Balance control methods, devices, equipment and storage media for wheeled-legged robots

By abstracting the wheeled-legged robot into a multi-stage inverted pendulum model and using sliding mode control technology to calculate the torque of the rotating joints, the problem of inflexible posture changes in the balance control of the wheeled-legged robot is solved, and its balance recovery ability and robustness under various disturbances are improved.

CN119322514BActive Publication Date: 2026-11-17TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310876039.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-07-14
Publication Date
2026-11-17
Estimated Expiration
2043-07-14

AI Technical Summary

Technical Problem

Existing wheeled robots have poor flexibility in posture changes during balance control, making it difficult to maintain balance under various disturbances and lacking robustness.

Method used

The wheeled robot is abstracted into a multi-stage inverted pendulum model. By acquiring state variables, establishing sliding surfaces, calculating the rotational torque of the rotating joints, and controlling the attitude adjustment of multiple rotating joints, balance control is achieved.

Benefits of technology

It enhances the flexibility of robot posture, improves the ability to recover balance under different disturbances, and improves the robustness of balance control.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119322514B_ABST
    Figure CN119322514B_ABST
Patent Text Reader

Abstract

A balance control method, apparatus, device, and storage medium for a wheeled-legged robot are disclosed, relating to the field of robotics. The wheeled-legged robot includes: a movable wheel, n links, and n rotary joints. The first link of the movable wheel and n links is connected via a first rotary joint of the n rotary joints. The n links are connected in series via n-1 rotary joints (excluding the first rotary joint), where n is a positive integer greater than or equal to 2. The method includes: acquiring the state variables of the wheeled-legged robot at a first moment; determining the dynamic model parameters based on the dynamic equations of the wheeled-legged robot and the state variables at the first moment; establishing a sliding surface based on the state variables at the first moment; calculating the rotational torques of the n rotary joints based on the sliding surface and the dynamic model parameters; and controlling the rotary joints based on the rotational torques of the n rotary joints. This makes the attitude of the n links in the robot adjustable during balance adjustment, improving the robustness of the balance adjustment process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a balance control method, device, equipment and storage medium for a wheeled robot. Background Technology

[0002] With the continuous development of robotics technology, a wheel-legged robot has emerged, in which the moving wheels and leg mechanism are connected by joints. This wheel-legged robot combines the mobility advantages of both wheeled and legged robots.

[0003] In related technologies, in order to improve the balance of wheeled robots during movement, the leg mechanism and torso mechanism of the wheeled robot are regarded as a whole. The torque of the joint motor between the moving wheel and the leg mechanism is calculated, and the angle between the moving wheel and the leg mechanism is changed by rotating the joint motor to maintain the balance of the wheeled robot.

[0004] However, in the balance control process of related technologies, the posture change flexibility of wheeled-legged robots is poor, making it difficult to maintain balance under various disturbances. The robustness of balance control for wheeled-legged robots needs to be further improved. Summary of the Invention

[0005] This application provides a method, apparatus, device, and storage medium for balancing a wheeled-legged robot. The technical solution is as follows:

[0006] According to one aspect of the embodiments of this application, a balance control method for a wheeled-legged robot is provided. The wheeled-legged robot includes: a movable wheel, n links, and n rotary joints. The movable wheel and a first link among the n links are connected through a first rotary joint among the n rotary joints. The n links are connected in series through n-1 rotary joints other than the first rotary joint, where n is a positive integer greater than or equal to 2. The method includes:

[0007] The state quantity of the wheeled-legged robot at a first moment is obtained, and the state quantity at the first moment is used to characterize the motion state of the wheeled-legged robot at the first moment.

[0008] Based on the dynamic equations of the wheel-legged robot and the state variables at the first moment, dynamic model parameters are determined. These dynamic model parameters define the mapping relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The angular acceleration at the first moment includes the angular acceleration of the n links and the angular acceleration of the moving wheels. The rotational torque at the second moment includes the rotational torque of the n rotating joints.

[0009] Based on the state variables at the first moment, a sliding surface is established, and the state variables of the wheeled robot gradually approach 0 on the sliding surface;

[0010] The rotational torques of the n rotating joints are calculated based on the sliding surface and the dynamic model parameters.

[0011] At the second moment, the n rotary joints are controlled according to their rotational torques.

[0012] According to one aspect of the embodiments of this application, a balance control device for a wheeled-legged robot is provided. The wheeled-legged robot includes: a movable wheel, n links, and n rotating joints. The movable wheel and a first link among the n links are connected via a first rotating joint among the n rotating joints. The n links are connected in series via n-1 rotating joints other than the first rotating joint, where n is a positive integer greater than or equal to 2. The device includes:

[0013] The state acquisition module is used to acquire the state quantity of the wheeled-legged robot at a first moment, and the state quantity at the first moment is used to characterize the motion state of the wheeled-legged robot at the first moment.

[0014] The parameter determination module is used to determine the dynamic model parameters based on the dynamic equations of the wheel-legged robot and the state variables at the first moment. The dynamic model parameters are used to define the mapping relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The angular acceleration at the first moment includes the angular acceleration of the n links and the angular acceleration of the moving wheels. The rotational torque at the second moment includes the rotational torque of the n rotating joints.

[0015] The sliding surface establishment module is used to establish a sliding surface based on the state variables at the first moment, wherein the state variables of the wheeled robot gradually approach 0 on the sliding surface;

[0016] The torque calculation module is used to calculate the rotational torque of the n rotating joints based on the sliding surface and the dynamic model parameters.

[0017] The joint rotation module is used to control the n rotating joints at the second moment based on the rotational torque of the n rotating joints.

[0018] According to one aspect of the embodiments of this application, a computer device is provided, the computer device including a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the above-described balance control method for a wheeled legged robot.

[0019] According to one aspect of the present application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the above-described balance control method for a wheeled legged robot.

[0020] According to one aspect of the embodiments of this application, a computer program product is provided, the computer program product including a computer program, the computer program being loaded and executed by a processor to implement the above-described balance control method for a wheeled robot.

[0021] The technical solution provided in this application can bring the following beneficial effects:

[0022] In summary, by abstracting the wheel-legged robot into a multi-stage inverted pendulum model, the deflection angles of multiple links in the wheel-legged robot can be adjusted during balance control, thereby changing the robot's posture. Compared to related technologies that can only adjust the angle between the moving wheels and leg links during balance control, this solution allows the posture of multiple links in the wheel-legged robot to be variable during balance adjustment, greatly enriching the robot's posture, facilitating rapid adjustment to a balanced state, and improving the robot's ability to recover to a balanced state under different disturbances, thus enhancing the robustness of the robot's balance control process. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the implementation environment of a solution provided in one embodiment of this application;

[0024] Figure 2 This is a schematic diagram of a quadrupedal wheeled robot provided in one embodiment of this application;

[0025] Figure 3 This is a schematic diagram of a quadrupedal wheeled robot provided in one embodiment of this application;

[0026] Figure 4 This is a schematic diagram of a quadrupedal wheeled robot in dual-wheel mode according to an embodiment of this application;

[0027] Figure 5 This is a schematic diagram of a quadrupedal wheeled robot climbing stairs according to an embodiment of this application;

[0028] Figure 6 This is a schematic diagram of the planar motion of a robot climbing stairs according to one embodiment of this application;

[0029] Figure 7 This is a schematic diagram of a robot's action sequence for going up and down stairs according to an embodiment of this application;

[0030] Figure 8This is a block diagram of a robot control system provided in one embodiment of this application;

[0031] Figure 9 This is a schematic diagram of a second-order inverted pendulum provided in one embodiment of this application;

[0032] Figure 10 This is a schematic diagram of an n-order inverted pendulum provided in one embodiment of this application;

[0033] Figure 11 This is a flowchart of a balance control method for a wheeled-legged robot provided in one embodiment of this application;

[0034] Figure 12 This is a schematic diagram of an equivalent process provided in one embodiment of this application;

[0035] Figure 13 This is a schematic diagram of an inverted pendulum with a center of mass related to an exemplary embodiment of this application;

[0036] Figure 14 This is a simulation diagram of a dual-wheel balance control method provided in an exemplary embodiment of this application;

[0037] Figure 15 This is a simulation diagram of a dual-wheel balance control method provided in another exemplary embodiment of this application;

[0038] Figure 16 This is a block diagram of a balance control device for a wheeled-legged robot provided in one embodiment of this application;

[0039] Figure 17 This is a structural block diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0042] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0043] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0044] This application's technical solution primarily relates to robotics within artificial intelligence, specifically intelligent robot control. A robot is a mechatronic electronic device that combines mechanical transmission and modern microelectronics technology to mimic certain human skills. Robots have evolved based on electronic, mechanical, and information technologies. A robot doesn't necessarily have to resemble a human; as long as it can autonomously complete tasks and commands assigned by humans, it belongs to the robot family. A robot is an automated machine possessing some intelligent abilities similar to humans or other living organisms, such as perception, planning, movement, and coordination. It is a highly flexible automated machine. With the development of computer and artificial intelligence technologies, robots have seen significant improvements in both functionality and technology. Mobile robots and technologies such as robot vision and touch are typical examples.

[0045] The technical solutions provided in this application embodiment can be executed by computer equipment, which refers to electronic devices with data computing, processing and storage capabilities.

[0046] Optionally, the computer device can be a PC (Personal Computer) device such as a desktop computer or laptop computer used to control the robot; or it can be a server used to control the robot. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The computer device and the robot can be connected via physical lines, networks, etc. For example, see reference... Figure 1The computer device 101 can calculate the rotational torque of the first rotational joint (equivalent to the ankle joint) connecting the moving wheel and the leg mechanism (corresponding to the first link) based on the current state of the robot 103, as well as the rotational torque of the other n-1 rotational joints that can be abstracted into an nth-order inverted pendulum model with the first link and the moving wheel. The robot 103's posture is adjusted by the rotational torque of the n rotational joints so that the robot 103 is in a balanced state.

[0047] Optionally, the computer device can also be the robot itself, that is, the entity executing each step in the technical solution provided in this application is a robot. For example, refer to Figure 1 The computer device 101 can send the state of the robot 103 at the first moment to the robot 103 through the network 102. The robot 103 calculates the torque of the n rotating joints based on the state of the first moment, and controls the rotation motors of the n rotating joints to adjust the posture of the robot 103 so that the robot 103 maintains a balanced state.

[0048] The robot in this application embodiment may refer to a wheel-legged robot, a legged robot, etc. A wheel-legged robot is a robot that combines a wheeled structure and legged capabilities, and a legged robot is a robot that uses feet as its legs. This application embodiment does not limit the specific robot to these terms.

[0049] In one example, robot 103 may include at least one movable wheel, and for each of the at least one movable wheel, the movable wheel is connected to its corresponding leg mechanism via a first rotary joint.

[0050] The wheeled robot in this embodiment includes a movable wheel and at least two links, each of which can be equivalent to the links in an n-order inverted pendulum model; that is, the wheeled robot includes a movable wheel and n links, which are connected in series through n-1 rotary joints, and the first link in the n mechanisms is connected to the movable wheel through a first rotary joint.

[0051] The wheeled robot includes at least one leg mechanism, and each leg mechanism has a corresponding moving wheel. For example, leg mechanism 1 corresponds to moving wheel 1, and leg mechanism 1 is connected to moving wheel 1 through a first rotary joint 1.

[0052] Optionally, at least one leg mechanism may be divided into an outer leg group and an inner leg group, wherein the outer leg group includes at least two leg mechanisms and the inner leg group includes at least one leg mechanism.

[0053] For example, a wheeled-legged robot will be used to illustrate the robot in the embodiments of this application.

[0054] At least one leg mechanism is divided into: an outer leg mechanism and an inner leg mechanism; wherein, the outer leg mechanism includes: at least two leg mechanisms, and the inner leg mechanism may also include at least two leg mechanisms, wherein the outer leg mechanism includes at least two leg mechanisms located on both sides of the central axis (i.e., sagittal plane) of the wheel-legged robot, and the inner leg mechanism includes at least two leg mechanisms located on both sides of the central axis of the wheel-legged robot, wherein the rotation center of the hip joint corresponding to the outer leg mechanism and the inner leg mechanism, and the rotation center of the hip joint corresponding to the inner leg mechanism, are located in the same vertical plane.

[0055] For example, the wheel-legged robot is a quadrupedal wheel-legged robot, meaning it includes two outer leg mechanisms and two inner leg mechanisms; it can also be a tripedal wheel-legged robot, meaning it includes two outer leg mechanisms and one inner leg mechanism. At least one hip joint of the inner leg mechanism group is located between the hip joints of the two outer leg mechanisms in the outer leg mechanism group, and the rotation centers of the hip joints corresponding to the outer and inner leg mechanisms are located in the same vertical plane. The wheel-legged robot can stand on a support surface using either the outer or inner leg mechanisms. It can also glide on the support surface using wheels on either the outer or inner leg mechanisms, and can move (i.e., walk) by alternately swinging the outer and inner leg mechanism groups.

[0056] For example, such as Figure 2 As shown, an exemplary schematic diagram of a quadrupedal wheeled robot is illustrated. The quadrupedal wheeled robot 200 may include: a torso structure, leg mechanisms, and wheels.

[0057] The quadrupedal wheeled robot 200 has four leg mechanisms: two outer leg mechanisms 201 (i.e., the first leg mechanism group) and two inner leg mechanisms 202 (i.e., the second leg mechanism group). The two inner leg mechanisms 202 are located between the two outer leg mechanisms 201. Each of the four leg mechanisms can extend and retract independently in the direction shown in the figure. Each of the four leg mechanisms has a wheel 203 installed at its end, and each wheel 203 can be driven independently. The quadrupedal wheeled robot 200 can stand on either the two inner leg mechanisms 202 or the two outer leg mechanisms 201 to be in a bipedal wheel-supported state; the quadrupedal wheeled robot 200 can also stand on both the two inner leg mechanisms 202 and the two outer leg mechanisms 201 simultaneously to be in a quadrupedal wheel-supported state, but this embodiment does not limit this.

[0058] Alternatively, the two inner leg mechanisms 202 can be implemented as a single unit, that is, the quadrupedal wheeled robot 200 can be implemented as a tripedal wheeled robot with only one inner leg mechanism.

[0059] Each of the four leg mechanisms is connected to a hip joint 204 at its other end. Each leg mechanism can rotate around its respective hip joint 204 and maintain linkage. In this embodiment, the rotation centers of the hip joints 204 corresponding to the quadrupedal wheeled robot 200 are located in the same vertical plane 205, and the rotation planes of the leg mechanisms corresponding to the quadrupedal wheeled robot 200 are parallel. The hip joints 204 corresponding to the two inner leg mechanisms 202 are located between the hip joints 204 corresponding to the two outer leg mechanisms 201.

[0060] Optionally, the hip joints 204 of the quadrupedal wheeled robot 200 can be coaxial, meaning the rotation centers of each hip joint 204 are located on the same straight line. Alternatively, the hip joints 204 of the quadrupedal wheeled robot 200 can be non-coaxial. For example, the hip joints 204 corresponding to the two inner leg mechanisms 202 can be coaxial, and the hip joints 204 corresponding to the two outer leg mechanisms 201 can be coaxial, but the hip joints 204 corresponding to the two inner leg mechanisms 202 are not coaxial with the hip joints 204 corresponding to the two outer leg mechanisms 201. The hip joint 204 can serve as one of the n rotating joints in this embodiment, connected to the first link. One end of the first link is connected to the first rotating joint, and the other end is connected to the hip joint 204.

[0061] Optionally, the hip joints 204 corresponding to the two outer leg mechanisms 201 share a single drive motor to enable synchronous movement of the two outer leg mechanisms 201; similarly, the hip joints 204 corresponding to the two inner leg mechanisms 202 share a single drive motor to enable synchronous movement of the two inner leg mechanisms 202. In a feasible example, each hip joint 204 of the quadrupedal wheel-legged robot 200 may also be driven independently by its respective drive motor; this embodiment does not limit this.

[0062] The body of the quadrupedal wheeled robot 200 may include a waist 206, a torso 207, an upper limb 208, and a head 209.

[0063] In this quadrupedal wheeled robot 200, each hip joint 204 is connected to the same end of the waist 206, and the other end of the waist 206 is connected to one end of the torso (torso mechanism) 207. The waist 206 has two rotation centers: a pitch rotation center that enables the torso 207 to pitch, and a lateral rotation center that enables the torso 207 to sway. The lateral rotation center and the pitch rotation center are designed in series and are located above the pitch rotation center, connected to the torso 207.

[0064] For example, if Figure 2 If the body 207 rotates 90 degrees along the lateral rotation center, the robot will then... Figure 2 Switch to Figure 3 The robot shown.

[0065] The other end of the torso 207 is connected to the upper limb (robotic arm link) 208 and the head 209. The upper limb 208 can be a multi-degree-of-freedom upper limb. In some embodiments, an end effector, such as a robotic gripper or a suction cup, is deployed on the upper limb 208. Data acquisition devices, such as image acquisition devices, video recording devices, and IMUs (Inertial Measurement Units), can be deployed in the head 209 to perceive the real environment. The IMU can be placed at the geometric center of the torso 207, the center point of the hip joint (i.e., the rotation center of the hip joint), etc., and can be used to measure the actual acceleration, actual angular velocity, and actual Euler angles of the torso 207.

[0066] In the technical solution provided in the embodiments of this application, the moving wheels, leg mechanism, hip joint, waist and torso mechanism (including IMU) of the quadruped wheeled robot 200 are necessary hardware for the control algorithm, and the rest are not necessary hardware.

[0067] Compared to bipedal wheeled robots, quadrupedal wheeled robots have a more stable structure and stronger resistance to external impacts and disturbances. Compared to hexapedal wheeled robots, they have fewer redundant joints, lower design complexity, and can bear heavy loads, move through narrow spaces, and perform tasks on objects of different heights. Quadrupedal wheeled robots have a strong adaptability to the environment.

[0068] The balance control method for wheeled-legged robots provided in this application is applicable to various scenarios, such as balancing wheeled-legged robots during movement. This method allows for flexible control of multiple joints within the wheeled-legged robot during balance control, enriching the robot's posture and facilitating rapid balancing.

[0069] When moving on flat ground, the robot can maintain the following posture: Figure 2 The four-wheel motion pattern shown can also be formed through the rotation center of the hip joint, as follows: Figure 4The diagram illustrates a two-wheeled, dynamically smooth movement mode. In four-wheeled mode, the robot remains in a stable state (does not tip over), facilitating the upper limbs to follow commands and perform tasks. However, when moving on flat ground, switching from four-wheeled to two-wheeled mode reduces the robot's footprint, making it more compatible with bipedal robots. For movement on uneven ground, such as climbing stairs or stepping over thresholds, the robot can... Figure 5 The two wheels alternate in a dynamic obstacle-crossing pattern, as shown.

[0070] In some embodiments, reference Figure 5 When the wheeled robot 501 needs to climb stairs, it can first construct an objective function corresponding to the stair-climbing task. Then, based on the joint torque information of the robot 501 when the objective function is minimized, it controls the outer leg mechanism group 502 (i.e., the first leg mechanism group) and the inner leg mechanism group 503 (i.e., the second leg mechanism group) of the robot 501 to swing alternately to perform the stair-climbing task. For example, the outer leg mechanism group 502 is used as the supporting leg mechanism group and the inner leg mechanism group 503 is used as the swinging leg mechanism group, so that the robot 501 climbs the first step. Then, the inner leg mechanism group 503 is used as the supporting leg mechanism group and the outer leg mechanism group 502 is used as the swinging leg mechanism group, so that the robot 501 climbs the second step. The outer leg mechanism group 502 and the inner leg mechanism group 503 swing alternately in sequence to complete the stair-climbing task.

