unified dynamics model: Establish the overall system dynamics equations including the robot body, bionic spine (N joints), and active tail fin (M joints).

[0088] State variables include the main body posture (roll angle, pitch angle), the angles of each joint of the spine, the angles of each joint of the tail, and all angular velocities.

[0089] Control input: A set of motor torque commands for all joints of the spine and tail.

[0090] Output: The most critical system state, usually the robot's center of mass position and body orientation angle.

[0091] Model Predictive Controller (MPC) - The Core of Joint Optimization: This invention uses MPC as the core controller, and its innovation lies in the fact that the solution range covers all joints from the spine to the tail.

[0092] Predictive Time Domain: Based on the current sensor state and the unified model described above, the controller predicts the system's behavior over a short period of time in the future.

[0093] Optimization solution: In the prediction time domain, find a set of optimal control input sequences that minimize the following cost function.

[0094] Rolling optimization: Only the first control instruction in the optimization sequence is executed, and then the state is updated with new sensor data in the next control cycle, and prediction and optimization are performed again. This achieves closed-loop control.

[0095] The MPC controller automatically calculates the optimal coordination strategy for the spine and tail. For example: Scenario A (slow, fine-tuning): This might primarily utilize the bending of the spine to fine-tune the center of gravity position, with the tail only needing slight coordination to conserve energy. Scenario B (withstanding severe impacts): The MPC will calculate a strategy requiring instantaneous bending of the spine while the tail accelerates and extends at full speed to maximize rotational inertia. This achieves true synergy, automatically generated by the optimization algorithm, rather than manually preset rules. A specific adjustment illustration is shown below. Figure 4 As shown.

[0096] More preferably, such as Figure 2 As shown, the step of predicting the expected disturbances that the robot will experience based on the environmental perception information includes:

[0097] S1021: Acquire elevation data of the terrain in front of the robot through visual sensors and LiDAR, and fuse them to generate a real-time terrain height map;

[0098] S1022: Based on the robot's current movement speed and gait planning, predict the expected contact point sequence of the robot's foot or wheel end within the next N control cycles;

[0099] S1023: Match the expected contact point sequence with the real-time terrain height map, and calculate the terrain normal vector and expected height difference for each expected contact point;

[0100] S1024: Based on the terrain normal vector and the expected height difference, calculate the expected disturbance force and disturbance torque generated by each expected contact point on the robot body through the robot inverse dynamics model;

[0101] S1025: Combine all the calculated expected disturbance forces and disturbance moments in a time series to form the expected disturbance.

[0102] This invention generates a real-time terrain height map by fusing visual sensors and LiDAR, which can comprehensively capture the three-dimensional features of the terrain in front of the robot and avoid the blind spots of a single sensor in complex terrain (such as potholes, slopes, and protruding obstacles). By combining the current movement speed and gait planning to predict the expected contact points of the foot / wheel end in multiple control cycles, and then matching the terrain height map to calculate the terrain normal vector and height difference, the passive response to the disturbance can be transformed into the active prediction of the disturbance. Furthermore, based on the inverse dynamics model, the disturbance force and torque can be calculated, which can accurately quantify the actual impact of the terrain on the robot body and significantly reduce the disturbance prediction error.

[0103] The solution of this invention, based on different terrain features (such as the inclination angle of a slope and the height of a protrusion), can predict potential disturbances such as impacts and tilting when the foot / wheel end contacts the terrain by calculating the terrain normal vector and the height difference. By combining the expected disturbances in a time sequence, the robot can know the pattern of disturbance changes in advance, and thus adjust the movements of the bionic spine and active tail fin in advance in balance control. For example, when facing continuous potholes, the robot can plan the movements in advance to offset the impact of each foot contact with the ground, effectively avoiding problems such as posture imbalance and movement jamming in complex terrain, and significantly improving its mobility in unstructured environments (such as the wild and ruins).

