Quadruped robot control method

Abstract

The application relates to a kind of quadruped robot control method.The quadruped robot includes base body, drive assembly, bearing assembly and control assembly, bearing assembly includes the shell rotating around first bearing rotation axis and the bearing rotating around second bearing rotation axis.The method generates compensation control quantity according to the attitude deviation of bearing or the article to be transported;Further obtain motion state information and the state information of bearing assembly, and the motion bearing collaborative control strategy model output corresponding to the feedforward compensation quantity of two bearing rotation axes is obtained by reinforcement learning training, and is fused with the feedback compensation quantity based on attitude deviation, to keep the bearing attitude stable during the movement of quadruped robot, reduce the risk of the article to be transported to shake, damage or liquid overflow.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and more specifically, to a quadruped robot control method, which is particularly suitable for quadruped robot bearing posture compensation control for stable transportation of fragile and spilled items under complex terrain conditions. Background Technology

[0002] With the development of smart catering, unmanned retail, office building delivery, hotel services, and commercial complex delivery, service robots for short-distance indoor and outdoor delivery are gradually being applied. Most existing delivery robots use wheeled mobility structures, which require a high degree of ground flatness. Quadruped robots can improve their ability to traverse steps, thresholds, ramps, ditches, or uneven ground, but they are prone to pitching, rolling, vibration, or acceleration / deceleration impacts during walking, turning, climbing, obstacle crossing, and starting / stopping. When transporting beverages, soups, reagents, precision items, or other fragile or easily spilled items, these bumps and posture changes can easily cause the items to shake, tilt, collide, or be damaged, or cause liquids to spill from containers.

[0003] To improve load-bearing stability, existing solutions can install two-degree-of-freedom load-bearing components on quadruped robots and control the balancing drive mechanism based on the posture deviation of the load-bearing component or the object to be transported. However, if feedback adjustment is only performed after the posture deviation has already occurred, the response to instantaneous disturbances such as gait switching, foot landing, obstacle crossing impact, sudden stop, or turning is often not timely enough.

[0004] Other approaches employ reinforcement learning to train quadrupedal or wheeled robots to walk and adapt to terrain. Their state and motion spaces are typically set for the base posture, driven leg joints, and wheel drive, generally excluding the rotational degrees of freedom of the load-bearing components around the two load-bearing rotation axes. If a control strategy trained solely on the robot's walking is directly applied, and load-bearing component posture feedback is then added separately, motion control and load compensation become independent, making it difficult to generate coordinated load compensation actions in advance to address walking disturbances.

[0005] Therefore, how to combine the state and movement of the carrier component around the two carrier rotation axes during the movement control of a quadruped robot, and perform more timely and coordinated compensatory control of the carrier component's posture to reduce the risk of damage to the transported goods or liquid spillage during transportation, has become a technical problem that needs to be solved. Summary of the Invention

[0006] The purpose of this application is to provide a quadruped robot control method to alleviate the problems of delayed response to bearing posture compensation and difficulty in coordinating body motion control and bearing compensation in existing solutions.

[0007] To achieve the above objectives, the technical solution provided in this application is as follows: This application provides a quadruped robot control method. The quadruped robot includes a base, a drive assembly disposed on the base, a support assembly disposed on the base, and a control assembly. The drive assembly includes at least four drive legs for driving the base to move. The support assembly includes a mounting frame, a housing, a first balancing drive component, a second balancing drive component, and a support component. The housing is rotatably disposed on the mounting frame about a first support rotation axis, and the support component is rotatably disposed on the housing about a second support rotation axis. The first support rotation axis and the second support rotation axis are not parallel. The first balancing drive component is drive-connected to the housing, and the second balancing drive component is drive-connected to the support component.