[0071] Figure 6 This is a schematic diagram of the planar motion of a robot climbing stairs, provided in an exemplary embodiment of this application. Figure 6 The left and right leg mechanisms of the robot are completely overlapped, therefore, Figure 6 The leg mechanisms connected to the robot's two wheels represent the robot's first and second leg mechanisms. Considering the specific form of climbing stairs, Figure 6 The robot shown does not simultaneously contact the same support surface when its wheels are on the ground, to match the actual movement of a staircase. The robot's movement can also be divided into a single-leg support phase (SSP) and a double-leg support phase (DSP).

[0072] Figure 6The first state shown is the single-leg support stage. The robot's first leg mechanism 601 is in contact with the support surface, while the second leg mechanism 602 is not. The first leg mechanism 601 uses the joints of the wheels and body to maintain the robot's balance. In the first stage, the robot swings the second leg mechanism 602, gradually lifting itself from the initial vertical downward direction of gravity until it rests on the next step, entering the second state, the two-leg support stage. During the first stage, the first leg mechanism 601 acts as the support wheel leg, and the second leg mechanism 602 acts as the swing wheel leg. In the second stage, the robot maintains two contact points between the wheels and the support surface within this plane. During this process, the robot changes the angles of its various joints, and while keeping the two contact points between the wheels and the support surface constant, gradually moves the projection of the robot's center of mass onto the support surface from near the center of the rear wheel on the next step to near the center of the front wheel on the previous step, entering the third state, the two-leg support stage. In the third stage, the robot lifts the first leg mechanism 601 from the lower step and uses the second leg mechanism 602 to maintain the robot's balance until the first leg mechanism 601 and the second leg mechanism 602 coincide in the vertical direction, entering the fourth state, the single-leg support stage. During the third stage, the second leg mechanism 602 acts as a supporting leg, and the first leg mechanism 601 acts as a swinging leg.

[0073] As the robot moves up one more step, the supporting leg mechanism and the swinging leg mechanism interchange. This periodic interchange occurs during the robot's upward movement. Similarly, the stages for the robot's downward movement are the same as those for the upward movement, except that the direction of movement of each leg mechanism is opposite. Therefore, the process for the robot's downward movement will not be elaborated here; please refer to [reference needed]. Figure 7 The diagram shown is a schematic of the robot's sequence of movements going up and down stairs, and will not be described in detail.

[0074] Please refer to Figure 8 It shows a block diagram of a robot control system provided in one embodiment of this application, and the robot body can be referenced. Figure 2 The robot's four wheels are driven by four rotary motors, and the lengths of its four legs are driven by four linear motors. Typically, the linear motors of the two outer legs have the same drive mode, and the linear motors of the two inner legs have the same drive mode.

[0075] Optionally, the two outer moving legs have equal leg lengths, and the two inner moving legs have equal leg lengths. The rates of change of leg length for the two outer moving legs are equal, and the rates of change of leg length for the two inner moving legs are equal.

[0076] The two outer moving legs are driven by the same rotary motor, which adjusts the angle between the outer moving legs and a straight line L1 perpendicular to the contact surface, ensuring that the angles between the two outer moving legs and the straight line L1 are equal. The two inner moving legs are also driven by the same rotary motor, which adjusts the angle between the inner moving legs and L1, ensuring that the angles between the two inner moving legs and the straight line L1 are equal.

[0077] In addition, each joint of the torso mechanism of the wheeled robot is equipped with a rotary motor, and the rotation of each joint can be actively driven by the rotary motor.

[0078] For all the rotary motors mentioned above, the rotary motor can receive rotation angle commands, rotation speed commands, rotation torque commands, etc.; among them, the rotation angle command is used to indicate the rotation angle of the rotary motor, the rotation speed command is used to indicate the rotation speed and rotation direction of the rotary motor, and the rotation torque is used to indicate the torque required for the rotary motor to rotate.

[0079] After the rotary motor receives any one of the above-mentioned command signals, such as rotation angle command, rotation speed command, or rotation torque command, the bottom drive board of the rotary motor will drive the motor to rotate according to the received command signal, thereby driving the movement of each joint.

[0080] For all the linear motors mentioned above, the linear motor can receive linear movement position commands, linear movement speed commands, driving force commands, etc. The underlying drive board of the linear motor will drive the motor to move linearly according to the received command signals, thereby driving the leg mechanism to extend and retract.

[0081] Rotary and linear motors alter the robot's posture and position in three-dimensional space, enabling balance control of wheeled robots. Rapidly changing joint angle commands can also cause highly dynamic changes in the robot's posture, altering its interaction with the environment.

[0082] Since the posture and state (such as motion state) of wheeled-legged robots are different at different times, it is necessary to collect the posture information and motion state information of the wheeled-legged robots through sensors installed on the robots in order to generate command signals and realize the balance control and motion state control of the wheeled-legged robots.

[0083] Optionally, the state of the wheeled robot can be acquired by different sensors mounted on the robot body. The sensor types include at least one of the following:

[0084] An IMU sensor is used to determine the current posture of the wheeled robot, including the robot's position and posture information.

[0085] The motor encoder is used to determine the rotational speed of each joint of the wheeled robot in the current state, the positional information related to movement, and the speed information of rotation and movement.

[0086] Force / torque sensor is used to determine the magnitude of the force, torque, and direction of the torque at the joint where the force / torque sensor is located at the current moment.

[0087] Tactile sensors are used to determine the contact position between the wheels of a wheeled robot and the contact surface, the pressure on body parts such as the body surface, inside the hand, and fingertips, and the characteristics of pressure changes over a period of time.

[0088] Visual sensors (such as cameras and infrared sensors) are used to identify obstacles appearing in the field of vision of wheeled robots, and also to determine the position information of the center of mass of wheeled robots and the corresponding velocity information of the center of mass of wheeled robots.

[0089] Optionally, the control system of the wheeled robot includes the following control modules:

[0090] 1. The State Estimation module fuses various posture and state information acquired by the wheeled robot. For example, fusing the robot's current posture from the IMU sensor, the odometer readings from the rotating wheels, and visual positioning information yields the robot's position in the world coordinate system; fusing information from force / torque sensors and tactile sensors provides the robot's contact with its environment; and fusing the robot's current posture from the IMU sensor with the angle information from each motor joint encoder, combined with the robot's own model parameters, estimates the robot's center of mass. This fused state information serves as feedback for the robot's motion generation, planning, and control.

[0091] 2. The Motion Generation module is used to generate motion strategies corresponding to the working mode of the wheeled robot when performing actions. The robot's working modes include, but are not limited to: four-wheel motion mode, two-wheel motion mode, four-wheel to two-wheel transition mode, stair climbing mode, four-wheel active suspension mode (four-wheel active suspension mode refers to a mode where the robot adjusts its upper body posture using its four wheels to maintain relative stability when the contact surface is uneven during four-wheel movement), and folding mode. Considering the complexity of the upper body, it can be used to complete various tasks, and the robot can include even more motion modes, which will not be listed here.

[0092] The methods for generating actions differ across different modes. These modes invoke some common underlying technologies and algorithm modules, including but not limited to: model-free controllers, model-based controllers, and model-based controllers such as LQR (Linear Quadratic Regulator), MPC (Model Predictive Control), adaptive controllers, and robust controllers. Optionally, the robust controller in this embodiment is implemented based on sliding mode control; a robust controller can also be called a sliding mode controller.

[0093] For example, when a wheeled robot needs to control the balance of its moving wheels in two-wheel mode, a model-free controller with PID (Proportional Integral Derivative) control can be used to generate a reference trajectory for the wheels and the robot's center of mass. Specifically, this can be achieved by calling one or more of the following: a model-free controller, a model-based controller (LQR, MPC, etc.), an adaptive controller, or a robust controller.

[0094] For example, in the dual-wheel control stage of a wheeled robot in the four-wheel to two-wheel mode, model-free controllers, model-based controllers (LQR, MPC, etc.), adaptive controllers, and robust controllers can be used to control the balance of the moving wheels.

[0095] For example, when a wheel-legged robot is in four-wheel mode, if the forward-extending leg mechanism and the rearward-extending leg mechanism are equated, an equivalent leg mechanism is obtained. This equivalent moving leg and torso mechanism are then combined to form a dynamic system described using an nth-order inverted pendulum. The resulting control trajectory can maintain the robot's balance in four-wheel mode. This method enables the wheel-legged robot to maintain relative stability even on uneven surfaces, with potholes or obstacles. For instance, when the wheel-legged robot is in four-wheel active suspension mode, the above module can also be used to control the relative stability of the robot's body.

[0096] The input to the motion generation module is a series of task information for the wheeled-legged robot. The types of task information include, but are not limited to: center of mass tasks, supporting leg tasks, swing leg tasks, waist tasks, etc. Similarly, considering the complexity of the upper body of the wheeled-legged robot, the upper body can independently complete a variety of actions and tasks, and the types of task information can be even more, which will not be listed here.

[0097] For example, these tasks can also serve as inputs to a Whole Body Controller (WBC) (also known as a whole-body dynamics controller). Within the WBC, the wheeled-legged robot undergoes detailed modeling and calibration. The robot's dynamic model and external force conditions are used as optimization constraints. Through an optimization process, target joint angle commands, target joint angular velocity commands, and target joint torque commands for at least one rotating joint of the wheeled-legged robot are calculated. Finally, the WBC sends these determined target joint angle commands, target joint angular velocity commands, and target joint torque commands to the actuators of each joint of the robot, thus completing the control closed loop for the wheeled-legged robot.

[0098] The balance control method for the wheeled-legged robot provided in this application mainly consists of... Figure 8 The robust controller in the control system completes the task. Optionally, the balance control method is completed jointly by the robust controller and the whole-body motion control module. For details of this process, please refer to the following embodiment.

[0099] Before introducing the specific steps of the balance control method provided in this application, the method for establishing the dynamic model corresponding to the balance control method provided in this application will be introduced first. In order to facilitate the understanding of the principle of the nth-order inverted pendulum model involved in this application, the second-order inverted pendulum model, which is easier to understand, will be introduced first.

[0100] Please refer to Figure 9 The diagram shows a schematic of a second-order inverted pendulum provided in one embodiment of this application.

[0101] To implement a balance control method for a wheeled-legged robot, it is necessary to model the robot's dynamics to obtain a dynamic model of the robot during the balance control process. For clarity and simplicity, and to facilitate understanding, the discussion will begin with the case where at least one leg mechanism of the wheeled-legged robot has the same length, at least one leg mechanism has the same angle with the ground, and all moving wheels have the same rotational speed and position. The specific details are as follows:

[0102] Establish a world coordinate system with the forward direction of the wheel-legged robot as the positive x-axis, the rightward direction as the positive y-axis, and the direction perpendicular to the contact surface upwards as the positive z-axis. Observing the wheel-legged robot from the y-axis direction, it will be observed that at least one leg mechanism of the wheel-legged robot coincides. This at least one leg mechanism can be the outer leg mechanism of the wheel-legged robot, or it can be all the leg mechanisms of the wheel-legged robot.

[0103] For example, the length changes of the four leg mechanisms of the wheel-legged robot are synchronized, and the angles between the four leg mechanisms and the robot's base are also synchronized. Viewed from the direction corresponding to the y-axis, the four leg mechanisms coincide, and the four wheels also coincide. In this case, on the yoz plane of the world coordinate system, the abstract two-dimensional model of the wheel-legged robot is as follows: Figure 9 As shown. This two-dimensional model belongs to the second-order inverted pendulum model, which includes: a moving wheel, a link B (corresponding to at least one moving leg of the wheel-legged robot), and a link P (corresponding to the torso mechanism of the wheel-legged robot).

[0104] like Figure 9 As shown, in the two-dimensional model, the direction in which the moving wheel rotates to the left of the contact surface is taken as the positive direction, the travel distance is x, and the angle through which the moving wheel rotates relative to the world coordinate system is defined as... The counterclockwise direction is taken as the positive direction of the rotation angle of the moving wheel. The angle through which the connecting rod 910 (connecting rod B) corresponding to the four leg mechanisms rotates relative to the world coordinate system is defined as α, with the counterclockwise direction being positive. The angle through which the connecting rod 930 (connecting rod P) corresponding to the torso mechanism rotates relative to the world coordinate system is defined as β, with the counterclockwise direction being taken as the positive direction of the rotation angle of the moving wheel.

[0105] definition They are respectively The time derivatives of α and β; where, This indicates the rotational speed of the moving wheel (also known as the angle of the moving wheel). This indicates the angular velocity of the leg mechanism. Represents the angular velocity of the torso mechanism; definition They are respectively The second derivatives of α and β with respect to time; where, This represents the angular acceleration of the moving wheel. This indicates the angular acceleration of the leg mechanism. This indicates the angular acceleration of the torso mechanism.

[0106] The angle through which the moving wheel rotates relative to the world coordinate system is determined by the first rotation joint (equivalent to...). Figure 9 The joint motor drives the first rotating joint (920), and the rotational torque of the first rotating joint is represented by τ1, with the clockwise direction being taken as the positive direction of the rotational torque of the first rotating joint.

[0107] The torso mechanism rotates relative to the world coordinate system by the angle determined by the second rotation joint (equivalent to...). Figure 9 The joint motor of the 940) is driven, and the rotational torque of the second rotating joint is represented by τ2, with the counterclockwise direction as the positive direction of the rotational torque of the second rotating joint.

[0108] The masses of the moving wheel, link B (equivalent to at least one overlapping leg mechanism), and link P (equivalent to the torso mechanism) are respectively expressed as m. W m B m P The moments of inertia of the moving wheel, link B, and link P are respectively expressed as: J W J B J P The radius of the moving wheel is denoted by r, and the length of link B is denoted by L. B The length of the line connecting the intersection of link B and the moving wheel to the geometric center of link B is represented by l. B The length of the line connecting the intersection of link B and the wheel to the geometric center of link P is represented by l. P express.

[0109] (1) Based on the definition of the above physical quantities, derive the total kinetic energy T of the moving wheel respectively. W The total kinetic energy T of link B B The total kinetic energy T of the connecting rod P P The expression for the total kinetic energy T of the wheel-legged robot is obtained; where the total kinetic energy T of the aforementioned moving wheels is... W The total kinetic energy T of link B B The total kinetic energy T of the connecting rod P P The kinetic energy calculations all include translational kinetic energy and rotational kinetic energy.

[0110]

[0111]

[0112]

[0113]

[0114] T = T W +T B +T P

[0115] (2) Calculate the total kinetic energy T of the system for each degree of freedom in the generalized coordinate system. The partial derivatives of the derivatives for each degree of freedom are calculated separately, and the results of the partial derivatives of the derivatives for each degree of freedom are then differentiated with respect to time. The specific process is as follows:

[0116] The total kinetic energy T of the system is related to the angular velocity of the moving wheel. partial derivatives Represented as:

[0117]

[0118] partial derivatives derivative with respect to time Represented as:

[0119]

[0120] The total kinetic energy T of the system with respect to the angular velocity of the leg mechanism partial derivatives Represented as:

[0121]

[0122] partial derivatives derivative with respect to time Represented as:

[0123]

[0124] The total kinetic energy T of the system relative to the angular velocity of the torso mechanism partial derivatives Represented as:

[0125]

[0126] partial derivatives The derivative with respect to time is expressed as:

[0127]

[0128]

[0129] The deflection angle of the moving wheel relative to the total kinetic energy T of the system partial derivatives Represented as:

[0130]

[0131] The partial derivative of the total kinetic energy T of the system with respect to the deflection angle α of the leg mechanism Represented as:

[0132]

[0133] The partial derivative of the total kinetic energy T of the system with respect to the deflection angle β of the torso mechanism Represented as:

[0134]

[0135] (3) Calculate the total potential energy U of the system:

[0136] U = m B gl B cosα+m P g(L B cosα+l P cosβ)

[0137] It should be noted that the plane where the moving wheel is located is taken as the zero potential energy surface in the above formula, that is, the potential energy of the moving wheel is considered to be 0. Therefore, the potential energy of the moving wheel does not appear in the formula for the total potential energy U of the system. Of course, other potential energy surfaces can be chosen, which will not affect the implementation of the balance control method. This application does not limit the potential energy surface used in the process of calculating the total potential energy of the system.

[0138] (4) Calculate the partial derivative of the total potential energy U of the system with respect to each degree of freedom in the generalized coordinates:

[0139] The deflection angle of the moving wheel due to the total potential energy U of the system partial derivatives Represented as:

[0140]

[0141] The partial derivative of the total potential energy U of the system with respect to the deflection angle α of the leg mechanism Represented as:

[0142]

[0143] The partial derivative of the total potential energy U of the system with respect to the deflection angle β of the torso mechanism Represented as:

[0144]

[0145] (5) Using the above formulas, the following dynamic equations for the wheel-legged robot under the second-order inverted pendulum model can be derived based on the Euler-Lagrange equations:

[0146]

[0147]

[0148]

[0149] m αα =(m B l B 2 +J B +m P L B 2 +J P )

[0150] m αβ =m P L B l P cosαcosβ+m P L B l Psinαsinβ+J P

[0151] m ββ =(m P l P 2 +J P )

[0152]

[0153]

[0154]

[0155]

[0156] g α =-(m B gl B +m P gL B sinα

[0157] g β =-m P gl P sinβ

[0158] The dynamic equations can be written in polynomial form:

[0159]

[0160]

[0161]

[0162] Rewrite the dynamic equations in polynomial form as dynamic equations in matrix form:

[0163]

[0164] Where M(α,β) is a 3*3 inertia matrix; specifically including

[0165] Elements in M(α,β) These are the elements in the inertia matrix, used to characterize the mass and moment of inertia of each joint rigid body that makes up the wheel-legged robot when the leg mechanism deflection angle is α and the torso mechanism deflection angle is β, as well as the equivalent inertial physical quantities of each mechanism under mutual influence. The deflection force matrix is ​​a 3*1 matrix, also known as the deflection force vector, and specifically includes: Used to characterize the Coriolis force and centripetal force acting on various mechanisms of a wheel-legged robot; G(α,β) represents a 3*1 gravity matrix, also known as a gravity vector, specifically including [0 g α g β ] T τ1 represents the rotational torque of the first rotary joint, and τ2 represents the rotational torque of the second rotary joint.

[0166] After deriving the matrix form of the dynamic equation through the above process, the subsequent balance control process can directly use the matrix form of the dynamic equation without repeating the above derivation process during the balance control process.

[0167] The following section introduces and explains the nth-order inverted pendulum model.

[0168] For ease of understanding, let's first consider the case where at least one leg mechanism of the wheel-legged robot coincides in the y-axis direction, the leg length of at least one leg mechanism changes synchronously, the included angle between at least one leg mechanism and the torso mechanism is the same, and the moving speed of at least one moving wheel is equal.

[0169] The world coordinate system used in this example is... Figure 9 While the definition of a world coordinate system is the same, other methods can also be used to define a world coordinate system, and this application does not impose any restrictions on this.

[0170] like Figure 10 The n-order inverted pendulum shown consists of n links (corresponding to the n links in the claims) and a moving wheel; wherein the moving wheel is connected to the first link among the n links, the n links are connected in series, and n is a positive integer greater than or equal to 2. For example, the moving wheel in the n-order inverted pendulum corresponds to the moving wheel of a wheeled robot. In the case of a wheeled robot with multiple moving wheels, the above assumption sets the motion states of the multiple moving wheels to be synchronized; therefore, multiple moving wheels can be equivalent to the same moving wheel in the n-order inverted pendulum model, and the rotation angle of the moving wheel is expressed as...

[0171] The n links of the n-order inverted pendulum model correspond to the n links of the wheel-legged robot. Two adjacent links in the n mechanisms are connected by a rotary joint. The deflection angles of the n links are q1, q2, ..., q... n Where q1 is the deflection angle of the first link, q2 is the deflection angle of the second link, and so on, q n Let be the deflection angle of the nth link. The deflection angle of the link is explained using the first link as an example: [Example...] Figure 10As shown, in this derivation process, the deflection angle of the first link refers to the deflection angle of the first link relative to the z-axis in the world coordinate system. Of course, other methods can also be used to define the deflection angle of the first link. In this case, the derivation process of the nth-order inverted pendulum model is similar to that of this embodiment.