[0104] The effectiveness of feedforward control is highly dependent on the accuracy and timeliness of the anticipated disturbance information. The disturbance force and torque data generated by this scheme, which includes time series and precise quantification, can provide a direct and reliable basis for the generation of feedforward control commands, enabling feedforward control to more accurately compensate for the impact of disturbances in advance. Compared with traditional control methods that rely on real-time feedback adjustments, feedforward control combined with this high-quality anticipated disturbance can reduce the lag in robot posture adjustment caused by feedback delays, making balance control faster and smoother. For example, before the robot steps onto a slope, feedforward commands are generated based on the predicted disturbance torque to adjust the spinal curvature angle and tail fin deflection state in advance, avoiding posture tilting the moment it steps onto the slope.

[0105] In addition to the methods mentioned above, wind speed sensors can also be used for wind speed measurement during implementation. Wind speed sensors can directly and dynamically collect wind data without the need for indirect derivation through complex terrain matching and inverse dynamics calculations. They can quickly convert real-time wind speed and direction into quantifiable wind-induced disturbance forces (such as lateral thrust) and disturbance torques (such as the torque that causes the robot to rotate around its vertical axis). Compared to relying on environmental models to indirectly estimate wind disturbances, direct measurement can reduce model errors and calculation delays. Especially in scenarios with sudden gusts and rapid changes in wind speed, it allows the robot to obtain wind disturbance information in real time, providing more timely and accurate input for feedforward control and avoiding balance instability caused by delayed prediction of wind disturbances.

[0106] When a robot is in a complex scenario with overlapping terrain and wind disturbances (such as encountering gusts while walking on an outdoor slope), wind data measured by a wind speed sensor can be fused with disturbance torque data acquired through terrain perception to form more complete multi-source disturbance information. Feedforward control commands generated based on this comprehensive disturbance information allow the bionic spine unit and active tail fin unit to adjust their posture to cope with terrain impacts while simultaneously compensating for the effects of wind loads. For example, when walking on a slope, the robot can both adjust its center of gravity by bending its spine to counteract the torque caused by the terrain tilt and generate a reverse torque by deflecting its tail fin to counteract the lateral tilting tendency caused by gusts. This allows the robot to maintain stability even in scenarios with multiple disturbances, further improving the robustness of overall balance control.

[0107] More preferably, the cost function further includes at least one of the following:

[0108] Energy consumption of the bionic spinal unit and the active tail fin joint;

[0109] The motion smoothness of the bionic spine unit and the active tail fin joint.

[0110] When the cost function includes joint energy consumption, the model predictive control algorithm automatically considers the energy consumption of the spine and tail joints as one of the optimization objectives during the solution of joint motion commands. This is achieved by constraining the output amplitude of joint drives, reducing unnecessary large movements or high-frequency start-stops, and avoiding energy waste. For example, when walking on flat terrain, the algorithm plans low-energy spine bending angles and tail wing deflection trajectories, rather than simply pursuing posture stability while ignoring energy consumption. For battery-powered mobile robots, this significantly reduces energy consumption per unit time, effectively extends the runtime of a single task, and reduces the impact of frequent charging on work efficiency.

[0111] The introduction of a motion smoothness term allows the algorithm to prioritize joint motion commands with continuous trajectories and gradual changes in velocity / acceleration during optimization, avoiding abrupt rotations, rapid accelerations, or sudden decelerations in the spine and tail joints. On one hand, smooth joint motion significantly reduces friction, impact, and vibration between mechanical structures, lowers the wear rate of components such as gears and bearings, extends the lifespan of the robot's actuators, and reduces maintenance costs and downtime frequency. On the other hand, smooth motion transmitted to the robot body avoids body posture fluctuations caused by abrupt joint movements. For example, when dealing with minor terrain disturbances, smooth tail adjustments can avoid additional impact on the robot's center of gravity, further ensuring the stability of balance control and improving the overall smoothness of robot movement, making it particularly suitable for scenarios requiring high motion accuracy and stability.