[0008] The control method includes: acquiring the posture information of the carrier or the item to be transported; determining the posture deviation of the carrier or the item to be transported relative to a preset posture based on the posture information; generating a first compensation control quantity corresponding to the first carrier rotation axis and a second compensation control quantity corresponding to the second carrier rotation axis based on the posture deviation; controlling the first balance drive to move according to the first compensation control quantity to drive the housing to rotate relative to the mounting frame around the first carrier rotation axis; controlling the second balance drive to move according to the second compensation control quantity to drive the carrier to rotate relative to the housing around the second carrier rotation axis; adjusting the posture of the carrier relative to the base through the rotation of the housing and the carrier, so that the item to be transported maintains the preset posture during the movement of the quadruped robot.

[0009] In the above technical solution, compensation control quantities in two directions are generated according to the attitude deviation, and the first balance drive and the second balance drive are controlled respectively, so that the bearing component can be adjusted in two degrees of freedom, thereby improving the bearing stability during transportation.

[0010] As a preferred embodiment, generating the first compensation control quantity and the second compensation control quantity based on the attitude deviation includes: acquiring the motion state information of the quadruped robot and the state information of the load-bearing component; inputting the motion state information and the state information of the load-bearing component into a motion-load-bearing cooperative control strategy model trained by reinforcement learning; outputting a first feedforward compensation quantity corresponding to the first load-bearing rotation axis and a second feedforward compensation quantity corresponding to the second load-bearing rotation axis through the motion-load-bearing cooperative control strategy model; generating a first feedback compensation quantity corresponding to the first load-bearing rotation axis and a second feedback compensation quantity corresponding to the second load-bearing rotation axis based on the attitude deviation; fusing the first feedforward compensation quantity and the first feedback compensation quantity to obtain the first compensation control quantity, and fusing the second feedforward compensation quantity and the second feedback compensation quantity to obtain the second compensation control quantity.

[0011] As a preferred embodiment, the motion-bearing collaborative control strategy model is trained as follows: a dynamic model including the substrate, the driving component, and the bearing component is established in a simulation training environment; rotations around the first bearing rotation axis and the second bearing rotation axis are added as controllable degrees of freedom to the motion space, and corresponding rotational state variables are added to the state space; training is performed in various motion modes or transient processes, during which the quadruped robot moves according to motion commands, and simultaneously performs attitude compensation for rotations around the first bearing rotation axis and the second bearing rotation axis; the movement control effect is evaluated based on the motion state information, and the bearing stability effect is evaluated based on the state information of the bearing component, thereby constructing a reward function for training.

[0012] As a preferred embodiment, the motion load coordinated control strategy model also outputs drive component control actions, which include at least one of the target angle, target angular velocity, target torque, or their increment for each drive foot joint.

[0013] As a preferred embodiment, when the item to be transported is a liquid container, the reward function further includes a penalty term related to at least one of the liquid sloshing amplitude and the degree of liquid surface tilt, the weight of the penalty term is adjusted according to the liquid loading rate, and the weight of the penalty term is increased during transient processes such as acceleration, deceleration, turning, or sudden stop.

[0014] As a preferred embodiment, when the item to be transported is a liquid container, the control component determines at least one of the pre-tilt angle, maximum permissible angular velocity, and maximum permissible angular acceleration of the liquid container based on at least one of the linear acceleration, angular velocity, turning radius, slope angle, and gait impact information of the quadruped robot, and controls the actions of the first balance drive and the second balance drive accordingly.

[0015] As a preferred embodiment, the quadruped robot further includes a gripping detection component, and the control method further includes: acquiring the gripping state detected by the gripping detection component; when the gripping state is abnormal, the control component outputs at least one of the following: an instruction to restrict the start of the drive component, an instruction to reduce the movement speed of the quadruped robot, and an instruction to restrict entry into obstacle-crossing mode. In the above technical solution, gripping abnormalities can be incorporated into motion control constraints, and the control component can output instructions to restrict start-up, reduce movement speed, or restrict entry into obstacle-crossing mode according to the abnormal situation, reducing the risk of high-speed movement or obstacle-crossing even when the object is not reliably secured.