[0172] For example, when the wheeled robot is in a standing posture, the distance l1 between the first link and the contact surface is less than the distance l2 between the second link and the contact surface, and so on, less than the distance ln between the nth link and the contact surface. n The first link (which can be a leg mechanism of a wheel-legged robot) among n links is connected to the movable wheel via a first revolute joint. Optionally, a link refers to a simplified model of the various mechanisms included in a wheel-legged robot. A wheel-legged robot can have multiple components, each of which can be called a "mechanism," and each mechanism can be simplified as a link. In some cases, the wheel-legged robot is in a posture where multiple movable wheels are in contact with the contact surface, and each of these movable wheels corresponds to its own leg mechanism; for any one of the multiple movable wheels, the revolute joint connecting that movable wheel to its corresponding leg mechanism is called the first revolute joint.

[0173] After understanding the nth-order inverted pendulum model, the dynamic equations of the nth-order inverted pendulum model are constructed in a manner similar to that used in constructing the dynamic equations of the second-order inverted pendulum model.

[0174] Specifically, determine the total kinetic energy (including translational kinetic energy and rotational kinetic energy) and potential energy corresponding to each of the n links and the moving wheels. Add the kinetic energies of the n links and the moving wheels to obtain the total kinetic energy of the wheel-legged robot system. Add the potential energies of the n links and the moving wheels to obtain the total potential energy of the wheel-legged robot system.

[0175] For any one of the multiple degrees of freedom in the generalized coordinate system (including the deflection angle of each link, the angular velocity of each link, the rotation angle of the transfer wheel, and the angular velocity of the transfer wheel), determine the partial derivative of the total kinetic energy of the system with respect to that degree of freedom, and determine the derivative of the partial derivative of that degree of freedom with respect to time.

[0176] For any one of the multiple degrees of freedom in the generalized coordinate system, determine the partial derivative of the total kinetic energy of the system with respect to that degree of freedom. Based on the partial derivatives of the system's kinetic energy with respect to each degree of freedom, the derivatives of the partial derivatives of each degree of freedom with respect to time, and the derivatives of the partial derivatives of each degree of freedom with respect to time, the system dynamic equations of the nth-order inverted pendulum are determined by the Euler-Lagrange equations:

[0177]

[0178] in, This refers to the inertia matrix, which includes the mass and moment of inertia of the n rotating joints (considering the rotating joints as rigid bodies) of the wheel-legged robot involved in the n-order inverted pendulum model. q is a vector, and the transpose of q is qinertia. T =[q1 q2…q n ], This refers to the angular velocity of the moving wheel. This refers to the angular acceleration of the moving wheel. This represents the angular velocity of n links. For vectors, transpose vector This represents the angular acceleration of n links. For vectors, The transpose matrix is This refers to the bias force matrix, used to characterize the angle of a wheeled robot at its moving wheel. The deflection angles of the n links are q1, q2, ..., q n When the device is in a certain orientation, each link experiences a biasing force, which includes centripetal force and Coriolis force. Here is the gravity matrix, used to characterize the angle of the wheeled robot at the moving wheel. The deflection angles of the n links are q1, q2, ..., q n When the n links are in the desired orientation, the gravitational forces acting on them are respectively.

[0179] Since the plane where the moving wheel is located is defined as a zero potential energy surface, the first term in the G matrix is ​​0, indicating that the gravity corresponding to the moving wheel on this zero potential energy surface is zero. Of course, other zero potential energy surfaces can also be chosen, and this application does not limit them here. n Let τ represent an n×n identity matrix. T =[τ1τ2…τ n ] represents the rotational torque corresponding to n rotating joints.

[0180] The following describes the balance control method for abstracting a wheeled-legged robot into an nth-order inverted pendulum. This method can be implemented by a robust controller in the wheeled-legged robot control system based on sensor observations, or it can be implemented collaboratively by the robust controller and other control modules such as the whole-body dynamics controller in the control system. Please refer to the embodiments below for details.

[0181] Please refer to Figure 11This document illustrates a flowchart of a balance control method for a wheeled-legged robot according to an embodiment of this application. The execution entity for each step of the method can be a computer device. The wheeled-legged robot includes: a movable wheel, n links, and n rotary joints. The movable wheel and a first link among the n links are connected via a first rotary joint among the n rotary joints. The n links are connected in series via n-1 rotary joints excluding the first rotary joint, where n is a positive integer greater than or equal to 2. The method may include at least one of the following steps (1110-1150):

[0182] Step 1110: Obtain the state variables of the wheeled-legged robot at the first moment. The state variables at the first moment are used to characterize the motion state of the wheeled-legged robot at the first moment.

[0183] In some embodiments, the movable wheel is used to contact the contact surface, the rotary joint refers to a joint driven by a joint motor, and the n links refer to links connected to at least one of the n rotary joints. Optionally, the n rotary joints connect the n links in series into a linear configuration, and the first link of the n links is connected to the movable wheel through the first rotary joint. That is, the movable wheel, n links, and n rotary joints of the wheel-legged robot can be abstracted into the nth-order inverted pendulum described above. The balance control method provided in this application embodiment is based on the nth-order inverted pendulum design.

[0184] Alternatively, the wheeled robot comprises multiple components, each of which (excluding the active joints) can be referred to as a mechanism, which can be simplified as a link.

[0185] Optionally, any link among the n mechanisms is equivalent to a link in an n-order inverted pendulum, meaning that ideally, this link will not bend or deform. For example, each of the n links can be an independent link (e.g., in a wheel-legged robot, the leg mechanism is one link and the torso mechanism is another link).

[0186] For example, the n links include at least one composite link consisting of m independent links connected by joints, where m is a positive integer greater than 1. In this case, the joints connecting the m independent links do not rotate or move linearly during balance control; for example, these joints are locked during balance control, so that the composite link consisting of the m independent links can be regarded as one of the n links.

[0187] In some embodiments, the first link refers to the link connected to the moving wheel of the wheeled robot via a first rotary joint. Optionally, among the n links, the first link has the smallest distance to the moving wheel in the direction perpendicular to the contact surface (i.e., the z-axis direction in the world coordinate system).

[0188] Optionally, if the leg mechanism of the wheel-legged robot is a straight leg, then the first link refers to the leg mechanism of the wheel-legged robot. If the leg mechanism of the wheel-legged robot is an articulated leg, that is, the leg mechanism includes two sub-mechanisms connected by a knee joint, and the knee joint is used to control the joint angle between the two sub-structures, then the first link refers to the sub-mechanism in the leg mechanism that is directly connected to the first rotary joint.

[0189] In this application, n is a positive integer greater than 1, for example, n equals 2, 3, 4, 5... The maximum value of n is related to the construction of the wheel-legged robot, and this application does not limit it here.

[0190] For example, when n equals 2, the n links include: the leg mechanism and torso mechanism of the wheel-legged robot. For example, when n equals 3, the n links include: the leg mechanism (legs are straight legs), torso mechanism, and head mechanism of the wheel-legged robot. For example, when n equals 3, the n links include: the leg mechanism (legs are straight legs), torso mechanism, and robotic arm mechanism of the wheel-legged robot. For example, when n equals 4, the n links include: the leg mechanism (legs are straight legs), torso mechanism, a first robotic arm link connected to the torso mechanism, and a second robotic arm link connected to the first robotic arm, or a robotic hand. For example, when n equals 4, the n links include: a first sub-mechanism of the leg mechanism (legs are straight legs), a second sub-mechanism connected to the first sub-mechanism via a knee joint, the torso mechanism, and a first robotic arm link connected to the torso mechanism.

[0191] Optionally, n revolute joints are used to connect n links in series. For example, any one of the n revolute joints refers to a type of revolute joint that performs the same function. For instance, in a wheeled robot comprising multiple wheels, there are multiple revolute joints used to connect each wheel to its corresponding first link; these multiple revolute joints all belong to the first revolute joint category.

[0192] It should be noted that the value of n is related to the configuration of the wheeled robot, the computing power of the computer, power consumption, and other practical requirements, and this application does not impose any restrictions on it. The larger the value of n, the more links can be adjusted in posture during the balance control process, resulting in a richer posture of the wheeled robot during balance adjustment, which helps to improve the balance ability of the wheeled robot to cope with external disturbances.

[0193] In some embodiments, the first moment is any moment in the motion process of the wheel-legged robot.

[0194] Optionally, at the first moment, the wheel-legged robot's posture is as follows: all first links have equal angles with the direction perpendicular to the contact surface, the angular velocities of the corresponding moving wheels of each first link are equal, and at least one first rotary joint connecting the first links and the moving wheels rotates synchronously. That is, viewed from the side of the wheel-legged robot, at least one first link overlaps, and at least one moving wheel overlaps. For example, the wheel-legged robot has four moving wheels, and at the first moment, the wheel-legged robot is in a dual-wheel balance mode (moving by rolling on the contact surface via two moving wheels, while the other two moving wheels are retracted).

[0195] In some embodiments, the state variables are used to describe the motion state of the wheeled-legged robot at a first moment. The attitude and motion state of the wheeled-legged robot at the first moment can be determined based on the state variables at that moment. Optionally, the state variables at the first moment are obtained from the sensors of the wheeled-legged robot.

[0196] Optionally, the state variables at the first moment include at least one of the following: the deflection angles q1, q2, ..., q of the n links. n angular velocity of the moving wheel angular velocities of n links

[0197] For example, the state variables include various physical quantities that are observed in the world coordinate system; among them, the deflection angle q of the i-th link out of n links. i (i∈[1,n]) refers to: the deflection angle of the i-th link relative to the z-axis in the world coordinate system; the angular velocity of the moving wheel. This refers to: the clockwise rotational speed of the moving wheel along the x-axis in the world coordinate system; and the angular velocity of the i-th link. Used to characterize the angular velocity of the i-th link, that is, the rate of change of the deflection angle of the i-th link.

[0198] It should be noted that the physical quantities in the state variables can also be represented by the relative positions between the links of the wheeled robot. For example, the deflection angle q of the j-th link. j ′(i∈[2,n]) refers to the deflection angle of the j-th link relative to the (j-1)-th link; wherein the (j-1)-th link and the j-th link are connected by a revolute joint. Optionally, the distance between the j-th link and the contact surface is greater than the distance between the (j-1)-th link and the contact surface. It should be noted that the observation coordinate system corresponding to each physical quantity in the state variables can be determined according to actual needs, and this application does not impose any restrictions here.

[0199] The physical quantities in the formulas in this application are obtained based on observations in the world coordinate system. Of course, the physical quantities determined by other observation coordinate systems can also achieve this balance control method. The relevant formulas can be obtained by equivalent substitution based on the formulas provided in this application, which will not be mentioned one by one in the text.

[0200] When the y-axis of the wheel-legged robot in contact with the contact surface coincides with that of the moving wheels in the world coordinate system, the angular velocities of the n moving wheels are the same, and the deflection angle of the first link connected to at least one moving wheel is the same. Please refer to the following embodiment for the method of obtaining the state variables at the first moment.

[0201] For example, the state variables are: the deflection angles q1, q2, ..., q of n links. n angular velocity of the moving wheel angular velocities of n links Composition. State variables can be represented by the symbol ξ, and the state variable ξ can be expressed as:

[0202]

[0203] Where, q T Let q be the transpose of vector q. T =[q1 q2 … q n ], Representing vectors The transpose of ,

[0204] Optionally, the state variables of the wheeled robot at the first moment are obtained, including: determining the deflection angles q1, q2, ..., q of the n links using IMU sensors and motor encoders. n The angular velocity of the moving wheel is determined by the motor encoder. angular velocities of n links For example, in a wheeled robot, the clock cycles of each IMU sensor and motor encoder are the same or multiples of each other, ensuring that all physical quantities included in the state variables are physical quantities at the first moment. For details regarding the IMU encoder and motor encoder, please refer to the above description; they will not be repeated here.

[0205] In some embodiments, the robust controller takes the state quantity at the first moment as input information and performs a balance control process based on the state quantity at the first moment.

[0206] Step 1120: Determine the dynamic model parameters based on the dynamic equations of the wheel-legged robot and the state variables at the first moment. The dynamic model parameters are used to define the mapping relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The angular acceleration at the first moment includes the angular acceleration of the n links and the angular acceleration of the moving wheels. The rotational torque at the second moment includes the rotational torque of the n rotating joints.

[0207] In some embodiments, the angular acceleration at the first moment refers to the rotational acceleration of the n links after the wheel-legged robot is abstracted into an nth-order inverted pendulum model (which can also be understood as the angular acceleration of the joint motors corresponding to the n rotating joints). The angular acceleration at the first moment can be determined by... express.

[0208] As can be seen from the above, after abstracting the wheel-legged robot into an nth-order inverted pendulum model, we can obtain the matrix-form dynamic equations:

[0209]

[0210] In the embodiments of this method, a robust controller is designed in the control system of the wheeled robot based on the equation, so that the robust controller can perform the balance control method provided in the embodiments of this application.

[0211] Based on the physical relationship between acceleration, velocity, and position, the acceleration at the first moment can predict the angular velocities of the n links of the wheel-legged robot in the next moment, as well as the robot's attitude in subsequent moments. The rotational torque of the joints can change the robot's velocity and attitude. In the design of the robust controller in this scheme, the focus is on establishing a mapping relationship between the robot's acceleration and the joint rotational torque based on matrix-form dynamic equations.

[0212] For example, the matrix-form dynamic equations are partially linearized with feedback, placing the matrix related to the acceleration of the wheel-legged robot on one side of the dynamic equations and the matrix related to the rotational torque of the rotating joints on the other side. This determines the mapping relationship between the acceleration and rotational torque at the first moment, which is represented by the dynamic parameters. This process is completed during the design of the robust controller, and the relevant formulas do not need to be derived again during actual balance control.

[0213] In some embodiments, the dynamic model parameters are used to characterize the linear mapping relationship between acceleration and torque at the first moment. That is, the dynamic model parameters ensure that the acceleration and torque at the first moment satisfy an equal relationship in the formula derived from the dynamic formula. Since there is more than one acceleration and torque at the first moment, the dynamic model parameters can be expressed in matrix form.

[0214] For example, the dynamic model parameters include: a scaling parameter matrix g[] and an offset parameter matrix f[]; wherein, the scaling parameter matrix g[] is used to characterize the proportional relationship between the acceleration and the rotational torque at the first moment, and the offset parameter matrix f[] is used to characterize the offset between the acceleration and the rotational torque at the first moment. For the calculation methods of the above two dynamic model parameter matrices, please refer to the following embodiment.

[0215] Optionally, in the design process of the robust controller, the representation of the dynamic model parameters is predetermined. After obtaining the state variables at the first moment, the robust controller substitutes the state variables at the first moment into the representation of the dynamic model parameters, thereby determining the values ​​of each element included in the dynamic model parameters.

[0216] Step 1130: Based on the state variables at the first moment, establish a sliding surface. The state variables of the wheeled robot gradually approach 0 on the sliding surface.

[0217] In some embodiments, the sliding surface can be understood as an imaginary vector plane in sliding mode control. The state variables of the wheeled robot gradually approach 0 on the sliding surface, meaning that the state of the wheeled robot can gradually approach equilibrium on the sliding surface. The sliding surface can be represented by the symbol s. Optionally, the sliding surface is a rational number, and its value can be greater than 0, equal to 0, or less than 0.

[0218] Optionally, the sliding surface is a linear sliding surface that has a linear relationship with the state variables, or the sliding surface belongs to a sliding surface where there is a higher-order mapping relationship between the state variables. Considering the purpose of reducing the computational burden on robust controllers and reducing the energy consumption of wheeled robots in practical applications, this embodiment mainly uses the creation of a linear sliding surface as an example to introduce and explain the method of determining the sliding surface based on the state variables.

[0219] In some embodiments, the number of sliding surfaces that the robust controller needs to establish is related to the number of torques to be determined. Optionally, the number of sliding surfaces that the robust controller needs to establish is equal to the number of torques to be determined. In this application, the wheel-legged robot is abstracted as an n-order inverted pendulum, involving n rotational joints, each with an independent torque. Therefore, the torque at the second moment includes the torques of the n rotational joints, requiring the establishment of n sliding surfaces; wherein any two sliding surfaces among the n sliding surfaces are different sliding surfaces. For specific steps on establishing sliding surfaces, please refer to the embodiments below.

[0220] Step 1140: Calculate the rotational torque of n rotating joints based on the sliding surface and dynamic model parameters.

[0221] In some embodiments, for any of the n rotary joints, the rotational torque of the rotary joint is used to control the rotation of the joint motor of that rotary joint. Optionally, the joint motor of the rotary joint can rotate at a certain angular velocity, speed, or torque, so that the rotary joint reaches the corresponding rotational torque.

[0222] In some embodiments, the robust controller is designed with an equation for calculating the rotational torque of n rotary joints using sliding surface and dynamic model parameters. By substituting the sliding surface and dynamic model parameters into this equation, the rotational torque of the n rotary joints can be calculated. The rotational torque of the n rotary joints controls the rotary motors corresponding to each rotary joint, adjusting the posture of the wheeled robot and helping it achieve or maintain a balanced state.

[0223] Optionally, in the process of calculating the rotational torque of the n rotary joints based on the sliding surface and dynamic model parameters, it is also necessary to consider the system stability criterion. The system stability criterion is used to constrain the robot's balance state. For details of this part, please refer to the following embodiment. For example, steps 1110-1140 are completed by a robust controller.

[0224] Step 1150: Based on the rotational torque of the n rotary joints, control the n rotary joints at the second moment.

[0225] In some embodiments, the second moment refers to any moment after the first moment. Optionally, the second moment refers to the moment after the rotational torques of the n rotary joints have been calculated.

[0226] In some embodiments, controlling n rotary joints is used to change the posture of the wheeled robot. Rotation of the rotary joints causes a change in the deflection angle of the links connected to the rotary joints, resulting in a change in the joint angle between the two links connected to the rotary joints. This changes the posture of the wheeled robot and the position of its center of mass, thereby helping the wheeled robot adjust its balance.

[0227] Optionally, after determining the rotational torque of n rotary joints, s rotary joints can be controlled, where s is a positive integer less than or equal to n.

[0228] Alternatively, the methods for controlling the n rotary joints based on their rotational torques include the following:

[0229] 1. The torque of each of the n rotary joints is sent to the joint motor of each of the n rotary joints through a robust controller, so that the joint motors of the n rotary joints rotate to achieve their respective corresponding torque.

[0230] 2. Calculate the task acceleration of the wheel-legged robot using the rotational torques of the n rotating joints; the whole-body dynamics controller in the control system of the wheel-legged robot calculates the force and torque commands corresponding to each joint of the whole body based on the task acceleration, thereby determining the force and torque commands corresponding to the n rotating joints in the n-order inverted pendulum model; the whole-body dynamics controller sends the force and torque commands corresponding to the n rotating joints to the n rotating joints, causing the n rotating joints to rotate according to their respective force and torque commands.

[0231] Full-body joints refer to the active joints in a wheeled-legged robot. After determining the force and torque commands for all the joints, it can not only control the rotation motors of n rotational joints, but also control the joint motors of other joints besides the n rotational joints according to the force and torque commands for all the joints. This makes the posture changes of the wheeled-legged robot more varied. Since the full-body dynamics controller is used to control the robot to perform motion tasks, this method can integrate the balance control method with the motion process of the wheeled-legged robot, which helps to improve the adaptability of the balance control method to different scenarios.

[0232] Optionally, the step of calculating the task acceleration of the wheel-legged robot using the rotational torques of the n rotary joints is performed by the observation module in the control system of the wheel-legged robot. The observation module is connected to the robust controller and the whole-body dynamics controller. The observation module obtains the rotational torques of the n rotary joints from the robust controller, calculates the task acceleration based on the rotational torques of the rotary joints, and transmits the task acceleration to the whole-body dynamics controller. The whole-body dynamics controller determines the force and torque corresponding to each motor of the whole-body joints to be executed based on the task acceleration. For the specific process of this step, please refer to the following embodiment.