[0112] More preferably, the cost function is defined as:

[0113] J = ∑[Wθ *(θ-θ ref ) 2 +W ω *(ω-ω ref ) 2 +W τ *τ 2 +W Δ τ*Δτ 2 ]

[0114] Where θ and ω are the robot's body attitude angle and angular velocity, respectively. ref and ω ref Here are the reference values ​​for attitude angle and angular velocity; τ is the output torque of each joint; Δτ is the rate of change of joint torque; W θ W ω W τ W Δτ These are the corresponding weighting coefficients;

[0115] The weighting coefficient W θ and W ω Determined in the following manner:

[0116] Determine the current machine stability parameters based on the robot's current state and the stability state formula;

[0117] When the machine stability parameter exceeds the first stability threshold, increase W. θ and W ω The weights;

[0118] When the machine stability parameter reaches the second stability threshold, W θ and W ω The weight is increased to the set maximum value.

[0119] The stability state formula is as follows:

[0120]

[0121] Where V(x) is the machine stability parameter, and ω and θ are the robot's angular velocity and attitude angle, respectively; θ ref and ω ref For attitude angle reference values ​​and angular velocity reference values; ω T 、(θ-θ ref ) T and (ω-ω) ref ) T They are respectively ω, (θ-θ) ref ) and (ω-ω ref The corresponding transpose; k θ and k ωThese are the weighting coefficients for the angular velocity term and the attitude deviation term, respectively; K is the weighting matrix for the attitude deviation term; and I is the moment of inertia matrix.

[0122] The aforementioned multi-dimensional weighted optimization can avoid the limitations of single-objective optimization. For example, it can prevent excessive joint torque and energy consumption from being caused by pursuing absolute posture stability, or the sacrifice of posture adjustment timeliness in order to reduce energy consumption, so as to enable the robot to achieve a dynamic balance of stability, energy consumption and smoothness in balance control.

[0123] The adjustable weighting coefficients enable the system to adapt to different task scenarios. For example, in routine patrol tasks, a lower W can be preset. θ and W ω This prioritizes energy efficiency and smoothness, extending battery life and reducing mechanical wear. During high-precision tasks (such as equipment maintenance), increasing the base weights ensures posture stability, preventing minor vibrations from affecting accuracy. Simultaneously, the dynamic adjustment mechanism allows the robot to autonomously adapt to environmental changes without human intervention. During transitions from smooth surfaces to rugged terrain, the robot automatically switches weight strategies, further enhancing the intelligence and practicality of the control scheme.

[0124] Dynamic adjustment strategy: When the system is stable, V(x) is small, the robot is nearly upright and almost motionless: W θ and W ω Maintain a low base value. At this point, the controller focuses more on W. τ and W Δτ This prioritizes energy efficiency and smooth movement. It may only require minor adjustments to the spine and tail.

[0125] When the system becomes unstable, V(x) exceeds the first threshold, and the robot begins to tilt noticeably: the controller linearly increases W. θ and W ω The weighting of these factors is significantly increased. Maintaining balance and suppressing swaying are given higher priority.

[0126] When the system is highly unstable, V(x) reaches the second threshold, and the robot is about to fall: W θ and W ω The weight is increased to its maximum fixed value. At this time, W τ and W Δτ The weight of the tail has little impact. The controller's sole objective is to restore balance. Therefore, it needs to control the spine and tail to move violently and rapidly, even though this is very power-intensive and involves strenuous activity.

[0127] Specifically, the essence of robot balance is the conversion between kinetic and potential energy. The process of falling is the conversion of potential energy into kinetic energy. V(x) directly measures the total energy of the system; the higher the energy, the more unstable the system.

[0128] However, in implementation, a problem arises: a stationary robot at a large angle and a robot rapidly returning to center may have similar machine stability values, but the former is far more risky than the latter. This problem is addressed in this embodiment by introducing a damping term; it directly penalizes any non-zero angular velocity. For the stationary robot, this term is 0; for the robot returning to center, this term is quite large. This allows V(x) to more finely distinguish between different types of instability. It now penalizes both the distance from the equilibrium point and the velocity away from the equilibrium point.