[0016] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic flowchart illustrating a quadruped robot control method provided in an embodiment of this application; Figure 2 A schematic diagram of the overall structure of a quadruped robot to which the control method of this application is applicable; Figure 3 A schematic diagram of the base and drive component structure of a quadruped robot to which the control method of this application is applicable; Figure 4 A schematic diagram of the structure of a drive unit to which the control method of this application is applicable in one embodiment; Figure 5 A schematic diagram of the structure of a carrier component to which the control method of this application is applied; Figure 6 A top view of an embodiment of the carrier component to which the control method of this application is applied.

[0019] Explanation of reference numerals in the attached figures: 1. Base; 11. First driving foot; 111. Abduction drive component; 112. Connecting joint; 113. Bending drive component; 114. Upper arm; 115. Direction change drive component; 116. Forearm; 117. Travel drive component; 118. Travel wheel; 12. Second driving foot; 13. Third driving foot; 14. Fourth driving foot; 2. Bearing assembly; 21. Mounting frame; 22. Housing; 23. First balancing drive component; 24. Second balancing drive component; 25. Bearing component; 3. Control assembly; 4. Sensing assembly; 41. Load inertial sensor; 42. Body inertial sensor; 43. Clamping detection component; A. First bearing rotation axis; B. Second bearing rotation axis; C1. First foot rotation axis; C2. Second foot rotation axis; C3. Third foot rotation axis; C4. Fourth foot rotation axis. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of this application, and not all embodiments.

[0021] In the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. Unless otherwise expressly defined, the terms "connection," "transmission connection," and "electrical connection" should be interpreted broadly.

[0022] In this application, the rotation about the first bearing rotation axis A and the second bearing rotation axis B is driven by the first balance drive member 23 and the second balance drive member 24, respectively, constituting two controllable rotational degrees of freedom of the bearing assembly. Example

[0023] This embodiment provides a quadruped robot control method, applied to a quadruped robot with a two-degree-of-freedom load-bearing component, for adjusting the posture of the load-bearing component 25 during the movement of the quadruped robot, so that the item to be transported maintains a preset posture.

[0024] like Figure 2 As shown, the quadruped robot includes a base 1, a drive assembly disposed on the base 1, a support assembly 2 disposed on the base 1, and a control assembly 3. Figure 3 As shown, the driving assembly is used to support the base 1 and drive the base 1 to move, and includes at least four driving feet; in this embodiment, the four driving feet correspond to the first driving foot 11, the second driving foot 12, the third driving foot 13 and the fourth driving foot 14 respectively, and are respectively arranged at the four corners of the base 1. A body inertial sensor 42 may also be arranged on the base 1; the control assembly 3 is arranged on the base 1.

[0025] like Figure 4As shown, each drive foot can be a wheel-type drive foot. The wheel-type drive foot includes an outward drive component 111, a connecting joint 112, a sway drive component 113, a boom 114, a steering drive component 115, a forearm 116, a travel drive component 117, and a travel wheel 118. An abduction drive 111 is disposed on the base 1, and its output end is driven to the connecting joint 112 to drive the connecting joint 112 to rotate relative to the base 1 around the first foot rotation axis C1; a sway drive 113 is disposed on the connecting joint 112, and its output end is driven to one end of the upper arm 114 to drive the upper arm 114 to rotate relative to the connecting joint 112 around the second foot rotation axis C2; a reversing drive 115 is disposed on the other end of the upper arm 114, and its output end is driven to one end of the forearm 116 to drive the forearm 116 to rotate relative to the upper arm 114 around the third foot rotation axis C3; a traveling drive 117 is disposed on the other end of the forearm 116, and its output end is driven to the traveling wheel 118 to drive the traveling wheel 118 to rotate around the fourth foot rotation axis C4. Preferably, the first foot rotation axis C1 is perpendicular or approximately perpendicular to the second foot rotation axis C2, and the second foot rotation axis C2, the third foot rotation axis C3, and the fourth foot rotation axis C4 are parallel or approximately parallel. Thus, the abduction drive 111 can realize the abduction or retraction of the driving foot relative to the base 1, the flexion drive 113 can realize the lifting or pressing down of the upper arm 114, the directional drive 115 can realize the swinging of the forearm 116 relative to the upper arm 114, and the travel drive 117 can realize the rotation of the travel wheel 118. The driving foot joints in this application correspond to the controllable rotational joints around the first foot rotation axis C1, the second foot rotation axis C2, the third foot rotation axis C3, and the fourth foot rotation axis C4; the target values ​​of each driving foot joint in the control action of the drive assembly can act on at least one of the aforementioned joints.