[0233] By associating the control results (the two torques mentioned above) of the robust controller with the whole-body dynamics controller, the whole-body dynamics controller can receive information related to balance control. This helps the whole-body dynamics controller to perform reasonable optimization based on the various input information it receives, generate optimized joint commands (i.e., force and torque commands), and send these joint commands to the corresponding joints. This allows more links in the wheeled-legged robot to rotate during balance control, which helps improve the balance control capability of the wheeled-legged robot during movement.

[0234] In summary, by abstracting the wheel-legged robot into a multi-stage inverted pendulum model, the deflection angles of multiple links in the wheel-legged robot can be adjusted during balance control, thereby changing the robot's posture. Compared to related technologies that can only adjust the angle between the moving wheels and the leg mechanism during balance control, this solution allows the posture of multiple links in the wheel-legged robot to be variable during balance adjustment, greatly enriching the robot's posture, facilitating rapid adjustment to a balanced state, and improving the robot's ability to recover to a balanced state under different disturbances, thus enhancing the robustness of the robot's balance control process.

[0235] The following examples illustrate the method for obtaining the state variables at the first moment.

[0236] In some embodiments, the state variables of the wheel-legged robot at a first moment are determined by sensors included in the wheel-legged robot. The state variables at the first moment include: the deflection angles of the n links, the angular velocities of the n links, and the angular velocities of the moving wheels.

[0237] Optionally, the method for obtaining the state variables at the first moment is related to the posture of the wheeled robot at the first moment. In one possible implementation, the state variables at the first moment are obtained from the observations of the wheeled robot's sensors; in another possible implementation, the observation results obtained from the sensor observations of the wheeled robot need to be processed to obtain the state variables at the first moment.

[0238] The wheel-legged robot's moving wheel, which is in contact with the contact surface, coincides with the y-axis in the world coordinate system, and the first link also coincides with the y-axis in the world coordinate system. Optionally, the wheel-legged robot is in a balanced state with two wheels, and step 1110 further includes the following sub-steps:

[0239] Sub-step 1110-a: Determine the deflection angles of n links using the inertial measurement unit and motor encoder of the wheel-legged robot.

[0240] Sub-step 1110-b: Determine the angular velocities of the n links and the angular velocities of the moving wheels using the motor encoder.

[0241] For information on inertial measurement units and motor encoders, please refer to the above description. The robust controller obtains the state quantities at the first moment from the IMUs and motor encoders corresponding to the n rotating joints respectively.

[0242] The following examples illustrate the process of determining the parameters of the dynamic model. Each step of this process can be performed by a computer device.

[0243] Step 1120 in the above embodiment: Determining the dynamic model parameters based on the dynamic equations of the wheeled robot and the state variables at the first moment may include the following steps:

[0244] Step 1123 (not shown in the figure): Substitute the state variables at the first moment into the dynamic equation to determine the inertia matrix, the bias force matrix, and the gravity matrix at the first moment. The inertia matrix is ​​used to characterize the mass and moment of inertia of the n rotating joints at the first moment. The bias force matrix is ​​used to characterize the bias force of the wheel-legged robot at the first moment. The gravity matrix is ​​used to characterize the gravity of the wheel-legged robot at the first moment.

[0245] In some embodiments, the inertia matrix is ​​used to characterize the inertia of the rigid bodies corresponding to the n rotating joints of the wheel-legged robot at the first moment of its posture. The inertia includes at least one of the following: mass and moment of inertia. Optionally, the inertia matrix can be calculated based on the dynamic equations. When the wheel-legged robot is abstracted into an nth-order inverted pendulum model, the inertia matrix is ​​a (n+1)*(n+1)-size matrix.

[0246] In some embodiments, the deflection force matrix is ​​used to characterize the Coriolis force and centripetal force acting on each link. Optionally, the deflection force matrix includes the deflection force caused by the deflection angle of the moving wheel and the deflection force caused by the deflection angle of the n-link.

[0247] In some embodiments, the gravity matrix is ​​used to characterize the gravity acting on each link. Optionally, the gravity matrix includes the gravity acting on the moving wheel and the gravity acting on each of the n links. Optionally, the moving wheel (such as the moving wheel of the outer leg mechanism) is always in contact with the contact surface, the height of the moving wheel remains unchanged during the balance control process, the gravity acting on the moving wheel remains unchanged, and the plane containing the center of mass of the moving wheel is taken as the zero potential energy surface, so the gravity acting on the moving wheel is 0, thereby reducing the computational overhead of the balance control process.

[0248] For details regarding the inertia matrix, corona force matrix, and gravity matrix, please refer to [link / reference needed]. Figure 10 The content corresponding to the instruction manual will not be repeated here. The expressions for each element in the aforementioned inertia matrix, eccentric force matrix, and gravity matrix are also derived in advance from the dynamic formulas. After determining the state variables at the first moment, the robust controller calculates the specific values ​​of each element in the inertia matrix, eccentric force matrix, and gravity matrix at the first moment based on the state variables at the first moment and the representation of each element in the inertia matrix, eccentric force matrix, and gravity matrix; thus obtaining the inertia matrix, eccentric force matrix, and gravity matrix at the first moment.

[0249] Step 1126 (not shown in the figure): Determine the dynamic model parameters based on the inertia matrix, the deflection force matrix, and the gravity matrix.

[0250] In some embodiments, the dynamic model parameters include: a proportional parameter matrix and an offset parameter matrix, wherein the proportional parameter matrix is ​​used to characterize the proportional relationship between the angular acceleration and the rotational torque at the first moment, and the offset parameter matrix is ​​used to characterize the offset relationship between the angular acceleration and the rotational torque at the first moment; optionally, the robust controller calculates the offset parameter matrix based on the inertia matrix, the bias force matrix, and the gravity matrix; and calculates the proportional parameter matrix based on the inertia matrix.

[0251] The rotational relationship between the acceleration of the wheeled robot and the rotational torque of n rotating joints can be determined by the dynamic model parameters. Subsequently, the rotational torque of n rotating joints can be determined relatively quickly based on the dynamic model parameters and the sliding surface. This helps to reduce unnecessary conversions in the process of determining the rotational torque of the rotating joints and reduce the computational burden on the computer equipment.

[0252] In some embodiments, step 1126 further includes the following sub-steps:

[0253] Sub-step 1126-a (not shown in the figure) uses the selection matrix to process the product between the inverse matrix of the inertia matrix and the bias force matrix, and the product between the inverse matrix of the inertia matrix and the gravity matrix, respectively, to obtain the offset parameter matrix. The selection matrix is ​​used to extract the rotational torque of n rotating joints from the dynamic equation.

[0254] Optionally, the elements in the selection matrix include three values: (0, 1, -1). The selection matrix is ​​used to ensure that each rotational torque in τ exists independently in the dynamic equation. That is, the selection matrix is ​​used to process the dynamic equation so that the coefficients of the rotational torques of each rotating joint in the dynamic equation have the same sign, and the dynamic equation does not include basic operations between the rotation matrices of any two rotating joints (such as τ1-τ2).

[0255] Sub-step 1126-b (not shown in the figure) uses the selection matrix to process the inverse matrix of the inertia matrix to obtain the scaling parameter matrix.

[0256] Following on the content regarding step 1120 in the previous embodiment, the two sub-steps included in step 1126 will be described below:

[0257] After performing partial feedback linearization on the matrix-form dynamic equations, we obtain Formula 1:

[0258]

[0259] Among them, M -1 (q) is the inverse of the inertia matrix, M -1 (q)*M(q)=E, where E is the identity matrix. For the physical meaning of the other parameters in Formula 1, please refer to the above embodiment, which will not be repeated here.

[0260] Furthermore, to simplify the execution logic in the balance control process and reduce the computational load of the robust controller, Equation 1 needs to be adjusted during the robust controller design process. Equation 1 is processed using a selection matrix to obtain Equation 2:

[0261]

[0262] in, Represents the offset parameter matrix, Represents the scaling parameter matrix. S T To select the matrix.

[0263] Optionally, Formula 2 is derived after the wheel-legged robot is abstracted into an n-order inverted pendulum model. During the balance control process, after obtaining the state variables at the first moment, the robust controller can calculate the proportional parameter matrix and the dynamic model parameter matrix based on Formula 2 and the expressions of each element in the inertia matrix, bias force matrix and gravity matrix obtained during the derivation of the dynamic equation of the n-order inverted pendulum model.

[0264] In some embodiments, the dynamic equations of the robot described above are derived from the Euler-Lagrange equations, which abstract the wheel-legged robot into an nth-order inverted pendulum model. The moving wheels of the wheel-legged robot correspond to the wheels in the nth-order inverted pendulum model, and at least one link of the wheel-legged robot corresponds to the n links in the nth-order inverted pendulum model. The derivation process of the dynamic equations is detailed in the embodiments above and will not be repeated here. After abstracting the wheel-legged robot into an nth-order inverted pendulum model, the dynamic equations include the rotational torques of n rotary joints. This allows the balance control process to control the rotation of the n rotary joints. The rotation of the rotary joints causes changes in the pose and motion state of the links connected to them, enabling all n mechanisms to participate in the balance control process. This makes the wheel-legged robot's posture more flexible during balance control, helping to adapt to balance control requirements in different scenarios and improving the robustness of the balance control method.

[0265] The following examples illustrate the method for establishing a sliding surface.

[0266] In some implementations, the sliding surface comprises n sliding surfaces; wherein, the n sliding surfaces are used to constrain the rotational torque of n rotary joints. Optionally, no two of the n sliding surfaces are the same, and the n sliding surfaces are used together in the calculation of the rotational torque.

[0267] Step 1130 in the above embodiment: Establishing the sliding surface based on the state variables at the first moment may include the following steps:

[0268] Step 1133 (not shown in the figure): For the i-th sliding surface among the n sliding surfaces, determine two sliding parameters, where i is a positive integer less than or equal to n.

[0269] In some embodiments, the sliding surface is associated with at least one element in the state variables, which can also be referred to as state parameters, such as state variables. Included They all belong to state parameters.

[0270] For example, a certain sliding surface is related to the deflection angle and acceleration of the first link in the state variables. Another example is that a certain sliding surface is related to the deflection angle and acceleration of the i-th link. Optionally, each of the n sliding surfaces is related to a first state parameter in the state variables. For example, the first state parameter is the angular velocity of the moving wheel, meaning all n sliding surfaces are related to the angular velocity of the moving wheel.

[0271] Optionally, the sliding mode parameter is a real number. The sliding mode parameter is used to characterize the proportional relationship between the state parameters in the state variables and the sliding surface. For example, the sliding mode parameters between different sliding surfaces are not shared; that is, among the n sliding surfaces, there are at least two sliding surfaces whose sliding mode parameters are not equal.

[0272] In some embodiments, the number of state parameters used in creating the sliding surface is greater than the number of sliding parameters. Optionally, for any one of the n sliding surfaces, the number of state parameters used in the creation process of that sliding surface is greater than the number of sliding parameters used. This method helps to ensure that a definite rotational torque can be calculated based on the sliding surface and the dynamic model parameters.

[0273] For example, when creating a sliding surface, three state parameters and two sliding parameters are required from the state variables. Alternatively, when creating a sliding surface, four state parameters and three sliding parameters are required from the state variables.

[0274] Since the number of sliding parameters used in creating a sliding surface is less than the number of state parameters, the coefficient of at least one state parameter in the sliding surface is equal to 1. That is, these sliding parameters do not have corresponding state parameters. The remaining state parameters required to create a certain sliding surface correspond one-to-one with the sliding parameters.

[0275] In some embodiments, for the i-th sliding surface among n sliding surfaces, the i-th sliding surface includes at least two sliding parameters. Optionally, the state parameters included in different sliding surfaces are not completely the same, and the number of state parameters required to create each sliding surface is equal. For example, the number of sliding parameters included in the n sliding surfaces is equal. Suppose that creating the i-th sliding surface uses 3 state parameters, creating the (i+k)-th sliding surface also uses 3 state parameters, where k is a positive integer, and i+k is less than or equal to n.

[0276] In some embodiments, the number of sliding parameters included in the i-th sliding surface is related to the number of state parameters used to establish the i-th sliding surface. Optionally, the number of state parameters used to establish the i-th sliding surface is one more than the number of sliding parameters used to establish the i-th sliding surface. To ensure that a definite solution for the rotational torque of n rotary joints can be calculated using the sliding surface and dynamic model parameters, the i-th sliding surface is associated with at least 3 state parameters in the state variables at the first moment. Optionally, the at least 3 state parameters associated with the i-th sliding surface are not exactly the same as the at least 3 state parameters associated with the l-th sliding surface, where l is a positive integer less than or equal to n.

[0277] To improve the reliability of the determined rotational torque, the sliding mode parameters need to meet the system stability conditions. For details on this part, please refer to the following example.

[0278] Optionally, the number of state parameters used to create the n sliding surfaces is equal. For example, any one of the n sliding surfaces includes 3 state parameters.

[0279] The following section uses the creation of the i-th sliding surface as an example to introduce and explain the method for creating a sliding surface.

[0280] Generally speaking, the more state parameters related to a certain sliding surface in the state variables at the first moment, the more reference information is used in the calculation of the rotational torque of n rotary joints; the mutual constraints between the various state variables help to improve the accuracy of the rotational torque of the n rotary joints determined based on the parameters of the n sliding surfaces and the dynamic model, and help to improve the effect of balance adjustment.

[0281] Optionally, creating the i-th sliding surface requires at least 3 state parameters and at least 2 sliding parameters.

[0282] For example, the state parameters used to create the i-th sliding surface include: the deflection angle of the i-th link, the angular velocity of the moving wheel, and the angular velocity of the i-th link. In this case, the two sliding parameters used to create the i-th sliding surface are: the sliding parameter corresponding to the deflection angle of the i-th link and the sliding parameter corresponding to the angular velocity of the moving wheel. For example, the i-th sliding surface is associated with four state parameters in the state variables. The i-th sliding surface is associated with the deflection angle of the i-th link, the angular velocity of the moving wheel, the angular velocity of the i-th link, and the deflection angle of the (i+m)%n-th link, where m is a positive integer not divisible by n, and "%" indicates the remainder of division. In this case, the i-th sliding surface includes at least three sliding parameters: the sliding parameter corresponding to the deflection angle of the i-th link, the sliding parameter corresponding to the angular velocity of the moving wheel, and the sliding parameter corresponding to the deflection angle of the (i+m)%n-th link. It should be noted that the number of state parameters related to the sliding surface and the number of sliding parameters are designed according to actual needs, and this application does not limit them here.

[0283] In some embodiments, the robust controller is designed with multiple sliding surface creation schemes. The number of sliding parameters and state parameters used to create the i-th sliding surface are not entirely the same in different schemes. For example, in the first sliding surface creation scheme, creating the i-th sliding surface requires 2 sliding parameters and 3 state parameters; in the second sliding surface creation scheme, creating the i-th sliding surface requires 5 sliding parameters and 6 state parameters. That is, the number of sliding parameters used in the first sliding surface creation scheme is not equal to the number of sliding parameters required in the second sliding surface creation scheme; the number of state parameters used in the first sliding surface creation scheme is not equal to the number of state parameters required in the second sliding surface creation scheme.

[0284] Optionally, during the balance control process, the robust controller selects the target creation scheme from multiple sliding surface creation schemes based on the actual situation, and determines the number of sliding parameters required to create the i-th sliding surface based on the target creation scheme.

[0285] For example, when the robust controller has high computational performance, it selects from multiple sliding surface creation schemes: the scheme involving more sliding parameters is chosen as the target creation scheme. For example, the robust controller determines the target creation scheme based on the task currently being performed by the wheeled robot; if the task requires high balance and the robust controller has good computational capabilities, it selects the sliding surface creation scheme involving more sliding parameters from the multiple schemes as the target creation scheme. For example, when the robust controller's computational performance is average or there are power consumption requirements, it selects the sliding surface creation scheme including fewer sliding parameters from at least one set of sliding surface creation schemes as the target creation scheme. For example, the robust controller receives control commands from the remote controller, which indicate the target creation scheme selected from the multiple sliding surface creation schemes.

[0286] Step 1136 (not shown in the figure): Based on at least two sliding mode parameters and the state variables at the first moment, establish the i-th sliding surface.

[0287] In some embodiments, the state variables include the deflection angles of the n links, the angular velocity of the moving wheel, and the angular velocities of the n links. Optionally, the robust controller creates the i-th sliding surface based on at least three state parameters and at least two sliding mode parameters from the state variables at the first moment.

[0288] Optionally, step 1136: Based on at least two sliding mode parameters and the state variables at the first moment, establish the i-th sliding surface, including the following sub-steps:

[0289] Sub-step 1136-a: Based on the first sliding mode parameter among at least two sliding mode parameters, process the deflection angle of the i-th link to obtain the processing result of the i-th link.

[0290] Optionally, the first sliding mode parameter is the coefficient corresponding to the deflection angle of the i-th link, and the second sliding mode parameter is the coefficient corresponding to the angular velocity of the moving wheel link.

[0291] Optionally, the processing result of the i-th link refers to the result obtained by processing the deflection angle of the i-th link using the first sliding mode parameter during the construction of the i-th sliding surface. For example, the method of processing the deflection angle of the i-th link using the first sliding mode parameter includes, but is not limited to: performing basic calculations (such as addition, subtraction, multiplication, and division) using the first sliding mode parameter and the deflection angle of the i-th mechanism to obtain the processing result of the i-th link. For instance, multiplying the first sliding mode parameter by the angular velocity of the i-th link yields the processing result of the first link.

[0292] Sub-step 1136-b involves processing the angular velocity of the moving wheel based on the second sliding mode parameter from at least two sliding mode parameters to obtain the processing result for the moving wheel. Optionally, during the establishment of the i-th sliding surface, the processing result for the moving wheel refers to the result obtained by processing the angular velocity of the moving wheel using the second sliding mode parameter. Exemplarily, methods for processing the angular velocity of the moving wheel using the second sliding mode parameter include, but are not limited to: performing basic calculations using the second sliding mode parameter and the angular velocity of the moving wheel. For example, multiplying the second sliding mode parameter by the angular velocity of the moving wheel yields the processing result for the moving wheel.

[0293] It should be noted that the second sliding parameters are different for different sliding surfaces. That is, for different sliding surface creation processes, the angular velocity of the moving wheel is processed using the second sliding parameters, and the processing results of the moving wheel are different.

[0294] Sub-step 1136-c: Based on the processing results of the i-th link, the processing results of the moving wheel, and the angular velocity of the i-th link, establish the i-th sliding surface.

[0295] In some embodiments, the i-th sliding surface is proportional to the processing result of the i-th link, the i-th sliding surface is proportional to the processing result of the moving wheel, and the first sliding surface is proportional to the angular velocity of the i-th link. For example, the robust controller adds the processing result of the i-th link, the processing result of the moving wheel, and the angular velocity of the i-th link to obtain the i-th sliding surface.

[0296] Optionally, the robust controller establishes the i-th sliding surface based on the processing result of the i-th link, the processing result of the moving wheel, the angular velocity of the i-th link, and the sliding mode constant. The sliding mode constant can be a preset value.

[0297] The robust controller determines n sliding surfaces using the method described above for creating the i-th sliding surface.

[0298] n sliding surfaces can be represented by the following formula:

[0299]

[0300] Where s represents a sliding surface matrix consisting of n sliding surfaces; This includes the first sliding mode parameters corresponding to the deflection angles of the n links. This includes the second sliding parameters corresponding to the angular velocity of the moving wheel in each of the n sliding surfaces. For explanations of other parameters, please refer to the above introduction, which will not be repeated here.

[0301] By establishing n sliding surfaces in this way, the rotational torque of the n rotating joints can be calculated based on the sliding surfaces and dynamic model parameters. At the same time, the number of sliding parameters and state parameters used in the creation of the screen is controlled. This helps to control the computational overhead of calculating the rotational torque of the n rotating joints, which helps to speed up the determination of the rotational torque, reduce power consumption, and achieve balance control of the robot.