[0129] In the simulation, typical perturbations (such as lateral thrust) are applied to the robot.

[0130] Observe the V(x) curve: A well-designed V(x) should exhibit a smooth, single-peaked curve during the robot's stabilization process. If the V(x) peaks and then declines slowly, it indicates insufficient damping, and k should be increased. ω Or Λ. If the peak value of V(x) is too high or rises too quickly, it indicates insufficient penalty for angle deviation, and k should be increased. ω .

[0131] In addition to the above methods, in the rolling optimization step using the model predictive control MPC algorithm, a virtual potential field term is introduced into the cost function, where:

[0132] Generate an attractive potential field around the robot's current center of mass; and

[0133] A repulsive potential field is generated based on the expected imbalance direction determined by the expected disturbance; wherein, the virtual potential field term is used to guide the MPC algorithm to calculate the joint motion command that makes the robot's center of mass face the attractive potential field and move away from the repulsive potential field.

[0134] The repulsive potential field is dynamically generated based on anticipated disturbances. For example, if a terrain protrusion on the left is predicted to cause imbalance to the left, a strong repulsive potential field is generated on the left. This allows the MPC algorithm to not only compensate for disturbances through feedforward control when planning the joint motion of the spine and tail, but also to form a double protection by actively moving the center of mass away from the dangerous direction. For example, when facing an anticipated disturbance on a downhill slope (which may cause the center of mass to shift forward and become unbalanced), the attractive potential field will pull the center of mass to stay at the support center, while the repulsive potential field will push the center of mass to adjust backward. Together, they drive the spine to bend slightly backward and the tail to tilt forward, forming a coordinated backward movement of the center of mass, which can avoid instability risks in advance more effectively than simply relying on attitude deviation correction. The virtual potential field term can dynamically adjust its contribution weight to the cost function through the potential field strength (e.g., the greater the intensity of the anticipated disturbance, the higher the weight of the repulsive potential field term), allowing the MPC algorithm to autonomously adapt to the disturbance level without manual parameter adjustment. In scenarios with slight disturbances, the potential field term has a lower weight, and the control center of gravity is biased towards energy consumption and smoothness; in scenarios with strong disturbances, the weight of the potential field term is automatically increased, forcing the center of gravity to move towards the stable region, and prioritizing the protection of balance.

[0135] More preferably, the balance control method further includes:

[0136] Based on the robot's current state information and environmental perception information, predict the expected set of disturbances for the robot's next stage, where each disturbance type corresponds to an expected probability and intensity of occurrence;

[0137] The multiple degrees of freedom of motion of the bionic spine unit and active tail fin unit of the robot are defined as multiple basic balance action primitives, each of which includes balance performance parameters and energy consumption parameters.

[0138] The corresponding hedging weight coefficient is determined based on the expected disturbance set, wherein the hedging weight coefficient corresponds to the contribution of the action primitive in responding to the expected disturbance;

[0139] Based on the hedging weight coefficient, a combined motion command for the bionic spine unit and the active tail fin unit is synthesized; and the combined motion command is output to the corresponding actuator.

[0140] By predicting the expected set of disturbances in the next stage (including the type of disturbance, probability of occurrence, and intensity), robots can prepare for various potential disturbances in advance, avoiding the limitations of predicting a single type of disturbance. For example, in outdoor scenarios, it can predict the impact probability of terrain protrusions and also consider the possibility of gusts of wind, enabling balance control to shift from passive response to active prediction of multiple scenarios, significantly improving its adaptability to complex and ever-changing environments.

[0141] The degrees of freedom of motion of the spine and tail fin are defined as basic balance motion primitives (each primitive includes efficiency and energy consumption parameters), which allows for flexible combination of motions according to disturbance requirements. By quantifying the contribution of each primitive to a specific disturbance through hedging weight coefficients, the efficiency of motions and the intensity of disturbances can be precisely matched. For example, when dealing with high-probability crosswind disturbances, the rapid deflection primitive of the tail fin is given a higher weight; when dealing with terrain tilt, the contribution of the spine bending primitive is increased. While ensuring balance, the energy consumption parameter constraint avoids motion redundancy, achieving the optimal ratio of efficiency to energy consumption.