[0026] like Figure 5 and Figure 6 As shown, the load-bearing assembly 2 includes a mounting frame 21, a housing 22, a first balancing drive component 23, a second balancing drive component 24, and a load-bearing component 25. The housing 22 is rotatably mounted on the mounting frame 21 about a first load-bearing rotation axis A, and the load-bearing component 25 is rotatably mounted on the housing 22 about a second load-bearing rotation axis B. The first load-bearing rotation axis A and the second load-bearing rotation axis B are not parallel, but preferably perpendicular to each other. The first balancing drive component 23 is drive-connected to the housing 22, and the second balancing drive component 24 is drive-connected to the load-bearing component 25. A clamping detection component 43 and a load inertial sensor 41 can be mounted on the load-bearing component 25. The control assembly 3 is electrically connected to the drive assembly, the first balancing drive component 23, the second balancing drive component 24, and the sensing assembly 4, respectively.

[0027] like Figure 1As shown, the control method of this embodiment includes the following steps: acquiring the attitude information of the carrier 25 or the item to be transported through the load inertial sensor 41. The attitude information may include at least one of tilt angle, pitch angle, and roll angle. The load inertial sensor 41 can also measure angular velocity and acceleration to assist in estimating the above attitude information. The attitude information of the base 1 can also be acquired through the body inertial sensor 42 as a reference for attitude compensation.

[0028] The control component 3 compares the acquired posture information with the preset posture to obtain the posture deviation of the carrier 25 or the item to be transported relative to the preset posture. The preset posture can be the posture in which the carrier 25 is kept horizontal, the posture in which the item to be transported has its opening facing upward, or other postures preset according to the type of item to be transported.

[0029] Based on the attitude deviation, the control component 3 generates a first compensation control quantity corresponding to the first load-bearing rotation axis A and a second compensation control quantity corresponding to the second load-bearing rotation axis B.

[0030] The control component 3 controls the first balancing drive component 23 to operate according to the first compensation control amount, driving the housing 22 to rotate relative to the mounting bracket 21 around the first bearing rotation axis A; and controls the second balancing drive component 24 to operate according to the second compensation control amount, driving the bearing component 25 to rotate relative to the housing 22 around the second bearing rotation axis B.

[0031] By rotating the housing 22 and the carrier 25, the posture of the carrier 25 relative to the base 1 is adjusted, so that the item to be transported maintains a preset posture during the movement of the quadruped robot. When the posture of the base 1 changes during movement, turning, climbing, obstacle crossing, starting and stopping, or passing through uneven ground, the carrier 25 can still reduce the swaying or tilting of the item to be transported through two degrees of freedom adjustment.

[0032] When the item to be transported is a liquid container, the preset posture can be such that the opening of the liquid container faces a preset direction, for example, the opening faces upwards. When the base 1 pitches, rolls, vibrates, or accelerates or decelerates, the control component 3 controls the first balancing drive 23 and the second balancing drive 24 to move, causing the carrier 25 to rotate relative to the base 1 to compensate for liquid sloshing or spillage. In this embodiment, the active posture compensation of the carrier 25 can be achieved through the cooperation of the load inertial sensor 41, the control component 3, the first balancing drive 23, and the second balancing drive 24, which is suitable for scenarios requiring stable load-bearing, such as coffee delivery, beverage delivery, and reagent handling. Example