[0302] After explaining the design method of the dynamic model parameters and n sliding surfaces in the robust controller, the following section introduces the relevant content of the robust controller design for step 1140: calculating the rotational torque of the n rotary joints based on the sliding surfaces and dynamic model parameters.

[0303] After establishing n sliding surfaces based on the state variables, the robust controller determines the first-order derivative formulas of the n sliding surfaces with respect to time, and uses torque to express the angular acceleration of the n links according to the dynamic equation (Formula 2), thus obtaining the following Formula 3:

[0304]

[0305] in, Let each of the n sliding surfaces have its first derivative with respect to time. It is the offset parameter matrix f[] in Formula 2 above. f q It is the offset parameter matrix f[] in Formula 2 above. It is the proportional parameter matrix g[] in Formula 2 above. g q In the formula 2 above, the proportionality parameter matrix g[] Please refer to the above description for other parameters, which will not be repeated here.

[0306] Rearranging terms in formula 3 yields formula 4:

[0307]

[0308]

[0309]

[0310] Where sgn() is the sign function, Let s be a constant matrix. T Let be the transpose of the sliding matrix s, which includes n sliding surfaces, where |s| represents the absolute value of the n sliding surfaces. This represents the first derivative of the n sliding surfaces with respect to time. Other parameters are described above and will not be repeated here.

[0311] Optionally, Formulas 3 and 4 are calculated in real time by the robust controller during the balance control process.

[0312] Utilizing the properties of the sliding surface, when the sliding surface is controlled to be equal to 0, the state parameters related to the sliding surface will gradually approach 0 along the sliding surface, meaning the wheeled robot returns to its equilibrium state. The state variables (state parameters in the model) satisfy the following on the sliding surface:

[0313]

[0314] The angular velocities of the n links can be expressed in reverse using the above formula:

[0315]

[0316] The robust controller determines the angular velocities of n links and calculates their first derivatives with respect to time, yielding Formula 5, which represents the angular acceleration of the n links:

[0317]

[0318] Since q is a vector containing the deflection angles of n links, we can decompose q: Let α represent q1 (the deflection angle of the first link), and let β represent... That is, β includes the deflection angles of all the links except the first link out of the n links. Therefore, the dynamic equations based on the nth-order inverted pendulum model can be written as follows:

[0319]

[0320]

[0321]

[0322] In this formula This represents the angular acceleration of the first link; This represents the angular acceleration of each of the n links except the first link; This represents the inertial quantities in the inertia matrix M related to the moving wheels; Represents the inertia matrix M The inertial quantity related to the first link; It refers to the inertia matrix M. The inertial quantity that is independent of the first link; m αα m represents the inertia matrix M qq The inertial quantity related to the first link; m ββ m represents the inertia matrix M qq The inertial quantity that is independent of the first link; c α c represents the deflection force received by the first link; β This represents the eccentric force acting on each of the n links except the first link; g α g represents the gravitational force acting on the first mechanism. β τ1 represents the gravitational force acting on each of the n links except the first link; τ2 represents the rotational torque of the first rotary joint; and τ3 represents the rotational torque of each of the n rotary joints except the first rotary joint.

[0323] Adding the three formulas included in the above dynamic equations yields Formula 6:

[0324]

[0325] In formula 6 merged into Formula 7 is obtained:

[0326]

[0327] Use formula 5 to apply formula 7 By performing an equal substitution, we obtain Formula 8:

[0328]

[0329] Equation 8 is used to represent... Formula 9:

[0330]

[0331] Formulas 5 and 9 are used to represent... Formula 10:

[0332]

[0333] right Combining the equations, we obtain the state equations in matrix form:

[0334]

[0335] in,

[0336] I is the identity matrix.

[0337] Optionally, the system stability criterion is related to the state equations determined based on the dynamic equations and the sliding surface. For example, the stability criterion is related to the coefficient matrix in the state equations; to ensure the system stability criterion is met, the sliding parameters need to satisfy the coefficient matrix of the state equations.

[0338] The eigenvalues ​​are all distributed in the left half of the complex plane. This state equation can be used as a matrix inequality constraint, that is, the values ​​of the sliding mode parameter are distributed in the left half of the complex plane of the coefficient matrix.

[0339] Optionally, in the balance control process of this application, the robust controller determines the rotational torque of the n rotary joints according to Equation 4 and the state equation. During the balance control process, the robust controller calculates the rotational torque τ of the n rotary joints according to Equation 4. Equation 4 involves the sliding surface matrix s composed of n sliding surfaces, the dynamic model parameters in Equation 2, and also the sliding parameters.

[0340] The robust controller obtains the state variables at the first moment and substitutes them into the corresponding formula to determine the dynamic model parameters and create n sliding surfaces. The sliding parameters in the sliding surfaces satisfy the system stability condition. Then, the robust controller ensures that the sliding parameters satisfy the matrix inequality constraints formed by the state equation above, while calculating the rotational torque τ of the n rotating joints based on formula 4.

[0341] For example, in the balance control method provided in this application, the sliding mode parameters that satisfy the matrix inequality constraints are first determined, and then n sliding surfaces are constructed using the sliding mode parameters. The robust controller determines the state equations based on the dynamic equations and the representation of the sliding surfaces (at this point, the sliding mode parameters in the sliding surfaces are unknown). The robust controller determines the values ​​or ranges of the sliding mode parameters under the system stability condition that the eigenvalues ​​of the coefficient matrix satisfying the state equations are in the left half-plane, thus obtaining the actual values ​​of the sliding mode parameters required in the process of creating the sliding surfaces. The robust controller substitutes the sliding mode parameters into the sliding surfaces and calculates the rotational torques of the n rotating joints based on the sliding surfaces and the dynamic model parameters.

[0342] For example, the balance control method provided in this application first selects n sliding mode parameters corresponding to each sliding mode surface within a certain numerical range, constructs n sliding mode surfaces using these sliding mode parameters, and calculates the rotational torque τ of n rotary joints using Formula 4; then, the robust controller verifies whether the four sliding mode parameters satisfy the above matrix inequality constraints. If all four sliding mode parameters satisfy the above matrix inequality constraints, then the rotational torque τ of the n rotary joints can be used to control the n rotary joints; if the n sliding mode parameters do not satisfy the above matrix inequality constraints, then the determined rotational torque τ of the n rotary joints cannot be used.

[0343] The following examples illustrate the method for determining sliding mode parameters.

[0344] In some embodiments, for the i-th sliding surface among n sliding surfaces, the robust controller determines at least two sliding parameters, including: determining a first sliding parameter from at least one sliding parameter from the (2i-1)-th predicted parameter set, and determining a second sliding parameter from at least one sliding parameter from the 2i-th predicted parameter set; wherein the sliding parameters included in the (2i-1)-th and 2i-th predicted parameter sets respectively satisfy the constraints of the stability criterion.

[0345] In this embodiment, after determining the state at the first moment, the computer device first uses the above state equation as a matrix inequality constraint to determine the solution sets corresponding to the sliding parameters required for each of the n sliding surfaces, thus obtaining 2n sets of predicted parameters. That is, for any one of the 2n sets of predicted parameters, the sliding parameters in the predicted parameter set satisfy the system stability principle.

[0346] During the creation of the i-th sliding surface, the first sliding parameter is determined from at least two sliding parameters in the (2i-1)-th prediction parameter set, and the second sliding parameter is determined from at least two sliding parameters in the 2i-th prediction parameter set. The i-th sliding surface is created based on the state variables at the first moment and the above at least two sliding parameters. Subsequently, the robust controller calculates the rotational torque of the n rotating joints according to Formula 4 above.

[0347] It should be noted that this embodiment uses the creation of a sliding surface using two sliding parameters as an example to illustrate the method for determining the sliding parameters. If creating a sliding surface requires k sliding parameters, then during the creation of the i-th sliding surface, one sliding parameter needs to be obtained from each of the k prediction parameter sets corresponding to the i-th sliding surface, resulting in k sliding parameters. When k equals 2, the method described in the previous embodiment applies. k is a positive integer greater than or equal to 2, and the maximum value of k is set according to actual needs; this application does not impose a limitation on this.

[0348] This method helps avoid selecting sliding mode parameters that do not meet system stability requirements, which would result in the torques of the n rotary joints calculated using these sliding mode parameters being unable to control the balance of the wheeled robot. It improves the usability of the determined torques of the rotary joints, helps avoid the calculations required when the computer equipment redetermines the torques, and helps shorten the time required to determine the torques of the n rotary joints.

[0349] The following examples illustrate a method for adjusting the posture of a wheeled robot using the rotational torque of n rotary joints.

[0350] In some implementations, step 1150 in the above embodiments, which involves controlling the n rotary joints based on their rotational torques, includes the following sub-steps:

[0351] Sub-step 1150a: For any one of the n rotary joints, control the corresponding rotary motor to rotate according to the rotational torque of the rotary joint.

[0352] In this embodiment, after the robust controller calculates the rotational torque of the n rotating joints, it sends the rotational torque of the n rotating joints to the corresponding rotating motors of each rotating joint. For any one of the n rotating joints, after receiving the rotational torque of the rotating joint, the corresponding rotating motor rotates according to the rotational torque of the rotating joint, driving at least one link connected to the rotating joint to move, so that the deflection angle and angular velocity of at least one link among the n links change.

[0353] Optionally, after receiving the rotational torque of their respective rotational joints, the rotational motors corresponding to the n rotational joints rotate simultaneously at the second moment, causing the posture of the wheel-legged robot to change.

[0354] The position and state of the wheeled robot are adjusted by n rotating motors, and the balance control of the wheeled robot is performed. This allows multiple links in the wheeled robot to participate in the balance control process, which helps to improve the efficiency of adjusting the posture and motion state of the wheeled robot according to the torque.

[0355] In one of the embodiments above, after determining the rotational torque of the n rotating joints, the robust controller directly sends the rotational torque of the n rotating joints to the rotating motors, controls the rotating motors corresponding to the n rotating joints, and changes the deflection angle of the n links, thereby changing the posture of the wheeled robot and adjusting its balance state.

[0356] After the robust controller determines the rotational torque of the n rotating joints, it can also determine the rotational torque corresponding to each joint of the wheeled robot based on the rotational torque of the n rotating joints. This allows more links in the wheeled robot, in addition to the n links, to participate in the balance control process, and realizes the combination of the balance control process and the motion process. The following are several examples to illustrate this method.

[0357] In some embodiments, step 1150 in the above embodiments: controlling the rotary motors corresponding to the n rotary joints according to the rotational torque of the n rotary joints at the second moment may include the following sub-steps:

[0358] Sub-step 1153: Calculate the task acceleration of the wheeled robot at the second moment based on the rotational torque of the n rotating joints. The task acceleration includes the acceleration related to the center of mass of the wheeled robot.

[0359] Optionally, the center of mass of a wheel-legged robot refers to an imaginary point that gathers all the mass of the wheel-legged robot. The center of mass of a wheel-legged robot is related to the mass distribution of the various mechanisms of the wheel-legged robot, the posture of the wheel-legged robot, and other factors.

[0360] Optionally, the task acceleration includes, but is not limited to, at least one of the following: the acceleration of the wheeled robot's center of mass along the x-axis, the acceleration of the wheeled robot's center of mass along the y-axis, the acceleration of the wheeled robot's center of mass along the z-axis, the angular acceleration of the wheeled robot's center of mass rotating about the x-axis, the angular acceleration of the wheeled robot's center of mass rotating about the y-axis, and the angular acceleration of the wheeled robot's center of mass rotating about the z-axis. Exemplarily, the task acceleration is observed in the operating space or in the joint space.

[0361] The operational space refers to the space related to the operational tasks of a wheeled robot. The physical quantities involved in this operation are observed in a Cartesian coordinate system (world coordinates), while joint space refers to the coordinate system used to observe the motion states of the joints. For example, physical quantities in the operational space and those in the task space can be converted. The position information of a mechanism in the operational space and its position in the joint space can be converted using kinematic formulas, and the velocity in the operational space and the velocity in the joint space can be converted using a Jacobian matrix; where the Jacobian matrix is ​​a matrix of first-order partial derivatives arranged in a specific manner.

[0362] In some embodiments, task acceleration includes: acceleration related to the center of mass of the wheeled-legged robot during the execution of a task in its operating space. Optionally, task acceleration is related to the position of the center of mass of the wheeled-legged robot relative to the robot; the method for determining task acceleration is described below.

[0363] Operational space tasks can be understood as the tasks that a wheeled-legged robot needs to perform in its operational space. At any given moment, the operational tasks that the wheeled-legged robot needs to perform are determined based on the robot's motion scenario. Optionally, the wheeled-legged robot may perform multiple operational space tasks at a single moment.

[0364] For example, the types of operational space tasks include, but are not limited to, at least one of the following: support wheel tasks, torso vertical tasks, torso posture tasks, swing wheel tasks, and center of mass tasks; wherein, support wheel tasks refer to tasks related to at least one of the moving wheels of a wheel-legged robot that is in contact with the contact surface; torso vertical tasks refer to tasks of the torso mechanism (also known as the body mechanism) of a wheel-legged robot in the z-axis direction of the world coordinate system; torso posture tasks refer to tasks related to the Euler angles of the torso mechanism of a wheel-legged robot, where Euler angles include roll angle, pitch angle, and yaw angle; swing wheel tasks refer to tasks related to at least one of the moving wheels of a wheel-legged robot that is not in contact with the contact surface; and center of mass tasks refer to tasks related to the center of mass of the wheel-legged robot.

[0365] Sub-step 1156: Based on the task acceleration, determine the force and torque commands for the whole body joints of the wheeled robot at the second moment. The whole body joints include n rotational joints.

[0366] In some embodiments, a whole-body joint refers to an active joint in a wheeled-legged robot. Optionally, the types of whole-body joints include: rotary joints and linear joints. For example, a rotary joint corresponds to a rotary motor, which changes the deflection angle, angular velocity, etc. of the link connected to the rotary joint by rotating the rotary motor. A linear joint corresponds to a linear motor, which is used to change the length of the mechanism. For example, the leg mechanism (first link) of a wheeled-legged robot includes a linear motor, which can adjust the length of the leg mechanism.

[0367] In some embodiments, after determining the task acceleration, the observation module sends the task acceleration to the whole-body dynamics controller; the whole-body dynamics controller determines the force and torque commands corresponding to each joint in the whole body based on the task acceleration. Optionally, if the wheel-legged robot performs multiple workspace tasks in the first moment, in addition to determining the task acceleration related to the center of mass of the wheel-legged robot, it is also necessary to calculate the other task accelerations corresponding to each of the other workspace tasks. Please refer to the embodiments below for details of this process.

[0368] Optionally, the whole-body dynamics controller incorporates whole-body dynamics equations for the wheel-legged robot. These equations characterize the relationship between the robot's posture and its overall state variables. The whole-body dynamics equations are derived by establishing a whole-body dynamics model of the wheel-legged robot and through kinematic derivation.

[0369] Alternatively, the whole-body dynamics model can be expressed by the following formula:

[0370]

[0371] in, H represents the joint space inertia matrix, which includes the inertia quantities of the rigid bodies of all joints in the body. The C matrix represents the offset forces acting on each joint in the joint space. It includes the sum of the Coriolis force, centripetal force, and gravity received by the complete wheel-legged robot. Represents the selection matrix; Represents the Jacobian matrix at the contact point; Represents the contact force vector, including at least one contact force acting on the wheel-legged robot; N represents generalized position, generalized velocity, and generalized acceleration, respectively. G N represents the total degrees of freedom of the robot. G =N f +N J N f For floating degrees of freedom, N J The active joint degree of freedom refers to a joint that can move automatically. N C N represents the number of contact forces. C Equal to the number of contact points n C Dimension N of a single contact force D The product of ∈{0,1,2,3}. It should be noted that τ is a vector including the force and torque commands of all joints in the body, while τ is a vector including the rotational torques of n rotational joints. The two meanings are not the same.

[0372] Optionally, the whole-body dynamics equations are constructed based on the rigid body dynamics model of the wheeled-legged robot. The whole-body dynamics equations are used to describe the relationship between the posture of the complete wheeled-legged robot and the motion state of each active joint and the external forces it is subjected to.

[0373] In some embodiments, the force and torque commands include at least one of the following commands: a force command and a torque command. Optionally, the force and torque commands for a rotary motor include a torque command, which is used to control the rotary motor to rotate according to the torque command; the force and torque commands for a linear motor include a force command, which is used to control the linear motor to perform linear motion.

[0374] Sub-step 1159: Control all joints of the body according to the force and torque commands at the second moment.

[0375] After determining the force and torque commands for all joints, the whole-body dynamics controller sends the force and torque commands to the joint motors corresponding to each joint. The joint motors corresponding to each joint then perform linear motion or rotation according to the received force and torque commands, causing the wheeled robot's posture to change and thus adjusting the wheeled robot's balance.

[0376] Optionally, after the rotation motors corresponding to the n rotation motors receive force and torque commands, the rotation of the rotation motors causes changes in the deflection angle, angular velocity, and angular acceleration of at least one wheeled robot corresponding to the n-order inverted pendulum model; for the active joints not considered in the n-order inverted pendulum (i.e., other active joints besides the n joints), these active joints begin to move after receiving force and torque commands.

[0377] For example, after receiving force and torque commands, the linear motors in the leg mechanism change the length of the leg mechanism, thereby changing the height of the wheeled robot and causing a change in the position of the robot's center of mass along the z-axis. For example, for the other rotary motors besides the n rotary motors, these other rotary motors rotate according to the received force and torque commands.

[0378] It should be noted that the rotational torque of n rotational joints, and the numerical values ​​of the force and torque commands corresponding to n rotational joints may be different. In the process of determining the force and torque commands of the whole body joints by the whole body dynamics controller, state variables related to the workspace task of the wheel-legged robot are used, and more rotational joints are involved.

[0379] The force and torque commands generated by the whole-body dynamics controller are used for balance control, which enables more links (more than n links) to participate in balance control, enriching the posture of the wheel-legged robot during balance adjustment. It also realizes the combination of balance adjustment and operation space tasks in motion process. Balance adjustment is completed by combining a robust controller and a whole-body dynamics controller, which provides a new approach to balance control.

[0380] The following examples illustrate the method for determining task acceleration.

[0381] In some embodiments, sub-step 1153: Calculate the task acceleration of the wheel-legged robot at the second moment based on the rotational torques of the n rotational joints. Sub-step 1153 can be performed by an observation module, which is used to determine the task acceleration. Optionally, the observation module is designed with a method for converting the rotational torques of the n rotational joints into task acceleration.

[0382] Optionally, sub-step 1153 may include the following sub-steps:

[0383] Sub-step 1153-a: Determine the angular acceleration at the second moment based on the dynamic equation and the rotational torque of the n rotating joints.

[0384] The dynamic equation refers to the dynamic equation obtained based on the nth-order inverted pendulum model. This dynamic equation is related to the motion state of n mechanisms and the rotational torque of n rotating joints.

[0385] Optionally, after the robust controller of the wheeled robot determines the torque of the n rotating joints based on the state variables at the first moment, it sends the torque of the n rotating joints to the observation module in the control system. The observation module then uses Formula 2 from above to determine the torque of the n rotating joints.

[0386]

[0387] Calculate the angular acceleration of the wheeled robot at the second moment.

[0388] Optionally, the angular acceleration at the second moment refers to the angular acceleration related to the nth-order inverted pendulum model. For example, the angular acceleration at the second moment includes the angular accelerations of the n links at the second moment and the angular acceleration of the moving wheel at the second moment.

[0389] Sub-step 1153-b: Based on the state variables at the first moment and the angular acceleration at the second moment, determine the expected incremental position and expected incremental velocity of the center of mass of the wheel-legged robot at the second moment. The expected incremental position is used to characterize the distance between the projection of the center of mass of the wheel-legged robot on the contact surface and the virtual contact point in the first direction. The expected incremental velocity is used to characterize the rate of change of the distance in the first direction. The virtual contact point refers to the center of each contact point between the wheel-legged robot and the contact surface.