[0142] The combination of probabilistic perturbation models and weighted synthesis mechanisms enables robots to make better decisions in uncertain environments. When multiple perturbations may occur simultaneously (such as uneven ground and gusts of wind), the system can dynamically adjust the weight coefficients to allow different motion primitives to work together (such as the spine adjusting the center of gravity and the tail fin counteracting crosswinds), avoiding the limitations of single actions in dealing with complex perturbations. This probabilistic combination strategy considers both the probability of perturbation occurrence and the cost-effectiveness of actions, making balance control more intelligent and robust, especially suitable for complex task scenarios in unstructured environments.

[0143] More preferably, such as Figure 3 The method shown includes, after predicting the expected disturbances the robot will experience based on the environmental perception information:

[0144] S102a: Obtain a pre-stored balance motion library, which stores optimized spine-tail coordinated motion sequences corresponding to various typical disturbance scenarios;

[0145] S102b: Match the predicted expected disturbance with the typical disturbance patterns in the balance action library and calculate the similarity; if the highest similarity exceeds a preset threshold, call the corresponding motion sequence from the balance action library;

[0146] S102c: Based on the intensity of the current expected disturbance, update the parameters in the invoked motion sequence to generate feedforward control commands.

[0147] The pre-stored balance motion library contains optimized spine-tail coordinated motion sequences for various typical disturbance scenarios (such as slope impacts, lateral gusts, and protruding obstacles). These sequences are optimal solutions obtained in advance through offline simulation and experimental verification, eliminating the need for resolving them using complex algorithms during real-time control. When the predicted expected disturbance matches a typical pattern in the library and the similarity meets the standard, the existing motion sequence can be directly called, significantly reducing the computational power consumption of real-time calculations such as model prediction and rolling optimization, avoiding command delays caused by complex calculations, and ensuring timely output of control commands.

[0148] The motion sequences corresponding to typical disturbances have been optimized and verified in advance, possessing mature and efficient balance adjustment logic. After being invoked, only minor adjustments to parameters based on the current disturbance intensity are needed to generate feedforward control commands, eliminating the time-consuming process of sensing disturbances, building models, and solving optimization problems. For example, when the robot predicts a lateral gust disturbance that highly matches the moderate-intensity lateral wind pattern in the motion library, it can directly invoke the corresponding tail wing rapid deflection + small-amplitude spinal lateral bending coordinated sequence. Then, based on the actual gust intensity, parameters such as the tail wing deflection angle and spinal bending rate are adjusted to quickly generate control commands. This allows the robot to initiate balance adjustment before or in the early stages of the disturbance, avoiding posture imbalance due to response lag and significantly improving the timeliness of disturbance resistance.

[0149] The motion sequences in the motion library have been optimized offline and verified in real-world scenarios. Their balance effect, motion coordination, and energy consumption rationality have all been tested. Compared to the deviations in optimization results that may be caused by model errors and parameter fluctuations during real-time calculations, calling mature sequences can reduce the uncertainty of control commands and improve the reliability of balance adjustments. At the same time, by matching typical patterns and updating parameters according to intensity, it can not only adapt to the typical disturbance scenarios already covered in the library, but also handle different intensities of the same disturbance type (such as light, moderate, and severe slope impacts) through parameter fine-tuning. This avoids control scheme failures due to differences in disturbance intensity, expands the scenario coverage of feedforward control, and enables the robot to stably output efficient balance control commands in diverse and recurring typical disturbance scenarios.