[0033] This embodiment provides a quadruped robot control method applied to a quadruped robot with a two-degree-of-freedom load-bearing component. Unlike schemes that only train the motion control of the driving legs, this embodiment simultaneously incorporates the rotational states and control actions around the first load-bearing rotation axis A and the second load-bearing rotation axis B during control and training. The control component 3 acquires motion state information and the state information of the load-bearing component, inputs it into the motion-load-bearing cooperative control strategy model to obtain feedforward compensation amounts corresponding to the two load-bearing rotation axes; simultaneously, it determines the posture deviation based on the posture information of the load-bearing component 25 or the item to be transported, generates feedback compensation amounts, and fuses them with the feedforward compensation amounts to control the first balancing drive component 23 and the second balancing drive component 24. The strategy model can also synchronously output the control actions of the drive components.

[0034] The motion state information may include at least one of the following: base posture, motion command, driving foot joint position, driving foot joint velocity, and foot contact state. The base posture, which falls under base orientation information, may include at least one of base pitch angle and roll angle.

[0035] The status information of the load-bearing component may include at least one of the angle and angular velocity about the first load-bearing rotation axis A, and the angle and angular velocity about the second load-bearing rotation axis B.

[0036] The attitude information of the carrier or the item to be transported may include at least one of the pitch angle and roll angle relative to the horizontal plane or the preset ground plane; the corresponding attitude deviation is at least one of the pitch deviation and roll deviation of the above attitude relative to the preset attitude.

[0037] The motion-bearing collaborative control strategy model can employ a policy network, an Actor-Critic network, or a deep reinforcement learning strategy model. Its output, the control action of the driving components, can be the target angle, target angular velocity, target torque, or their increment for each driving foot joint. The driving foot joints include at least one of the following: an abduction joint about a first foot rotation axis C1, a flexion joint about a second foot rotation axis C2, a directional joint about a third foot rotation axis C3, and a travel wheel drive joint about a fourth foot rotation axis C4. The first feedforward compensation amount and the second feedforward compensation amount correspond to the first load-bearing rotation axis A and the second load-bearing rotation axis B, respectively, and can be the target angle, target angular velocity, target torque, or their increment around the corresponding load-bearing rotation axis, which are different from the control amounts of the driving foot joints.

[0038] In one embodiment, the control component 3 can also generate a first feedback compensation amount and a second feedback compensation amount based on the attitude deviation measured by the inertial sensor 41, and fuse them with the first feedforward compensation amount and the second feedforward compensation amount output by the strategy model to obtain the first compensation control amount and the second compensation control amount, and then send them to the first balance drive 23 and the second balance drive 24.

[0039] During training, in a simulation environment that includes quadruped robot dynamics, load-bearing component dynamics, and the swaying characteristics of the transported object, rotations around the first load-bearing rotation axis A and the second load-bearing rotation axis B are added as controllable degrees of freedom to the motion space, and the corresponding angles and angular velocities are added to the state space. The terrain can include flat ground, slopes, steps, obstacles, and uneven ground, and can be superimposed with acceleration, deceleration, turning, sudden stops, as well as changes in the load's mass, center of gravity, and moment of inertia.

[0040] The reward function can simultaneously evaluate the motion control effect based on motion state information and the load-bearing stability effect based on the state information of the load-bearing components. Motion-related items include speed command tracking, base attitude stability, foot contact stability, joint motion smoothness, and energy consumption; load-bearing stability-related items include pitch and roll deviations of the load-bearing components or the transported item, and the rate of change of compensated motion around the two load-bearing rotation axes. When the transported item is a liquid container, the reward function may also include penalty items related to the amplitude of liquid sloshing and the degree of liquid surface tilt; the weight of the penalty item can be adjusted according to the liquid loading rate. The higher the loading rate, the greater the risk of spillage under the same sloshing or tilting, and the stronger the penalty accordingly; and the weight of this penalty item is further increased during transient processes such as acceleration, deceleration, turning, or sudden stopping.