[0390] In some embodiments, when at least two moving wheels of a wheeled robot are in contact with a contact surface, the virtual contact point refers to the center of the line connecting the contact points of the at least two moving wheels with the contact surface.

[0391] In some embodiments, the desired position increment is used to characterize the position of the center of mass of the legged robot relative to the legged robot. The desired increment velocity is the first derivative of the desired position increment with respect to time, that is, the desired position increment is used to characterize the rate of change of the position of the legged robot's center of mass relative to the legged robot.

[0392] Optionally, the desired position increment and desired velocity increment refer to the relative position and relative velocity of the wheeled robot's center of mass in the first direction, respectively. The first direction refers to the forward direction of the wheeled robot (i.e., the x-axis direction in the world coordinate system). If the wheeled robot involves left-right motion (i.e., the y-axis direction in the world coordinate system), then the desired velocity increment refers to the component of the relative velocity of the wheeled robot's center of mass in the first direction.

[0393] In some embodiments, after determining the angular acceleration at the second moment, the observation module uses the forward kinematics method to determine the expected incremental position and expected incremental velocity of the center of mass of the wheel-legged robot at the second moment, based on the acceleration at the second moment, the deflection angles of the n links, the rotation angles of the moving wheels, the angular velocities of the n links, the configuration, model, and parameters of the wheel-legged robot.

[0394] Optionally, given that the acceleration at the second moment, the deflection angles of the n mechanisms, the rotation angles of the moving wheels, and the angular velocities of the n mechanisms are known, the position and velocity of the end effector of the wheel-legged robot are calculated based on forward kinematics, thus determining the posture of the wheel-legged robot at the second moment. Based on the position and posture of the wheel-legged robot in the operating space, the position of the center of mass of the wheel-legged robot in the operating space can be calculated, thereby obtaining the desired incremental position and the desired incremental velocity.

[0395] Sub-step 1153-c: Determine the task acceleration of the wheeled robot at the second moment based on the desired incremental position and desired incremental velocity.

[0396] In some embodiments, the observation module determines the reference state variable set of the center of mass of the wheel-legged robot based on the desired incremental position and the desired incremental velocity; the observation module determines the task acceleration based on the reference state variable set of the center of mass of the legged robot and the actual state variable set.

[0397] The reference state variable set is used to characterize the state variables of the wheel-legged robot's center of mass at the second moment, which are inferred from kinematics. Optionally, the reference state variable set includes the desired incremental position Δx. r Expected incremental speed Desired location Expected speed The desired position refers to the position of the center of mass of the wheeled robot in the operation space at the second moment, or the position of the joint in the joint space at the second moment. The desired velocity refers to the first derivative of the desired position with respect to time.

[0398] The actual state variable set is used to characterize the state variables of the wheeled-legged robot obtained through sensor observations. Optionally, the actual state variable set includes: the actual incremental position Δx. a Actual incremental speed actual location actual speed Actual position refers to the location of the center of mass of the wheeled robot in the operating space, and actual velocity refers to the first derivative of the actual position with respect to time.

[0399] Optionally, the wheeled robot's task acceleration at the second moment. The following formula can be used for calculation:

[0400]

[0401] Where K represents the feedback gain matrix. For example, the objective function of the inverted pendulum dynamic equation is constructed using LQR, and then the feedback gain matrix of the inverted pendulum dynamic equation is obtained iteratively with the goal of minimizing the objective function. LQR is constructed based on the quadratic programming optimization method, the essence of which is: under linear constraints, find a multidimensional vector such that the quadratic objective function of the multidimensional vector is minimized (or maximized).

[0402] In some embodiments, the observation module sends the calculated task acceleration of the wheeled-legged robot to the whole-body dynamics controller. The whole-body dynamics controller generates force and torque commands for the joints of the wheeled-legged robot based on the task acceleration. The whole-body dynamics controller sends the force and torque commands for the joints to the joint motors corresponding to each active joint, thereby controlling the movement of the whole-body joints and adjusting the posture of the wheeled-legged robot to maintain its balance.

[0403] In some embodiments, the observation module generates force and torque commands for the whole-body joints of the wheeled-legged robot based on the task acceleration, including the following steps: the whole-body dynamics controller determines the desired operational space task of the wheeled-legged robot at a second moment, the desired operational space task including at least one task acceleration corresponding to the operational space task; the whole-body dynamics controller performs spatial transformation on the desired operational space task to obtain the desired joint space task; the whole-body dynamics controller inputs the desired joint space task into the whole-body dynamics equations of the wheeled-legged robot to obtain the force and torque commands for the whole-body joints.

[0404] Optionally, the desired operation space task includes the task acceleration at the second time step. Desired operation space task It can be expressed as:

[0405]

[0406] in, This refers to the first acceleration corresponding to the task of the support wheel. This refers to the second acceleration corresponding to the vertical direction of the torso during the task. This refers to the third acceleration corresponding to the torso posture task. The fourth acceleration corresponding to the swing wheel task. This refers to the task acceleration corresponding to the center of mass task. Optionally, the operational space task of the wheeled robot is different at different times, that is, it is expected that some acceleration values ​​included in the operational space task can be 0.

[0407] Optionally, the relationship between the acceleration in the operation space included in the operation space task and the joint space velocity and acceleration in the joint space task is expected to be as follows:

[0408]

[0409] Among them, J t The Jacobian matrix represents the task in the operation space. The derivative of the Jacobian matrix number of the operation space task.

[0410] According to the whole-body dynamics equation: With the expected joint space task Able to calculate Here, τ represents the force and torque commands for all joints in the body.

[0411] To improve the rationality of force and torque commands for the joints throughout the body, at least one constraint was added during the solution of the whole-body dynamics equations. The constraint includes, but is not limited to, at least one of the following: joint physical constraints and frictional constraints.

[0412] Optionally, joint physical constraints are used to limit the force and torque commands based on the characteristics of the joint motors corresponding to each joint in the wheeled robot. Joint physical constraints can be expressed as:

[0413] τ lb ≤τ≤τ ub

[0414] Where τ represents a matrix containing force and torque commands for all joints in the body, τ lb τ represents a matrix containing the minimum values ​​of the force and torque commands corresponding to each active joint. ub This represents a matrix containing the maximum values ​​of the force and torque commands corresponding to each active joint.

[0415] Optionally, friction constraint refers to limiting the contact force f at the i-th contact point. i The friction cone constraint should be satisfied. To reduce nonlinearity, the friction cone can be approximated as a friction angle cone, and the friction inequality constraint can be expressed as:

[0416]

[0417] Where, n x n y n z Let μ represent the unit orthogonal basis of the lower edge contact surface of the operating space system, respectively. i f represents the coefficient of friction. i,zlb f i,zub These represent the minimum and maximum values ​​of the non-negative positive pressure perpendicular to the contact surface, respectively.

[0418] The following describes the solution process for the whole-body dynamics equations. Optionally, this process can be completed by an optimizer: the whole-body dynamics equations are expressed as a polynomial solution in the form AX = B.

[0419] in, Using a quadratic programming optimizer, we construct the objective function, which can be expressed as: J = min(AX - B) T Q(AX-B)+X T RX; where Q and R are weight matrices, the minimum value of the objective function is calculated using a quadratic programming optimizer, resulting in... The variables in the data are obtained, which means the force and torque commands τ of all the joints in the body are obtained.

[0420] This method enables the independent determination of the force and torque commands corresponding to each joint in the body. Even in scenarios where multiple real wheels of a wheel-legged robot do not overlap in the first direction, it is possible to control at least one real wheel and at least one real first rotational joint.

[0421] The following describes the method for determining each desired acceleration in the desired operation space task. The method for determining each element in the desired operation space task is preset. Optionally, the desired acceleration is determined based on the actual acceleration and the reference acceleration.

[0422] 1. The first acceleration corresponding to the support wheel task The method for determining it.

[0423] Since the support wheel rolls on the contact surface and there is no relative sliding between it and the contact surface, the expected acceleration of the support wheel is... It is always equal to 0; the expected acceleration of the support wheel can be expressed as If the wheel-legged robot operates in four-wheel mode, it has four support wheels, each with a task in the x-direction. Assuming the support wheels remain on the ground, their height along the z-axis in the world coordinate system is constant; therefore, the tasks on the four support wheels in the z-axis direction are not considered for now. Similarly, assuming the wheel-legged robot does not perform left or right turns or translations along the y-axis in four-wheel mode, the tasks on the four support wheels in the y-axis direction are also not considered for now.

[0424] If the wheel-legged robot operates in dual-wheel balance mode, it has two support wheels and two oscillating wheels. The tasks of the support wheels are not considered here, as these are already covered in the "Center of Mass Tasks" section below. The tasks of the two oscillating wheels each include two dimensions: the x-axis and the z-axis. If we disregard the left-right turning motion and the translational motion along the y-axis of the wheel-legged robot in dual-wheel balance mode, then the tasks of the two oscillating wheels in the y-direction are not considered here.

[0425] 2. The second acceleration corresponding to the vertical direction of the torso during the task. The method for determining it.

[0426] In wheel-legged robots, cameras capture changes in the height of the contact surface within the field of view to obtain positional information of the supporting wheels at multiple future moments. Using planning methods such as spline curve interpolation, formulas are determined to express these future positional information, and the vertical reference position of the torso mechanism is then determined based on these formulas. speed acceleration Based on the IMU sensor readings and the deflection angles (joint angles) and angular velocities of the mechanism, the actual vertical position of the torso mechanism is calculated using forward kinematics. speed Construct a PD (Proportional Derivative) feedback controller for the task in the vertical direction of the torso, and calculate the desired acceleration in the vertical direction of the torso, i.e., the second acceleration.

[0427]

[0428] 3. The third acceleration corresponding to the torso posture task The method for determining it.

[0429] In some scenarios (such as climbing stairs), the torso mechanism of a wheel-legged robot needs to remain as vertical as possible and should not rotate. This means the Euler angle reference trajectory, composed of the roll, pitch, and yaw angles of the torso mechanism, must be zero. The actual Euler angles of the wheel-legged robot are calculated based on the observations from the IMU sensor using the forward kinematics formula. angular velocity The desired angular acceleration of the torso can be calculated using the PD feedback controller corresponding to the torso posture task; that is, the third acceleration.

[0430]

[0431] 4. The fourth acceleration corresponding to the swing wheel task. The method for determining it.

[0432] In scenarios involving dual-wheel balancing, such as when the two oscillating wheels are required to assume specific postures and movements while moving on the ground, or when there is a switching process between the supporting wheel leg and the oscillating wheel leg during forward strides on the ground, or when there is a switching process between the supporting wheel leg and the oscillating wheel leg during ascents and descents on stairs, there are corresponding fourth accelerations.

[0433] The reference position of the oscillating wheel in the operating space is planned using spline curve interpolation. speed acceleration Based on the observations from the IMU sensor, joint angles, and angular velocity information, the actual position of the oscillating wheel is calculated using forward kinematics. speed By constructing a PD feedback controller corresponding to the oscillating wheel task, the desired acceleration of the oscillating wheel, i.e., the third acceleration, can be calculated. The PD feedback controller calculates the third acceleration. The formula is as follows:

[0434]

[0435] 5. Task acceleration corresponding to the center of mass task The method for determining it.

[0436] Alternatively, the reference position of the centroid along the forward direction can be planned using heuristic or model-based methods. speed During the movement of a wheeled robot, the center of mass of the wheeled robot not only needs to move continuously along the direction of travel, but also needs to be within the range where the wheeled robot maintains dynamic balance. Based on this, a method for determining the expected acceleration of the center of mass (i.e., task acceleration) is designed.

[0437] Assuming the mass of the wheel-legged robot is concentrated at its center of mass, let the center of the line connecting at least two supporting wheels be the virtual contact point between the inverted pendulum and the contact surface related to the center of mass. Connect the center of mass and the virtual contact point to construct the inverted pendulum model with the center of mass.

[0438] Figure 12 This is a schematic diagram of an inverted pendulum with a center of mass related to an exemplary embodiment of this application.

[0439] The center of mass of the wheel-legged robot is 1210, and the virtual contact point is 1220. Based on the inverted pendulum model related to this center of mass, the dynamic equations related to the center of mass are determined as follows:

[0440]

[0441] The feedback gain matrix K is determined using a linear quadratic programming regulator. The input to the state equation can then be obtained based on the LQR controller.

[0442] in, Represents the actual state variable group. This represents the actual state variable group. Please refer to the above description for other parameters, which will not be repeated here.

[0443] In some embodiments, the balance control method provided in this application is continuously executed during the operation of the wheeled robot. That is, the robust controller continuously acquires the state variables at the current moment, determines the dynamic model parameters and establishes the sliding surface based on the state variables at the current moment, and determines the rotational torque of n rotating joints based on the dynamic model parameters and the sliding surface control.

[0444] In other embodiments, when the IMU sensor detects a drastic change in the deflection angle of certain links (such as the first link) in a wheeled robot, the robust controller executes the balance control method starting from step 1110. For example, if at a certain moment the IMU sensor detects that the change in the deflection angle of the first link is greater than an angle threshold, the IMU sensor sends the detected state parameters to the robust controller, and the robust controller begins the balance adjustment process. The angle threshold may be preset.

[0445] In other embodiments, if the moving wheel does not receive a control signal from the controller and the motor encoder detects that the moving wheel has moved, the robust controller begins to execute the balance control method starting from step 1110.

[0446] This method intermittently uses a robust controller to control the balance of the wheeled robot, which helps save energy consumption and improves the robot's resistance to external interference without significantly affecting the balance control effect.

[0447] The following example illustrates the process of the balance control method:

[0448] In some embodiments, if the IMU sensor detects that the deflection angle of the first link changes more than the angle threshold, it indicates that the wheeled robot is experiencing a large change in sway under external force disturbance, which affects the balance of the wheeled robot. In this case, the robust controller starts the balance control process.

[0449] Step A10: The robust controller acquires the first time step t. i The state variables include: the deflection angles q1, q2, ..., q of the n links. n angular velocity of the moving wheel angular velocities of n links

[0450] Step A20: The robust controller substitutes the state variables at the first moment into the dynamic equation to determine the inertia matrix, deflection force matrix, and gravity matrix at the first moment.

[0451] This step mainly uses the following formula:

[0452]

[0453] The robust controller uses selection matrices to process the product between the inverse of the inertia matrix and the bias force matrix, and the product between the inverse of the inertia matrix and the gravity matrix, respectively, to obtain the offset parameter matrix.

[0454] The main formulas for calculating the offset parameter matrix are:

[0455]

[0456] The robust controller uses the selection matrix to process the inverse of the inertia matrix to obtain the proportional parameter matrix.

[0457] The main formulas for calculating the scaling parameter matrix are:

[0458]

[0459] Step A30: The robust controller determines the sliding parameters required for each of the n sliding surfaces from the set of predicted parameters.

[0460] Step A40: The robust controller creates n sliding surfaces respectively:

[0461] Step A50: Calculate the rotational torques of the n rotating joints based on the sliding surface and dynamic model parameters. The main formulas used in calculating the rotational torques of the n joints are Formulas 3 and 4 mentioned above; specifically, Formula 3 is as follows: Formula 4:

[0462] Optionally, after determining the rotational torque of the n rotary joints, the robust controller can first determine whether to continue the balance adjustment process. For example, the robust controller determines whether the rotational torque of the n rotary joints is greater than a first threshold. If the rotational torque of the n rotary joints is greater than the first threshold, the robust controller continues to execute step A60; if the rotational torque of the n rotary joints is less than the first threshold, the robust controller does not execute step A60, and the balance control ends.

[0463] Step A60: At the second time t i+1The wheel-legged robot controls the rotation motors of n rotation joints based on the rotational torque of n rotation joints, thereby adjusting the angle between the torso link and the leg link.

[0464] Repeat steps A10-A60 until the balance control process is completed.

[0465] It should be noted that all formulas used in this example are the same as those in the above embodiments. Please refer to the above embodiments for explanations of the parameters in these formulas, which will not be repeated here.

[0466] By abstracting the wheeled robot into an nth-order inverted pendulum model, the joint motors of multiple rotating joints can be adjusted during the balance adjustment process. This allows for a richer range of postures of the multiple links during the balance adjustment process, which helps to reach a balance state more quickly and improves the robustness of the balance control method.

[0467] The following example illustrates the process of the balance control method:

[0468] In the balance control process, step B10: The robust controller acquires the first moment t. i The state variables include: the deflection angles q1, q2, ..., q of the n links. n angular velocity of the moving wheel angular velocities of n links

[0469] Step B20: The robust controller substitutes the state variables at the first moment into the dynamic equations to determine the inertia matrix, eccentric force matrix, and gravity matrix at the first moment. The robust controller uses selection matrices to process the product between the inverse of the inertia matrix and the eccentric force matrix, and the product between the inverse of the inertia matrix and the gravity matrix, respectively, to obtain the offset parameter matrix. For details of this process, please refer to the above embodiment; further elaboration is not provided here.

[0470] Step B30: The robust controller determines the sliding parameters required for each of the n sliding surfaces from the set of predicted parameters.

[0471] Step B40: The robust controller creates n sliding surfaces based on the sliding mode parameters and state variables:

[0472] Step B50: Calculate the rotational torques of the n rotary joints based on the sliding surface and dynamic model parameters. Optionally, after determining the rotational torques of the n rotary joints, the robust controller checks whether the rotational torques of the n rotary joints are greater than a first threshold. If the rotational torques of the n rotary joints are greater than the first threshold, the robust controller continues to execute step B50; if the rotational torques of the n rotary joints are all less than the first threshold, the robust controller does not execute step B50, and the balance control ends.

[0473] Step B60: The robust controller sends the rotational torques of the n rotational joints to the observation module. The observation module determines the task acceleration based on the rotational torques of the n rotational joints. The observation module sends the task acceleration to the whole-body dynamics controller. The whole-body dynamics controller determines the force and torque commands for the whole-body joints based on the task acceleration. The whole-body dynamics controller sends the corresponding force and torque commands to each joint motor.

[0474] Step B70: At the second time t i+1 The wheeled robot controls the movement of the motors corresponding to each joint of its body according to force and torque commands, thereby adjusting the robot's posture.

[0475] Repeat steps B10-B70 until the balance control process is completed.

[0476] It should be noted that all formulas used in this example are the same as those in the above embodiments. Please refer to the above embodiments for explanations of the parameters in these formulas, which will not be repeated here.

[0477] By abstracting the wheel-legged robot into an n-order inverted pendulum model, calculating the rotational torque of the n joints, and determining the task acceleration of the wheel-legged robot based on the rotational torque of the n joints, the input of the whole-body dynamics controller during the balance control process is obtained. This allows the whole-body dynamics controller to calculate the force and torque commands of the joints throughout the body, enabling each mechanism in the wheel-legged robot to participate in the balance control process. This further enriches the posture of the wheel-legged robot during the balance control process and helps to improve the robustness of the balance control method.

[0478] The following section uses an example of n=2 to introduce and explain the balance control method.

[0479] Depend on Figure 9It can be seen that when n equals 2, the nth-order inverted pendulum model includes a moving wheel, two links, and two rotational joints. Optionally, the two links refer to the leg mechanism and torso mechanism of the wheel-legged robot. If the wheel-legged robot has multiple moving wheels, the first moving wheel that contacts the contact surface among the multiple moving wheels coincides in the y-axis direction of the world coordinate system, and the leg mechanism connected to at least one of the first moving wheels coincides in the y-axis direction. The leg mechanism is connected to the moving wheel through a first rotational joint, and the torso mechanism is connected to the leg mechanism through a second rotational joint.

[0480] The balance control method includes the following steps:

[0481] Step 1310: Obtain the state variables of the wheeled-legged robot at the first moment. The state variables at the first moment are used to characterize the motion state of the wheeled-legged robot at the first moment.