[0150] More preferably, the balance control method further includes:

[0151] Real-time monitoring of changes in robot system parameters, including total mass, center of mass position, and moment of inertia about the main axis;

[0152] The system parameters are identified online using a Bayesian filtering algorithm. The specific steps are as follows:

[0153] The system parameters are constructed as state variables;

[0154] In each control cycle, the robot motion state predicted by the model predictive control algorithm is compared with the actual motion state measured by the sensor components, and the observation residual is calculated.

[0155] The posterior probability distribution of the state variables is updated based on the observed residuals;

[0156] The expected value of the posterior distribution is used as the latest identified system parameter to update the unified dynamic model.

[0157] In actual operation, system parameters such as total mass, center of mass position, and moment of inertia about the main axis (e.g., changes in inertial parameters due to component wear or attitude adjustments) of a robot are prone to dynamic changes with the working conditions. If relying on an initial fixed model, control errors will occur due to parameter mismatch. By constructing system parameters as state variables through Bayesian filtering, the posterior probability distribution of parameters is updated each cycle based on the observation residuals of the model's predicted motion state and the sensor's measured state. The expected value is used as the latest identified parameter, which can correct the core parameters of the unified dynamic model in real time, eliminate model deviations caused by parameter drift, and ensure that the model always remains consistent with the robot's actual physical state. This provides a reliable model foundation for the accurate generation of subsequent balance control commands.

[0158] The solution of this invention enables robots to autonomously adapt to changes in system parameters under different operating conditions without the need for manual model recalibration. For example, when a robot switches from an unloaded state to a loaded state with detection equipment, its total mass and center of mass position change. Bayesian filtering can quickly identify the new parameters and update the model by observing the residuals, allowing subsequent spine-tail coordinated control commands to be replanned based on the new parameters (such as adjusting the torque output amplitude to adapt to load changes), avoiding lag or failure of balance control due to outdated parameters. Furthermore, if the robot's rotational inertia changes due to component wear after long-term operation, online identification can compensate for this difference in real time, ensuring that the control strategy adapts to the robot's physical state throughout its entire lifecycle, significantly improving control adaptability under multiple operating conditions.

[0159] The control method in this embodiment of the invention comprehensively considers multiple factors such as the robot's posture information, environmental perception information, and dynamic model. By combining feedforward control and model predictive control, the system can maintain a good balance when facing different working environments and complex task requirements. This method can also ensure the robot's balance through rolling optimization and real-time adjustment, thereby improving the system's robustness.

[0160] Example 2

[0161] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of the robot dynamic balance control system disclosed in an embodiment of the present invention. Figure 5 As shown, the robot's dynamic balance control system may include:

[0162] Acquisition module 21: used to acquire the robot's current state information, which includes the robot's posture information and environmental perception information;

[0163] Command generation module 22: used to predict the expected disturbances that the robot will experience based on the environmental perception information, and generate feedforward control commands based on the expected disturbances;

[0164] Model building module 23: used to establish a unified dynamic model of the balance actuator, taking the bionic spine unit and active tail fin unit of the robot as a unified balance actuator;

[0165] Optimization calculation module 24: Based on the current state information and the feedforward control command, it performs rolling optimization on the unified dynamics model through model predictive control algorithm to solve the joint motion command of the bionic spine unit and the active tail unit in a future time period, so as to minimize the predefined cost function, wherein the cost function includes at least the robot posture deviation term;

[0166] Cooperative control module 25: used to output the joint motion command to the actuators of the bionic spine unit and the active tail fin unit to control their cooperative motion.

[0167] The control method in this embodiment of the invention comprehensively considers multiple factors such as the robot's posture information, environmental perception information, and dynamic model. By combining feedforward control and model predictive control, the system can maintain a good balance when facing different working environments and complex task requirements. This method can also ensure the robot's balance through rolling optimization and real-time adjustment, thereby improving the system's robustness.