[0041] After deployment, the control component 3 inputs the motion state information of the current motion mode or transient process and the state information of the load-bearing component into the motion load-bearing collaborative control strategy model. The model then synchronously provides the control actions of the drive component and the first and second feedforward compensation amounts. The output tendencies of the drive component and the first and second balancing drive components 23 and 24 under different motion modes are shown in Table 1; the output tendencies under different transient processes are shown in Table 2.

[0042] Table 1. Output Tendencies of Policy Models under Different Motion Modes

[0043] Table 2. Output Tendencies of Strategy Models under Different Motion Modes

[0044] When the item to be transported is a liquid container, the control component 3 can also determine at least one of the pre-tilt angle, maximum permissible angular velocity, and maximum permissible angular acceleration of the liquid container based on at least one of the linear acceleration, angular velocity, turning radius, slope angle, and gait impact information of the quadruped robot, and control the actions of the first balance drive 23 and the second balance drive 24 accordingly. The liquid loading rate can be calculated from the liquid level height relative to the container volume, or estimated from the load weight combined with the container volume; when the loading rate is high, the maximum permissible angular velocity and maximum permissible angular acceleration can be further reduced, or the pre-tilt angle can be adjusted. Among them, the gait impact information is used to characterize the impact strength on the base and bearing components when the foot lands or gait changes, and can include at least one of the foot impact force, landing impact acceleration, and ground contact vibration peak value. The foot impact force can be estimated by the foot force sensor or by the contact state between the joint torque and the foot; the landing impact acceleration can be measured by the acceleration near the base or foot; the ground contact vibration peak value can be characterized by the peak value of the vibration amplitude measured by the inertial sensor during the ground contact period. When the gait impact is strong, the control component 3 can reduce the maximum permissible angular velocity and the maximum permissible angular acceleration, or adjust the pretilt angle, to avoid the superposition of impacts causing violent shaking of the liquid surface.

[0045] If a clamping detection element 43 is provided, the control component 3 acquires the clamping state detected by the clamping detection element 43 and determines whether the clamping state is abnormal. The clamping detection element 43 is used to detect the clamping force or the position of the clamping mechanism. The corresponding abnormal states include not clamping in place and insufficient clamping force. Not clamping in place corresponds to a clamping force of zero or the clamping mechanism not reaching the preset clamping position; insufficient clamping force corresponds to the clamping mechanism being in place but the clamping force being lower than a preset threshold.

[0046] When the clamping state is abnormal, control component 3 outputs the corresponding control command, as follows: When it is determined that the gripper is not in place, the control component 3 outputs a command to restrict the start of the drive component, prohibiting the quadruped robot from starting or cutting off the driving enable of the drive component until the gripper is in place. When the clamping force is determined to be insufficient, the control component 3 outputs a command to reduce the moving speed and a command to restrict entry into the obstacle crossing mode, so that the quadruped robot is only allowed to move at a low speed on flat ground, avoiding entering the legged obstacle crossing or going up and down steps when the clamping is unreliable.

[0047] The aforementioned control commands can be output individually or in combination depending on the severity of the anomaly. This embodiment incorporates drive foot control and compensation control around the two load-bearing rotation axes into the same strategy. Compared to methods that rely solely on attitude deviation feedback, this approach makes it easier to provide load compensation in the early stages of disturbances, reducing the risk of swaying, damage, or liquid spillage during transportation in complex terrain.

[0048] This embodiment illustrates a training and deployment process for a motion load collaborative control strategy model.

[0049] In a simulation training environment, an overall dynamic model of the substrate, drive components, load-bearing components, and the object to be transported is established. The load-bearing component dynamics describes the rotational response of the shell 22 and the load-bearing component 25 about the first load-bearing rotation axis A and the second load-bearing rotation axis B. The object to be transported model describes the attitude changes of solid loads, or the sloshing, tilting, and liquid loading rates within a liquid container, and can randomize the load's mass, center of gravity, and moment of inertia.