[0482] Optionally, the state variables at the first moment include at least one of the following: the deflection angle α of the leg mechanism, the deflection angle β of the torso mechanism, and the angular velocity of the moving wheel. angular velocity of the leg mechanism and the angular velocity of the torso mechanism State variables can be represented by the symbol ξ, and the state variable ξ can be expressed as:

[0483]

[0484] The state variables include various physical quantities observed in the world coordinate system; among them, the deflection angle α of the leg mechanism refers to the deflection angle of the leg mechanism relative to the z-axis in the world coordinate system, the deflection angle β of the torso mechanism refers to the deflection angle of the torso mechanism relative to the z-axis in the world coordinate system, and the angular velocity of the moving wheel... This refers to: the rotational speed of the moving wheel along the x-axis in the world coordinate system in the counterclockwise (or clockwise) direction, and the angular velocity of the leg mechanism. Used to characterize the rate of change of deflection angle of the leg mechanism and the angular velocity of the torso mechanism. Used to characterize the rate of change of deflection angle of the leg-torso mechanism.

[0485] It should be noted that the physical quantities in the state variables can also be determined by the relative positions between the various mechanisms of the wheeled-legged robot. For example, the deflection angle α of the leg mechanism refers to the deflection angle of the leg mechanism relative to the z-axis in the world coordinate system, and the deflection angle β′ of the torso mechanism refers to the deflection angle of the torso mechanism relative to the leg mechanism, that is, β=α+β′ (β′ and β have the same positive direction). The coordinate system used for observing the various physical quantities in the state variables is determined according to actual needs, and this application does not impose any restrictions here.

[0486] In some embodiments, the computer device acquires the state variables of the wheeled robot at a first moment, including: determining the deflection angle α of the leg mechanism and the deflection angle β of the torso mechanism through IMU sensors and motor encoders; and determining the angular velocity of the moving wheels through the motor encoders. angular velocity of the leg mechanism and the angular velocity of the torso mechanism

[0487] Step 1320: Determine the dynamic model parameters based on the dynamic equations of the wheeled robot and the state variables at the first moment.

[0488] Step 1320 includes the following sub-steps: Sub-step 1323: The robust controller substitutes the state variables at the first moment into the dynamic equation to determine the inertia matrix, the bias force matrix, and the gravity matrix at the first moment. The inertia matrix is ​​used to characterize the mass and moment of inertia of each joint rigid body that makes up the wheel-legged robot at the first moment. The bias force matrix is ​​used to characterize the bias force of the wheel-legged robot at the first moment. The gravity matrix is ​​used to characterize the gravity of the wheel-legged robot at the first moment.

[0489] Sub-step 1326: Determine the dynamic model parameters based on the inertia matrix, the deflection force matrix, and the gravity matrix.

[0490] When n equals 2, after abstracting the wheel-legged robot into a second-order inverted pendulum model, the dynamic equation used in the balance control process, derived based on the Euler-Lagrange equations, can be expressed as:

[0491]

[0492] Optionally, the dynamic model parameters are determined based on the second-order inverted pendulum model used for balance control; the dynamic model parameters include: a proportional parameter matrix and an offset parameter matrix.

[0493] In the design process of robust controllers, the dynamic equations are analyzed as follows:

[0494]

[0495] After performing partial linearization, we obtain formula a:

[0496]

[0497] Among them, M -1 (α,β) is the inverse matrix of the inertia matrix, M -1 (α,β)*M(α,β)=E, where E is the identity matrix. For the physical meaning of other parameters in the equation, please refer to the above example, which will not be repeated here.

[0498] Formula a is processed to unify the rotational torque τ1 of the first rotary joint and the rotational torque τ2 of the second rotary joint in formula a. Formula b is obtained by processing formula a using a selection matrix:

[0499]

[0500] in, That is, the offset parameter matrix f[]. That is, the proportionality parameter matrix g[], S T Let S be the transpose of matrix S. T The selection matrix is ​​a 3x2 matrix, for example, S T The matrix is ​​specifically as follows

[0501] Formula b can be pre-designed after the wheel-legged robot is abstracted into a second-order inverted pendulum model. During the balance control process, after the robust controller obtains the state variables at the first moment, it substitutes the state variables at the first moment into the dynamic equation to determine the inertia matrix, bias force matrix and gravity matrix at the first moment; based on the inertia matrix, bias force matrix and gravity matrix, the dynamic model parameters are determined.

[0502] Optionally, sub-step 1326 can be implemented through the following steps: The robust controller uses a selection matrix to process the product between the inverse of the inertia matrix and the bias force matrix, and the product between the inverse of the inertia matrix and the gravity matrix, respectively, to obtain an offset parameter matrix. The selection matrix is ​​used to extract the rotational torque of the first and second rotating joints from the dynamic equations. The robust controller uses the selection matrix to process the inverse of the inertia matrix to obtain a proportional parameter matrix. For details of this process, please refer to the above embodiment, which will not be repeated here.

[0503] In other words, during the balance control process, the robust controller can operate according to formula b and... Figure 9 The proportional parameter matrix and dynamic model parameter matrix are calculated from the equations in section (5) of the corresponding instruction manual.

[0504] Step 1330: Based on the state variables at the first moment, establish a sliding surface. The state variables of the wheeled robot gradually approach 0 on the sliding surface.

[0505] When n equals 2, it is necessary to determine the rotational torque of the first and second rotary joints, requiring a robust controller to establish two sliding surfaces. These two sliding surfaces include: 1. The sliding surface of the first rotary joint, i.e., the state variables related to the first rotary joint gradually approach 0 on this sliding surface; 2. The sliding surface of the second rotary joint, i.e., the state variables related to the second rotary joint gradually approach 0 on this sliding surface.

[0506] Optionally, the first sliding surface and the second sliding surface are different. The purpose of setting the first sliding surface and the second sliding surface is to calculate the rotational torque of the first rotary joint and the rotational torque of the second rotary joint. The first sliding surface and the second sliding surface jointly participate in the calculation process of the rotational torque of the first rotary joint and the second rotary joint. The definition of the first sliding surface is not limited to the sliding surface corresponding to the first rotary joint. Similarly, the definition of the second sliding surface is not limited to the sliding surface corresponding to the second rotary joint. The first sliding surface and the second sliding surface can be defined according to actual needs, and this application does not limit them here.

[0507] As introduced in the section on the nth-order inverted pendulum model above, to ensure the calculation of a definite rotational torque, establishing the sliding surface requires at least three state parameters and at least two sliding parameters in the state variables; where the state variables are written as... Then α, β, All of these are state parameters.

[0508] In some embodiments, step 1330 includes sub-step 1333: the robust controller determines at least four sliding mode parameters, which are used to constrain the rotational torque of the first rotational joint and the rotational torque of the second rotational joint to meet the system stability conditions of the wheeled robot.

[0509] Optionally, different state parameters and different sliding parameters are used to establish different sliding surfaces. For example, establishing a first sliding surface requires at least the deflection angle of the leg mechanism, the angular velocity of the moving wheel, and the angular velocity of the leg mechanism in the state variables; establishing a second sliding surface requires at least the deflection angle of the torso mechanism, the angular velocity of the moving wheel, and the angular velocity of the torso mechanism in the state variables at the first moment. In this case, the first sliding surface includes at least two sliding parameters: the sliding parameter corresponding to the deflection angle of the leg mechanism and the sliding parameter corresponding to the angular velocity of the moving wheel. The second sliding surface includes at least two sliding parameters: the sliding parameter corresponding to the deflection angle of the torso mechanism and the sliding parameter corresponding to the angular velocity of the moving wheel.

[0510] In some embodiments, step 1330 further includes sub-step 1336, in which the robust controller establishes a first sliding surface and a second sliding surface based on the state variables at the first moment and at least four sliding parameters.

[0511] Optionally, the robust controller establishes a first sliding surface based on the first sliding surface parameters, the second sliding surface parameters, the deflection angle of the leg mechanism, the angular velocity of the moving wheel, and the angular velocity of the leg mechanism; and establishes a second sliding surface based on the third sliding surface parameters, the fourth sliding surface parameters, the deflection angle of the torso mechanism, the angular velocity of the torso mechanism, and the angular acceleration of the leg mechanism.

[0512] For example, the robust controller processes the deflection angle of the leg mechanism using a first sliding mode parameter to obtain the processing result of the leg mechanism; it processes the angular velocity of the moving wheel using a second sliding mode parameter to obtain the processing result of the moving wheel. The robust controller determines the first sliding surface based on the processing result of the leg mechanism, the processing result of the moving wheel, and the angular velocity of the leg mechanism. Optionally, the first sliding surface is proportional to the processing result of the leg mechanism, the processing result of the moving wheel, and the angular velocity of the leg mechanism.

[0513] For example, the robust controller adds the processing results of the leg mechanism, the processing results of the moving wheel, and the angular velocity of the leg mechanism to obtain the first sliding surface.

[0514] For example, the robust controller processes the deflection angle of the torso mechanism using a third sliding mode parameter to obtain the processing result of the torso mechanism; it processes the angular velocity of the moving wheel using a fourth sliding mode parameter to obtain the processing result of the moving wheel; the robust controller determines the second sliding surface based on the processing result of the torso mechanism, the processing result of the moving wheel, and the angular velocity of the torso mechanism. Optionally, the second sliding surface is proportional to the processing result of the torso mechanism, proportional to the processing result of the moving wheel, and proportional to the angular velocity of the leg mechanism.

[0515] For example, the robust controller adds the processing results of the torso mechanism, the processing results of the moving wheel, and the angular velocity of the torso mechanism to obtain the second sliding surface.

[0516] In some embodiments, during the balance control process, the robust controller establishes the first sliding surface s1 and the second sliding surface s2 according to the following two formulas:

[0517] s1=λ1ξ1+λ2ξ3+ξ4

[0518] s2=λ3ξ2+λ4ξ3+ξ5

[0519] Where λ1 represents the first sliding mode parameter, λ2 represents the second sliding mode parameter, λ3 represents the third sliding mode parameter, λ4 represents the fourth sliding mode parameter, and ξ1 represents the state variable. The first element α, ξ2 in the equation represents the state variables. The second element β, ξ3 in the equation represents the state variables. The third element ξ4 represents the state variable. The 4th element ξ5 represents the state variable. The 5th element

[0520] By establishing the first and second sliding surfaces in this way, and controlling the number of sliding parameters in the first and second sliding surfaces, the rotational torques of the first and second rotary joints that satisfy the system stability principle can be determined with relatively low computational overhead. This helps to improve the speed of determining the rotational torques and achieve balance control of the robot.

[0521] Step 1340: Calculate the rotational torque of the first rotary joint and the rotational torque of the second rotary joint based on the sliding surface and dynamic model parameters.

[0522] After determining the first and second sliding surfaces, the robust controller calculates the first derivative formula of the first sliding surface s1 with respect to time and the first derivative formula of the second sliding surface s2 with respect to time:

[0523]

[0524]

[0525] in, This represents the first derivative of the first sliding surface s1 with respect to time. f represents the first derivative of the second sliding surface s2 with respect to time. k (k∈(1,2,3)) are elements in the offset parameter matrix f[] in Formula 2 above, g mn (m∈(1,2,3),n∈(1,2,3)) are elements in the proportional parameter matrix g[] in Formula 2 above.

[0526] Rearranging the formulas for the first derivative of the first sliding surface s1 with respect to time and the first derivative of the second sliding surface s2 with respect to time into matrix form, we obtain formula c:

[0527]

[0528] Rearranging formula c yields formula d:

[0529]

[0530] in, g s -1 For g s The inverse matrix, sgn() is the sign function,

[0531] Formula d can be further written as formula e:

[0532]

[0533] That is, in the nth order inverted pendulum model when n=2, Formula 4 above becomes Formula d, Formula 5.

[0534] This is achieved using formula e. In the balance control method based on the second-order inverted pendulum model, the calculation of the rotational torque of the first and second rotational joints requires the use of formulas d and e.

[0535] Optionally, the robust controller needs to consider system stability conditions during the calculation of torque. For the method of satisfying system stability conditions in the balance control process of the second-order inverted pendulum model, please refer to the following embodiments.

[0536] Step 1350: At the second moment, adjust the angle between the leg mechanism and the moving wheel according to the rotational torque of the first rotating joint, and adjust the angle between the torso mechanism and the leg mechanism according to the rotational torque of the second rotating joint.

[0537] In one implementation, the robust controller sends the rotational torque of the first rotary joint and the rotational torque of the second rotary joint to the joint motors of the first and second rotary joints. That is, step 1350 is implemented as follows: the robust controller adjusts the angle between the leg mechanism and the moving wheel according to the rotational torque of the first rotary joint via the rotational motor of the first rotary joint; the robust controller adjusts the angle between the torso mechanism and the leg mechanism according to the rotational torque of the second rotary joint via the rotational motor of the second rotary joint.

[0538] In another implementation, after the robust controller determines the rotational torques of the first and second rotary joints, it sends these torques to the observation module. The observation module calculates the angular acceleration at a first moment using the torques of the first and second rotary joints and the dynamic equations. Based on this angular acceleration, it calculates the task acceleration of the wheel-legged robot. The observation module then sends the task acceleration to the whole-body dynamics controller. Based on the task acceleration, the whole-body dynamics controller determines the force and torque commands for all joints of the wheel-legged robot. Based on the force and torque commands for the first rotary joint, the whole-body dynamics controller controls the joint motors of the first rotary joint to adjust the angle between the leg mechanism and the moving wheel. Based on the force and torque commands for the second rotary joint, the whole-body dynamics controller controls the joint motors of the second rotary joint to adjust the angle between the torso mechanism and the leg mechanism.

[0539] Optionally, the whole-body dynamics controller can also control the movement of other joints in the wheeled-legged robot, except for the first and second rotary joints, according to the force and torque commands of the whole-body joints, so that more active joints can participate in the balance control process, enriching the posture of the wheeled-legged robot in the balance control process.

[0540] For a detailed introduction to this topic, please refer to the relevant content in the nth-order inverted pendulum model above; it will not be repeated here.

[0541] In summary, by abstracting the wheeled-legged robot into a second-order inverted pendulum model, the angle between the moving wheels and the leg mechanism can be adjusted during balance control, as well as the angle between the torso mechanism and the leg mechanism can be adaptively adjusted. This enriches the robot's posture during balance adjustment, making the robot's posture changes more flexible, helping to quickly adjust the robot to a balanced state, improving the robot's ability to recover to a balanced state under different disturbances, and enhancing the robustness of the robot's balance control process.

[0542] The following example illustrates the process of satisfying the system stability criterion during the calculation of the rotational torque of the first and second rotating joints.

[0543] From the definition and properties of sliding surfaces, it can be seen that the state parameters on the sliding surface satisfy the following:

[0544] s1=λ1ξ1+λ2ξ3+ξ4=0

[0545] s²=λ³ξ²+λ⁴ξ³+ξ⁵=0

[0546] The above two formulas can be used to express ξ4 and ξ5 in reverse:

[0547] ξ4=-λ1ξ1-λ2ξ3

[0548] ξ5=-λ3ξ2-λ4ξ3

[0549] Then the first derivative of ξ4 with respect to time The first derivative of ξ5 with respect to time It can be represented as:

[0550]

[0551]

[0552] Since ξ4 represents ξ1 represents α, ξ5 represents ξ2 represents β; therefore, ξ1 and ξ2 can be expressed by the following formula:

[0553]

[0554]

[0555]

[0556] Let's review the dynamic equations of the second-order inverted pendulum model in polynomial form mentioned above:

[0557]

[0558]

[0559]

[0560] Adding the three formulas included in the polynomial form of the dynamic equations yields formula f:

[0561]

[0562] In formula f By substitution, we obtain formula g:

[0563]

[0564] Simplifying formula g, we obtain formula h:

[0565]

[0566] The formula h is expressed in reverse. get:

[0567]

[0568] For each including Solving the three formulas simultaneously yields system of equations 1:

[0569]

[0570] Due to space limitations, equation system 1... The representation on the right side of the equation is somewhat distorted. The brackets "[]" represent algebraic expressions and do not indicate matrices or any other meaning. Please refer to the brackets above equation system 1. To understand the system of equations 1 using the following representation method The way to represent it.

[0571] Equation system 1 can be written as state equations in terms of the state variables (ξ1, ξ2, ξ3). Optionally, state equations refer to equations related to the state quantities.

[0572] For ease of reading, let's use a1 to represent... m1 represents m2 represents m3 indicates The state equation can be written as:

[0573]

[0574] Optionally, the system stability criterion refers to the fact that the eigenvalues ​​of the coefficient matrix in the state equation determined based on the sliding surface and the dynamic equation are negative, that is, the eigenvalues ​​are distributed in the left half of the phase plane.

[0575] To ensure the system meets the stability criterion, the coefficient matrix must satisfy the following conditions. The eigenvalues ​​are all distributed in the left half of the complex plane. This state equation can be used as a matrix inequality constraint, meaning that the values ​​of the sliding mode parameters need to ensure that the eigenvalues ​​of the coefficient matrix are negative.

[0576] Optionally, in the balance control process of this application, the robust controller determines the rotational torque of the first rotary joint and the rotational torque of the second rotary joint according to formula d and equation set 1.

[0577] During balance control, the robust controller calculates the rotational torque τ1 of the first rotary joint and the rotational torque τ2 of the second rotary joint according to formula d. Formula d involves the sliding surface s1, the second sliding surface s2, the dynamic model parameters in formula 2, and the sliding parameters. The robust controller obtains the state variables at the first moment from the sensor and substitutes them into the corresponding formula to determine the sliding surface s1, the second sliding surface s2, and the dynamic model parameters. It then identifies the sliding parameters that satisfy the system stability conditions and calculates the rotational torque τ1 of the first rotary joint and the rotational torque τ2 of the second rotary joint using formula d, while ensuring that the sliding parameters satisfy the matrix inequality constraints formed by the state equations above.

[0578] In the balance control method provided in this application, four sliding mode parameters that satisfy matrix inequality constraints can be determined first, and then the first sliding surface and the second sliding surface can be constructed through the sliding mode parameters.

[0579] Alternatively, four sliding parameters can be randomly selected within a certain numerical range. These four sliding parameters can be used to construct the first and second sliding surfaces. The rotational torque τ1 of the first rotary joint and the rotational torque τ2 of the second rotary joint can be calculated using formula d. Then, it can be verified whether the four sliding parameters satisfy the matrix inequality constraints mentioned above.

[0580] Optionally, if the four sliding mode parameters satisfy the above matrix inequality constraints, the determined rotational torque τ1 of the first rotating joint and the rotational torque τ2 of the second rotating joint can be used; if the four sliding mode parameters do not satisfy the above matrix inequality constraints, the determined rotational torque τ1 of the first rotating joint and the rotational torque τ2 of the second rotating joint cannot be used.

[0581] The following examples illustrate the method for determining sliding mode parameters.

[0582] In some embodiments, determining at least four sliding mode parameters includes: a computer device determining a first sliding mode parameter from a first set of prediction parameters, a second sliding mode parameter from a second set of prediction parameters, a third sliding mode parameter from a third set of prediction parameters, and a fourth sliding mode parameter from a fourth set of prediction parameters; wherein the sliding mode parameters included in the first set of prediction parameters, the second set of prediction parameters, the third set of prediction parameters, and the fourth set of prediction parameters respectively satisfy the constraints of the stability criterion.

[0583] In this embodiment, after determining the state at the first moment, the robust controller first uses the above state equation as a matrix inequality constraint to determine the solution sets corresponding to λ1, λ2, λ3, and λ4, namely the first prediction parameter set, the second prediction parameter set, the third prediction parameter set, and the fourth prediction parameter set. Then, it selects the first sliding mode parameter, the second sliding mode parameter, the third sliding mode parameter, and the fourth sliding mode parameter from each parameter set. Through the first sliding mode parameter, the second sliding mode parameter, the third sliding mode parameter, and the fourth sliding mode parameter and the state quantity at the first moment, it establishes the first sliding surface and the second sliding surface. Then, according to the formula d above, it calculates the rotational torque of the first rotary joint and the rotational torque of the second rotary joint.