[0168] Example 3

[0169] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device can be a computer, a server, etc. Of course, in certain cases, it can also be a mobile phone, tablet computer, monitoring terminal, or other smart device, as well as an image acquisition device with processing capabilities. Figure 6 As shown, the electronic device may include:

[0170] Memory 510 storing executable program code;

[0171] Processor 520 coupled to memory 510;

[0172] The processor 520 calls the executable program code stored in the memory 510 to execute some or all of the steps in the robot dynamic balance control method in Embodiment 1.

[0173] This invention discloses a computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps in the robot dynamic balance control method of Embodiment 1.

[0174] This invention also discloses a computer program product, wherein when the computer program product is run on a computer, the computer executes some or all of the steps in the robot dynamic balance control method of Embodiment 1.

[0175] This invention also discloses an application publishing platform, which is used to publish computer program products. When the computer program products are run on a computer, the computer executes some or all of the steps in the robot dynamic balance control method in Embodiment 1.

[0176] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0177] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they can be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0178] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0179] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several requests to cause a computer device (which can be a personal computer, server, or network device, specifically a processor in the computer device) to execute some or all of the steps of the methods described in the various embodiments of the present invention.

[0180] In the embodiments provided by this invention, it should be understood that B corresponding to A means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0181] Those skilled in the art will understand that some or all of the steps in the various methods of the embodiments described can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0182] The robot dynamic balance control method, system, electronic device, and storage medium disclosed in the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for dynamic balance control of a robot, characterized in that, include: Obtain the robot's current state information, which includes the robot's posture information and environmental perception information; Based on the environmental perception information, predict the expected disturbances that the robot will experience, and generate feedforward control commands based on the expected disturbances. Using the robot's bionic spine unit and active tail fin unit as a unified balancing actuator, a unified dynamic model of the balancing actuator is established. Based on the current state information and the feedforward control command, a model predictive control algorithm is used to perform rolling optimization on the unified dynamics model to solve for the joint motion command of the bionic spine unit and the active tail unit in a future time period, so as to minimize a predefined cost function, wherein the cost function includes at least a robot posture deviation term; the cost function also includes at least one of the following: Energy consumption of the bionic spinal unit and the active tail fin joint; The motion smoothness term of the biomimetic spinal unit and the active tail fin joint; the cost function is defined as: ； Where θ and ω are the robot's body attitude angle and angular velocity, respectively. and Here are the reference values ​​for attitude angle and angular velocity; τ is the output torque of each joint; Δτ is the rate of change of joint torque. , , , These are the corresponding weighting coefficients; The weighting coefficient and Determined in the following manner: Determine the current machine stability parameters based on the robot's current state and the stability state formula; When the machine stability parameter exceeds the first stability threshold, increase and The weights; When the machine stability parameter reaches the second stability threshold, and The weights are increased to the set maximum value; where the stability state formula is: ； in, For machine stability parameters, and These are the robot's angular velocity and attitude angle, respectively. and These are the reference values ​​for attitude angle and angular velocity. , and They are respectively , and The corresponding transpose; and These are the weighting coefficients for the angular velocity term and the attitude deviation term, respectively. Here is the weight matrix for the attitude deviation term. Here is the rotational inertia matrix; The combined motion command is output to the actuators of the bionic spine unit and the active tail fin unit to control their coordinated motion.

2. The robot dynamic balance control method as described in claim 1, characterized in that, The method of predicting the expected disturbances that the robot will experience based on the environmental perception information includes: The robot acquires elevation data of the terrain in front of it using visual sensors and LiDAR, and then fuses the data to generate a real-time terrain height map. Based on the robot's current movement speed and gait planning, predict the expected contact point sequence of the robot's foot or wheel end within the next N control cycles; The expected contact point sequence is matched with the real-time terrain height map to calculate the terrain normal vector and expected height difference for each expected contact point; Based on the terrain normal vector and the expected height difference, the expected disturbance force and disturbance torque generated by each expected contact point on the robot body are calculated by the robot inverse dynamics model. The calculated expected disturbance forces and disturbance moments are combined in time series to form the expected disturbance.