[0050] In addition to the base posture, motion command, driving foot joint position, driving foot joint velocity, and foot contact state, the state space includes at least one of the angles and angular velocities about the first load-bearing rotation axis A and the second load-bearing rotation axis B. In addition to the driving foot joint motion, the action space includes the target angle, target angular velocity, target torque, or their increments about the two load-bearing rotation axes.

[0051] Under conditions including flat ground, slopes, steps, obstacles, and uneven ground, the quadruped robot is trained using various movement modes such as flat ground walking, legged obstacle crossing, uphill and downhill walking, and uphill and downhill walking, as well as transient processes such as acceleration, deceleration, turning, and sudden stopping. This trains the quadruped robot to move according to motion commands and simultaneously trains compensation movements around two load-bearing rotation axes, ensuring that the pitch and roll deviations of the load-bearing components relative to the horizontal plane or a preset ground plane remain within preset ranges. The reward function evaluates the walking control effect based on motion state information and the load-bearing stability effect based on the state information of the load-bearing components; specifically, this includes motion command tracking, base posture stability, foot contact stability, and the pitch and roll deviations of the load-bearing components or the transported items, as well as the smoothness of compensation movements. Training algorithms can employ methods such as policy gradient, Actor-Critic, or proximal policy optimization.

[0052] After training, the strategy model is deployed on control component 3. During runtime, based on the motion state information acquired in real time by the sensors and the state information of the load-bearing components, the system outputs control actions for the drive components, as well as first and second feedforward compensation amounts corresponding to the two load-bearing rotation axes. After fusing with the attitude deviation feedback, the system controls the first and second balance drive components 23 and 24. The output tendencies for each motion mode are shown in Table 1, and the output tendencies for each transient process are shown in Table 2. After fusion, these tendencies are applied to the drive components and the first and second balance drive components 23 and 24, respectively.

[0053] The above embodiments are merely preferred embodiments of this application and are not intended to limit this application. Modifications, equivalent substitutions, and improvements made by those skilled in the art within the spirit and principles of this application should be included within the protection scope of this application. Where there is no conflict, the features of the various embodiments of this application can be combined with each other.

Claims

1. A control method for a quadruped robot, characterized in that, The quadruped robot includes a base, a drive assembly disposed on the base, a support assembly disposed on the base, and a control assembly. The drive assembly includes at least four drive legs for moving the base. The support assembly includes a mounting frame, a housing, a first balancing drive component, a second balancing drive component, and a support component. The housing is rotatably disposed on the mounting frame about a first support rotation axis, and the support component is rotatably disposed on the housing about a second support rotation axis. The first support rotation axis and the second support rotation axis are not parallel. The first balancing drive component is drivenly connected to the housing, and the second balancing drive component is drivenly connected to the support component. The control method includes: Obtain the attitude information of the carrier or the item to be transported; The attitude deviation of the carrier or the item to be transported relative to the preset attitude is determined based on the attitude information; Based on the attitude deviation, a first compensation control quantity corresponding to the first load-bearing rotation axis and a second compensation control quantity corresponding to the second load-bearing rotation axis are generated. The first balance drive is controlled to operate according to the first compensation control amount, so as to drive the housing to rotate relative to the mounting bracket about the first bearing rotation axis; The second balance drive is controlled to operate according to the second compensation control amount, so as to drive the bearing member to rotate relative to the housing about the second bearing rotation axis; The orientation of the carrier relative to the base is adjusted by rotating the shell and the carrier, so that the item to be transported maintains the preset orientation during the movement of the quadruped robot.