[0584] In some implementations, the process of determining any one of the four prediction parameter sets in the above prediction solution sets is as follows: calculate the coefficient matrix. The eigenvalues, and all the eigenvalues ​​corresponding to the coefficient matrix, are used as the prediction parameter set.

[0585] This method avoids selecting sliding mode parameters that do not meet system stability requirements, which would prevent the torques of the first and second rotating joints calculated using these parameters from being unable to control the balance of the wheeled robot. It also avoids unnecessary calculations by the computer equipment and helps to shorten the time required to determine the torques of the first and second rotating joints.

[0586] The robustness of the balance control effect of the wheeled robot under the balance control method provided in this embodiment will be experimentally verified.

[0587] Figure 14This is a simulation diagram of a dual-wheel balance control method provided in an exemplary embodiment of this application.

[0588] Figure 14 The background is as follows: the wheel-legged robot is in a two-wheeled mode and is subjected to lateral disturbance forces. Based on abstracting the wheel-legged robot as an nth-order inverted pendulum, a robust controller acquires the robot's state variables. The rotational torque of the nth rotational joint is calculated using the methods described in the above embodiments. The posture of the wheel-legged robot's leg mechanism and torso mechanism are then adjusted based on the rotational torque of the nth rotational joint. Figure 14 As shown, under lateral disturbance, the wheel-legged robot simultaneously adjusts the posture of n links of the wheel-legged robot, so that the wheel-legged robot swings back and forth, the moving wheels connected to the two outer leg mechanisms maintain balance on the contact surface, and the torso mechanism also moves in a timely manner to adjust the center of gravity and force of the wheel-legged robot, helping the wheel-legged robot to quickly restore its overall balance.

[0589] Figure 15 This is a simulation diagram of a dual-wheel balance control method provided in another exemplary embodiment of this application. Figure 15 The background is as follows: The wheel-legged robot is in a two-wheeled motion mode. In this example, n equals 2, meaning the balance control of the wheel-legged robot is based on a second-order inverted pendulum model. When the wheel-legged robot falls from a height, there is an initial angle between the leg mechanism and the z-axis upon contact with the contact surface. As the wheel-legged robot begins its movement from an initial position relatively far from its equilibrium point and falls to the ground, the two outer leg mechanisms act like springs. Due to the large impact, the first rotary joint controls the two outer leg mechanisms to sway back and forth significantly to maintain the robot's balance. The (n-1) links above the leg mechanism (i.e., the first mechanism) also move significantly in a timely manner to adjust the center of gravity and forces, helping the robot quickly regain its balance.

[0590] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0591] Please refer to Figure 16This diagram illustrates a block diagram of a balance control device for a wheeled-legged robot according to an embodiment of this application. The device has the function of implementing the balance control method for the wheeled-legged robot described above. This function can be implemented in hardware or by hardware executing corresponding software. The device can be the computer device described above, or it can be installed within a computer device. The wheeled-legged robot includes: a movable wheel, n links, and n rotating joints. The movable wheel and a first link among the n links are connected via a first rotating joint among the n rotating joints. The n links are connected in series via n-1 rotating joints other than the first rotating joint, where n is a positive integer greater than or equal to 3.

[0592] like Figure 16 As shown, the device 1600 may include: a status acquisition module 1610, a parameter determination module 1620, a sliding surface establishment module 1630, a torque calculation module 1640, and a joint rotation module 1650.

[0593] The state acquisition module 1610 is used to acquire the state quantity of the wheeled-legged robot at a first moment, and the state quantity at the first moment is used to characterize the motion state of the wheeled-legged robot at the first moment.

[0594] The parameter determination module 1620 is used to determine the dynamic model parameters based on the dynamic equation of the wheel-legged robot and the state variables at the first moment. The dynamic model parameters are used to define the mapping relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The angular acceleration at the first moment includes the angular acceleration of the n links and the angular acceleration of the moving wheels. The rotational torque at the second moment includes the rotational torque of the n rotating joints.

[0595] The sliding surface establishment module 1630 is used to establish a sliding surface based on the state variables at the first moment, wherein the state variables of the wheeled robot gradually approach 0 on the sliding surface.

[0596] The torque calculation module 1640 is used to calculate the rotational torque of the n rotating joints based on the sliding surface and the dynamic model parameters.

[0597] The joint rotation module 1650 is used to control the n rotating joints according to the rotational torque of the n rotating joints at the second moment.

[0598] In some embodiments, the parameter determination module 1620 includes: a matrix calculation unit, configured to substitute the state variables at the first moment into the dynamic equation to determine the inertia matrix, the bias force matrix, and the gravity matrix at the first moment, wherein the inertia matrix is ​​used to characterize the mass and moment of inertia of the n rotating joints at the first moment, the bias force matrix is ​​used to characterize the bias force of the wheel-legged robot at the first moment, and the gravity matrix is ​​used to characterize the gravity of the wheel-legged robot at the first moment; and a parameter determination unit, configured to determine the dynamic model parameters based on the inertia matrix, the bias force matrix, and the gravity matrix.

[0599] In some embodiments, the dynamic model parameters include: a scaling parameter matrix and an offset parameter matrix. The scaling parameter matrix characterizes the proportional relationship between the angular acceleration at the first moment and the rotational torque at the second moment, and the offset parameter matrix characterizes the offset relationship between the angular acceleration at the first moment and the rotational torque at the second moment. A parameter determination unit is used to process the product between the inverse of the inertia matrix and the bias force matrix, and the product between the inverse of the inertia matrix and the gravity matrix, respectively, using a selection matrix to obtain the offset parameter matrix. The selection matrix is ​​used to extract the rotational torque of the n rotating joints from the dynamic equation. The scaling parameter matrix is ​​obtained by processing the inverse of the inertia matrix using the selection matrix.

[0600] In some embodiments, the dynamic equations of the robot are derived by abstracting the wheel-legged robot into an nth-order inverted pendulum model based on the Euler-Lagrange equations.

[0601] In some embodiments, the sliding surface includes n sliding surfaces, which are used to constrain the rotational torque of the n rotary joints; the sliding surface establishment module 1630 includes: a sliding parameter determination unit, used to determine at least two sliding parameters for the i-th sliding surface among the n sliding surfaces, where i is a positive integer less than or equal to n; and a sliding surface establishment unit, used to establish the i-th sliding surface based on the at least two sliding parameters and the state quantity at the first moment.

[0602] In some embodiments, the sliding mode parameter determination unit is configured to determine a first sliding mode parameter from the at least one sliding mode parameter set from the (2i-1)th prediction parameter set, and to determine a second sliding mode parameter from the at least one sliding mode parameter set from the (2i-1)th prediction parameter set; wherein the sliding mode parameters included in the (2i-1)th prediction parameter set and the (2i)th prediction parameter set respectively satisfy the constraint conditions of the stability criterion.

[0603] In some embodiments, the state quantities at the first moment include: the deflection angles of the n links, the angular velocities of the n links, and the angular velocity of the moving wheel; the sliding surface establishment unit is used to process the deflection angle of the i-th link according to the first sliding parameter among the at least two sliding parameters to obtain the processing result of the i-th link; process the angular velocity of the moving wheel according to the second sliding parameter among the at least two sliding parameters to obtain the processing result of the moving wheel; and establish the i-th sliding surface according to the processing result of the i-th link, the processing result of the moving wheel, and the angular velocity of the i-th link.

[0604] In some embodiments, the joint rotation module 1650 is used to control the rotation motor corresponding to any one of the n rotation joints to rotate according to the rotational torque of the rotation joint.

[0605] In some embodiments, the joint rotation module 1650 is configured to calculate the task acceleration of the wheel-legged robot at the second moment based on the rotational torque of the n rotational joints, wherein the task acceleration refers to the acceleration related to the center of mass of the wheel-legged robot; determine the force and torque commands of the whole-body joints of the wheel-legged robot at the second moment based on the task acceleration, wherein the whole-body joints include the n rotational joints; and control the whole-body joints based on the force and torque commands at the second moment.

[0606] In some embodiments, the joint rotation module 1650 is configured to: determine the angular acceleration at the second moment based on the dynamic equation and the rotational torque of the n rotating joints; determine the expected incremental position and expected incremental velocity of the center of mass of the wheel-legged robot at the second moment based on the state variables at the first moment and the angular acceleration at the second moment, wherein the expected incremental position is used to characterize the distance between the projection of the center of mass of the wheel-legged robot on the contact surface and the virtual contact point in a first direction, and the expected incremental velocity is used to characterize the rate of change of the distance in the first direction, wherein the virtual contact point refers to the center of each contact point between the wheel-legged robot and the contact surface; and determine the task acceleration of the wheel-legged robot at the second moment based on the expected incremental position and expected incremental velocity.

[0607] In some embodiments, the state quantities include: the deflection angles of the n links, the angular velocities of the n links, and the angular velocities of the moving wheels; the state acquisition module 1610 is used to determine the deflection angles of the n links through the inertial measurement unit and the motor encoder of the wheel-legged robot; and to determine the angular velocities of the n links and the angular velocities of the moving wheels through the motor encoder.

[0608] In some embodiments, the moving wheel refers to a virtual wheel formed by superimposing at least two real wheels of the wheel-legged robot that do not coincide in a first direction, and the first link refers to a virtual first link formed by superimposing at least two real first links of the wheel-legged robot that do not coincide in the first direction, where the first direction refers to the forward direction of the wheel-legged robot.

[0609] In some embodiments, the state quantities include: the deflection angles of the n links, the angular velocities of the n links, and the angular velocity of the moving wheel; the state acquisition module 1610 is used to determine, through the inertial measurement unit and motor encoder of the wheel-legged robot, the deflection angles of the other links among the n links excluding the first link, and the deflection angles of each of the actual first links; to determine, through the motor encoder, the angular velocities of the other links among the n links excluding the first link, the angular velocities of each of the actual first links, and the angular velocities of each of the actual wheels; to determine the angular velocity of the moving wheel based on the angular velocities of each of the actual wheels; and to determine the deflection angle and angular velocity of the first link based on the length, deflection angle, and angular velocity of each of the actual first links.

[0610] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0611] Please refer to Figure 16 This diagram illustrates a structural block diagram of a computer device 1600 provided in one embodiment of this application. The computer device 1600 can be any electronic device capable of data calculation, processing, and storage. The computer device 1600 can be used to implement the balance control method for the wheeled robot provided in the above embodiments.

[0612] Typically, computer device 1600 includes a processor 1601 and a memory 1602.

[0613] Processor 1601 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1601 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). Processor 1601 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1601 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 1601 may also include an AI processor for handling computational operations related to machine learning.

[0614] The memory 1602 may include one or more computer-readable storage media, which may be non-transitory. The memory 1602 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1602 is used to store a computer program configured to be executed by one or more processors to implement the aforementioned balance control method for a wheeled robot.

[0615] Those skilled in the art will understand that Figure 16 The structure shown does not constitute a limitation on the computer device 1600, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0616] In an illustrative embodiment, a computer-readable storage medium is also provided, wherein a computer program is stored in the storage medium, and the computer program, when executed by a processor of a computer device, implements the aforementioned balance control method for a wheeled-legged robot. Optionally, the aforementioned computer-readable storage medium may be ROM (Read-Only Memory), RAM (Random Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic tape, floppy disk, and optical data storage device, etc.

[0617] In an exemplary embodiment, a computer program product is also provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the aforementioned balance control method for a wheeled robot.

[0618] It should be understood that "multiple" as used herein refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, the step numbers described herein are merely illustrative of one possible execution order. In some other embodiments, the steps may not be executed in numerical order, such as two steps with different numbers being executed simultaneously, or two steps with different numbers being executed in the reverse order of the illustration. This application does not limit this.

[0619] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A balance control method for a wheeled-legged robot, characterized in that, The wheel-legged robot includes: a movable wheel, n links, and n rotating joints. The movable wheel and a first link among the n links are connected via a first rotating joint among the n rotating joints. The n links are connected in series via n-1 rotating joints other than the first rotating joint, where n is a positive integer greater than or equal to 2. The method includes: The state variables of the wheeled robot at a first moment are obtained. The state variables at the first moment are used to characterize the motion state of the wheeled robot at the first moment. The state variables include: the deflection angle of the n links, the angular velocity of the n links and the angular velocity of the moving wheel. Based on the dynamic equations of the wheel-legged robot and the state variables at the first moment, dynamic model parameters are determined. These dynamic model parameters define the mapping relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The angular acceleration at the first moment includes the angular acceleration of the n links and the angular acceleration of the moving wheels. The rotational torque at the second moment includes the rotational torque of the n rotating joints. Based on the state variables at the first moment, n sliding surfaces are established. For the i-th sliding surface among the n sliding surfaces, the i-th sliding surface is established based on at least two sliding parameters determined for the i-th sliding surface and the state variables at the first moment, where i is a positive integer less than or equal to n. The n sliding surfaces are used to constrain the rotational torque of the n rotating joints. The sliding parameters are used to characterize the proportional relationship between the state parameters in the state variables and the sliding surfaces. The state variables of the wheel-legged robot gradually approach 0 on the sliding surfaces. The process of establishing the i-th sliding surface includes: processing the deflection angle of the i-th link according to the first sliding parameter among the at least two sliding parameters to obtain the processing result of the i-th link; processing the angular velocity of the moving wheel according to the second sliding parameter among the at least two sliding parameters to obtain the processing result of the moving wheel; and establishing the i-th sliding surface based on the processing result of the i-th link, the processing result of the moving wheel, and the angular velocity of the i-th link. The rotational torques of the n rotating joints are calculated based on the n sliding surfaces and the dynamic model parameters. At the second moment, the n rotary joints are controlled according to their rotational torques.

2. The method according to claim 1, characterized in that, The step of determining the dynamic model parameters based on the dynamic equations of the wheeled-legged robot and the state variables at the first moment includes: Substituting the state variables at the first moment into the dynamic equation, the inertia matrix, the bias force matrix, and the gravity matrix at the first moment are determined. The inertia matrix is ​​used to characterize the mass and moment of inertia of the n rotating joints at the first moment, the bias force matrix is ​​used to characterize the bias force of the wheel-legged robot at the first moment, and the gravity matrix is ​​used to characterize the gravity of the wheel-legged robot at the first moment. The dynamic model parameters are determined based on the inertia matrix, the deflection force matrix, and the gravity matrix.

3. The method according to claim 2, characterized in that, The dynamic model parameters include a scaling parameter matrix and an offset parameter matrix. The scaling parameter matrix is ​​used to characterize the proportional relationship between the angular acceleration at the first moment and the rotational torque at the second moment, and the offset parameter matrix is ​​used to characterize the offset relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The step of determining the dynamic model parameters based on the inertia matrix, the deflection force matrix, and the gravity matrix includes: The selection matrix is ​​used to process the product between the inverse matrix of the inertia matrix and the deflection force matrix, and the product between the inverse matrix of the inertia matrix and the gravity matrix, respectively, to obtain the offset parameter matrix. The selection matrix is ​​used to extract the rotational torque of the n rotating joints from the dynamic equation. The inverse of the inertia matrix is ​​processed using the selection matrix to obtain the scaling parameter matrix.

4. The method according to claim 1, characterized in that, The dynamic equations of the robot are derived by abstracting the wheel-legged robot into an nth-order inverted pendulum model based on the Euler-Lagrange equations.

5. The method according to claim 1, characterized in that, The method further includes: The first sliding mode parameter is determined from the (2i-1)th set of predicted parameters, and the second sliding mode parameter is determined from the (2i-1)th set of predicted parameters. The sliding mode parameters included in the (2i-1)th prediction parameter set and the 2ith prediction parameter set respectively satisfy the constraints of the stability criterion.

6. The method according to claim 1, characterized in that, The step of controlling the n rotary joints based on their rotational torques includes: For any one of the n rotary joints, the corresponding rotary motor is controlled to rotate according to the rotational torque of the rotary joint.

7. The method according to claim 1, characterized in that, The step of controlling the n rotary joints based on their rotational torques includes: Based on the rotational torque of the n rotating joints, the task acceleration of the wheeled robot at the second moment is calculated, and the task acceleration includes the acceleration related to the center of mass of the wheeled robot. Based on the task acceleration, the force and torque commands of the whole joints of the wheel-legged robot at the second moment are determined, and the whole joints include the n rotational joints; The joints throughout the body are controlled according to the force and torque commands at the second moment.

8. The method according to claim 7, characterized in that, The step of calculating the task acceleration of the wheeled robot at the second moment based on the rotational torque of the n rotating joints includes: Based on the dynamic equations and the rotational torques of the n rotating joints, determine the angular acceleration at the second moment; Based on the state quantity at the first moment and the angular acceleration at the second moment, the expected incremental position and expected incremental velocity of the center of mass of the wheel-legged robot at the second moment are determined. The expected incremental position is used to characterize the distance between the projection of the center of mass of the wheel-legged robot on the contact surface and the virtual contact point in the first direction. The expected incremental velocity is used to characterize the rate of change of the distance in the first direction. The virtual contact point refers to the center of each contact point between the wheel-legged robot and the contact surface. The task acceleration of the wheeled robot at the second moment is determined based on the expected incremental position and the expected incremental velocity.

9. The method according to claim 1, characterized in that, The process of obtaining the state variables of the wheeled robot at the first moment includes: The deflection angles of the n links are determined by the inertial measurement unit and motor encoder of the wheel-legged robot. The angular velocities of the n links and the angular velocity of the moving wheel are determined by the motor encoder.

10. A balance control device for a wheeled-legged robot, characterized in that, The wheeled robot includes: a movable wheel, n links, and n rotating joints. The movable wheel and a first link of the n links are connected via a first rotating joint of the n rotating joints. The n links are connected in series via n-1 rotating joints other than the first rotating joint, where n is a positive integer greater than or equal to 2. The device includes: The state acquisition module is used to acquire the state quantities of the wheeled robot at a first moment. The state quantities at the first moment are used to characterize the motion state of the wheeled robot at the first moment. The state quantities include: the deflection angle of the n links, the angular velocity of the n links and the angular velocity of the moving wheel. The parameter determination module is used to determine the dynamic model parameters based on the dynamic equations of the wheel-legged robot and the state variables at the first moment. The dynamic model parameters are used to define the mapping relationship between the angular acceleration at the first moment and the rotational torque at the second moment. The angular acceleration at the first moment includes the angular acceleration of the n links and the angular acceleration of the moving wheels. The rotational torque at the second moment includes the rotational torque of the n rotating joints. A sliding surface establishment module is used to establish n sliding surfaces based on the state variables at the first moment. For the i-th sliding surface among the n sliding surfaces, the i-th sliding surface is established based on at least two sliding parameters determined for the i-th sliding surface and the state variables at the first moment, where i is a positive integer less than or equal to n. The n sliding surfaces are used to constrain the rotational torque of the n rotary joints. The sliding parameters are used to characterize the proportional relationship between the state parameters in the state variables and the sliding surfaces. The state variables of the wheel-legged robot are in... The sliding surface gradually approaches 0; wherein, the process of establishing the i-th sliding surface includes: processing the deflection angle of the i-th link according to the first sliding parameter among the at least two sliding parameters to obtain the processing result of the i-th link; processing the angular velocity of the moving wheel according to the second sliding parameter among the at least two sliding parameters to obtain the processing result of the moving wheel; and establishing the i-th sliding surface according to the processing result of the i-th link, the processing result of the moving wheel, and the angular velocity of the i-th link. The torque calculation module is used to calculate the rotational torque of the n rotating joints based on the n sliding surfaces and the dynamic model parameters. The joint rotation module is used to control the n rotating joints at the second moment based on the rotational torque of the n rotating joints.

11. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program, which is loaded and executed by the processor to implement the balance control method for the wheeled robot as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is loaded and executed by a processor to implement the balance control method for the wheeled-legged robot as described in any one of claims 1 to 9.

13. A computer program product, characterized in that, The computer program product includes a computer program that is loaded and executed by a processor to implement the balance control method for a wheeled robot as described in any one of claims 1 to 9.