3. The robot dynamic balance control method as described in claim 1, characterized in that, The balance control method further includes: Based on the robot's current state information and environmental perception information, predict the expected set of disturbances for the robot's next stage, where each disturbance type corresponds to an expected probability and intensity of occurrence; The multiple degrees of freedom of motion of the bionic spine unit and active tail fin unit of the robot are defined as multiple basic balance action primitives, each of which includes balance performance parameters and energy consumption parameters. The corresponding hedging weight coefficient is determined based on the expected disturbance set, wherein the hedging weight coefficient corresponds to the contribution of the action primitive in responding to the expected disturbance; Based on the hedging weight coefficient, a combined motion command for the bionic spine unit and the active tail fin unit is synthesized; and the combined motion command is output to the corresponding actuator.

4. The robot dynamic balance control method as described in claim 1, characterized in that, Following the prediction of the expected disturbances the robot will experience based on the environmental perception information, the method further includes: Obtain a pre-stored balance motion library, which stores optimized spine-tail coordinated motion sequences corresponding to various typical disturbance scenarios; The predicted expected disturbance is matched with the typical disturbance patterns in the balance action library, and the similarity is calculated; if the highest similarity exceeds a preset threshold, the corresponding motion sequence is called from the balance action library. Based on the strength of the expected disturbance, the parameters in the invoked motion sequence are updated to generate feedforward control commands.

5. The robot dynamic balance control method as described in claim 1, characterized in that, The balance control method further includes: Real-time monitoring of changes in robot system parameters, including total mass, center of mass position, and moment of inertia about the main axis; The system parameters are identified online using a Bayesian filtering algorithm. The specific steps are as follows: The system parameters are constructed as state variables; In each control cycle, the robot motion state predicted by the model predictive control algorithm is compared with the actual motion state measured by the sensor components, and the observation residual is calculated. The posterior probability distribution of the state variables is updated based on the observed residuals; The expected value of the posterior distribution is used as the latest identified system parameter to update the unified dynamic model.

6. A dynamic balance control system for a robot, characterized in that, include: Acquisition module: used to acquire the robot's current state information, which includes the robot's posture information and environmental perception information; Command generation module: used to predict the expected disturbances that the robot will experience based on the environmental perception information, and to generate feedforward control commands based on the expected disturbances; Model building module: used to establish a unified dynamic model of the robot's bionic spine unit and active tail fin unit as a unified balancing actuator; The optimization calculation module is used to perform rolling optimization on the unified dynamics model based on the current state information and the feedforward control command, using a model predictive control algorithm to solve for the joint motion command of the bionic spine unit and the active tail unit within a future time period, so as to minimize a predefined cost function, wherein the cost function includes at least a robot posture deviation term; the cost function also includes at least one of the following: Energy consumption of the bionic spinal unit and the active tail fin joint; The motion smoothness term of the biomimetic spinal unit and the active tail fin joint; the cost function is defined as: ； in, and These are the robot's body attitude angle and angular velocity, respectively. and Attitude angle reference value and angular velocity reference value; The output torque of each joint; The rate of change of joint torque; , , , These are the corresponding weighting coefficients; The weighting coefficient and Determined in the following manner: Determine the current machine stability parameters based on the robot's current state and the stability state formula; When the machine stability parameter exceeds the first stability threshold, increase and The weights; When the machine stability parameter reaches the second stability threshold, and The weights are increased to the set maximum value; where the stability state formula is: ； in, For machine stability parameters, and These are the robot's angular velocity and attitude angle, respectively. and These are the reference values ​​for attitude angle and angular velocity. , and They are respectively , and The corresponding transpose; and These are the weighting coefficients for the angular velocity term and the attitude deviation term, respectively. Here is the weight matrix for the attitude deviation term. Here is the rotational inertia matrix; Cooperative control module: used to output the joint motion command to the actuators of the bionic spine unit and the active tail fin unit to control their cooperative motion.

7. An electronic device, characterized in that, include: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the robot dynamic balance control method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to perform the robot dynamic balance control method according to any one of claims 1 to 5.

Images