2. The quadruped robot control method according to claim 1, characterized in that, The step of generating the first compensation control quantity and the second compensation control quantity based on the attitude deviation includes: Acquire the motion state information of the quadruped robot and the state information of the supporting component; The motion state information and the state information of the load-bearing component are input into the motion load-coordinated control strategy model obtained through reinforcement learning training. The motion load cooperative control strategy model outputs a first feedforward compensation amount corresponding to the first load rotation axis and a second feedforward compensation amount corresponding to the second load rotation axis; Based on the attitude deviation, a first feedback compensation amount corresponding to the first load-bearing rotation axis and a second feedback compensation amount corresponding to the second load-bearing rotation axis are generated. The first feedforward compensation amount and the first feedback compensation amount are fused to obtain the first compensation control amount, and the second feedforward compensation amount and the second feedback compensation amount are fused to obtain the second compensation control amount. In the above technical solution, the strategy model output is used as the feedforward compensation amount, which is fused with the feedback compensation amount based on attitude deviation to form the final compensation control amount. This can both respond to motion disturbances in advance and suppress residual deviations.

3. The quadruped robot control method according to claim 2, characterized in that, The motion load cooperative control strategy model is trained in the following manner: A dynamic model including the substrate, the driving component, and the load-bearing component is established in a simulation training environment; The rotation around the first load-bearing rotation axis and the second load-bearing rotation axis is added as a controllable degree of freedom to the motion space, and the corresponding rotation state quantity is added to the state space. Training is conducted in various motion modes or transient processes. During training, the quadruped robot moves according to motion commands, and posture compensation is performed on the rotation around the first load-bearing rotation axis and the second load-bearing rotation axis. The effectiveness of the travel control is evaluated based on the motion state information, and the stability of the load-bearing components is evaluated based on the state information. A reward function is then constructed and trained accordingly. In this technical solution, the training covers multiple motion modes or typical transient processes, which helps the strategy adapt to different travel disturbances.

4. The quadruped robot control method according to claim 2 or 3, characterized in that, The motion-bearing coordinated control strategy model also outputs drive component control actions, which include at least one of the target angle, target angular velocity, target torque, or their increment for each drive foot joint. In the above technical solution, the drive component control actions act on each drive foot joint and are distinct from the feedforward compensation amounts corresponding to the two load-bearing rotation axes.

5. The quadruped robot control method according to claim 3, characterized in that, When the item to be transported is a liquid container, the reward function also includes a penalty term related to at least one of the liquid sloshing amplitude and the degree of liquid surface tilt. The weight of the penalty term is adjusted according to the liquid loading rate, and the weight of the penalty term is increased during transient processes such as acceleration, deceleration, turning, or sudden stopping. In the above technical solution, a liquid surface-related constraint is added to the reward function during liquid transportation, and the penalty weight is adjusted by the loading rate, with the penalty weight further increased during transient processes.

6. The quadruped robot control method according to claim 2, characterized in that, When the item to be transported is a liquid container, the control component determines at least one of the pre-tilt angle, maximum permissible angular velocity, and maximum permissible angular acceleration of the liquid container based on at least one of the quadruped robot's linear acceleration, angular velocity, turning radius, slope angle, and gait impact information, and controls the actions of the first and second balancing drive components accordingly. In the above technical solution, inertia, slope turning information, and gait impact information can be combined to apply motion constraints adapted to liquid transportation to the two balancing drive components.

7. The quadruped robot control method according to claim 2, characterized in that, The quadruped robot also includes a gripping detection component, and the control method further includes: Obtain the clamping state detected by the clamping detection component; When the clamping state is abnormal, the control component outputs at least one of the following commands: restricting the start of the drive component, reducing the movement speed of the quadruped robot, and restricting entry into obstacle-crossing mode. In the above technical solution, clamping abnormalities can be incorporated into motion control constraints. The control component outputs commands to restrict start-up, reduce movement speed, or restrict entry into obstacle-crossing mode according to the abnormal situation, reducing the risk of high-speed movement or obstacle crossing even when the item is not reliably secured. Other features and advantages of this application will be described in detail in the following detailed description section.

Images