Multi-sensor fusion attitude control method for wheeled humanoid robot

Abstract

The invention discloses a multi-sensor fusion attitude control method for a wheeled humanoid robot, and relates to the technical field of robot motion control. The problems that in the prior art, chassis and upper body posture sensing are separated, and whole-body coordination is insufficient can be at least partially solved. The method comprises the following steps: establishing a whole body state vector including chassis, upper body and head postures; collecting data of each inertial measurement unit and generating an adaptive fusion weight based on topographic features; performing multi-source data fusion through extended Kalman filtering to obtain a whole body attitude estimation value; calculating a terrain compensation torque according to the terrain inclination angle and the curvature; a hierarchical control law is constructed based on quaternion errors to generate control instructions of all the parts; predicting center-of-gravity shift caused by upper limb movement and calculating feed-forward compensation torque; and carrying out residual monitoring and zero offset calibration on the sensor. Whole-body posture coordination control over the chassis, the upper body, the head and the two arms is achieved, and the robot has the terrain self-adaptive capacity and the anti-disturbance capacity.

Description

Technical Field

[0001] This invention relates to the field of robot motion control technology, specifically to a multi-sensor fusion posture control method for a wheeled humanoid robot. Background Technology

[0002] As humanoid robots gradually evolve towards practicality and scenario-based applications, wheeled humanoid robots have become the mainstream structure of the new generation of platforms due to their combination of mobility stability and upper limb manipulation capabilities. Compared with bipedal robots, they have advantages in terms of energy consumption and structural reliability. However, due to the independent motion drive characteristics of the chassis and upper body, the posture perception and control of the two parts are often separated, resulting in insufficient overall coordination.

[0003] In existing systems, chassis attitude estimation is mostly performed by the chassis inertial measurement unit (IMU), while the upper body or head attitude is measured by the upper body IMU. However, the inertial coupling effect between the two has not been effectively modeled. When the robot moves on a slope or is subjected to external disturbances, the change in chassis angular velocity and the superposition of the upper body inertial response will cause lag or drift in upper body attitude estimation, further affecting visual perception, arm trajectory, and overall balance performance.

[0004] Traditional complementary filtering can only provide approximate attitude estimation in low-dynamic scenarios and cannot describe the dynamic coupling relationship between the chassis, upper body, and head. Although extended Kalman filtering has multi-sensor fusion capabilities, it is prone to estimation divergence when the sampling frequencies, installation attitudes, and delays of multiple inertial measurement units are mismatched.

[0005] Existing research has made improvements at the single-layer control level, using chassis inertial measurement units for ground tilt compensation or utilizing joint angles to constrain upper body posture, but it has still failed to form a unified posture fusion and coordination control framework for the whole body. Especially in complex terrain, where chassis posture changes frequently, the dynamic coordination of the upper body, arms, and head has become a technical bottleneck affecting the overall stability of the robot.

[0006] Therefore, how to establish an attitude control algorithm that integrates multi-inertial measurement unit collaboration, whole-body dynamics coupling, and terrain disturbance adaptation to achieve stability and coordination of the robot's whole-body posture is a core problem that has not yet been solved in the field of wheeled humanoid robots. Summary of the Invention

[0007] The present invention aims to solve at least one of the technical problems existing in the prior art, and provides a multi-sensor fusion posture control method for wheeled humanoid robots.

[0008] To achieve the above objectives, the present invention provides a multi-sensor fusion attitude control method for a wheeled humanoid robot, comprising: Establish a state vector containing quaternions of chassis, upper body, and head postures, angular velocities, and joint angles; Acceleration and angular velocity data of each inertial measurement unit, as well as angle data of the joint encoder; The terrain tilt angle is calculated based on the acceleration data from the chassis inertial measurement unit, the terrain curvature is calculated based on the wheel speed difference, and terrain adaptive fusion weights are generated. Based on the fusion weights, an extended Kalman filter is used to weight and fuse the pose data to obtain the whole body pose estimate. Based on the terrain slope angle, the terrain curvature and its rate of change, a disturbance vector is constructed, and the terrain compensation torque is calculated. Calculate the quaternion attitude error of each part, adaptively adjust the control gain based on the disturbance amplitude, and generate a control torque command in combination with the compensation torque. Predict the upper limb center of gravity offset based on the joint angle, and calculate the feedforward compensation torque to be superimposed on the chassis control command. The inertial measurement unit is subjected to residual monitoring and weight reduction processing, and the zero bias compensation value is updated periodically.

[0009] Furthermore, the formula for calculating the terrain slope angle is: ; The formula for calculating the curvature of the terrain is: ; in, The terrain slope angle, , , These represent the acceleration components of the chassis inertial measurement unit along three axes. The curvature of the terrain, The speed of the right wheel. For the speed of the left wheel, This refers to the wheelbase of the left and right wheels.

[0010] Furthermore, the terrain adaptive fusion weights are generated as follows: When the terrain tilt angle is greater than a preset tilt angle threshold or the terrain curvature is greater than a preset curvature threshold, the fusion weight of the upper body and head inertial measurement units is increased, and the fusion weight of the chassis inertial measurement unit is decreased. When the terrain tilt angle is less than or equal to the preset tilt angle threshold and the terrain curvature is less than or equal to the preset curvature threshold, each inertial measurement unit adopts a preset reference fusion weight.

[0011] Furthermore, the formula for the weighted fusion is: ; in, The fused pose matrix, , , The terrain adaptive fusion weights are respectively for the chassis, upper body, and head inertial measurement units. , , These are the attitude matrices measured by the corresponding inertial measurement units.

[0012] Furthermore, the disturbance vector is: ; The formula for calculating the terrain compensation moment is: ; in, Let be the disturbance vector. The terrain slope angle, The curvature of the terrain, The rate of change of terrain slope. For the terrain compensation torque, This is the terrain compensation gain matrix.

[0013] Furthermore, the formula for calculating the control torque command is as follows: ; in, For the first Control torque command for the part, The value can be the chassis, upper body, or head. The quaternion attitude error is... The rate of change of the quaternion attitude error. For proportional gain, The differential gain is the percentage gain, and both the proportional gain and the differential gain are disturbance amplitudes. The function.

[0014] Furthermore, the adaptive adjustment method of the control gain is as follows: When the disturbance amplitude is less than a preset disturbance threshold, a low stiffness gain value is used; When the disturbance amplitude is greater than or equal to the preset disturbance threshold, a high stiffness gain value is adopted.

[0015] Furthermore, the formula for predicting the center of gravity offset is: ; The formula for calculating the feedforward compensation torque is as follows: ; in, This is the centroid offset. For the first The mass of the upper limb linkage For the first The length of the upper limb link For the first The angle of an upper limb joint, The feedforward compensation torque, To compensate for the gain at the center of gravity, The rate of change of the center of gravity offset.

[0016] Furthermore, the determination method for the residual monitoring is as follows: When the absolute value of the difference between the acceleration magnitude measured by the inertial measurement unit and the gravitational acceleration is greater than the preset residual threshold, the inertial measurement unit is determined to be abnormal, and it is downweighted or removed.

[0017] Furthermore, the update formula for the zero-bias compensation value is: ; in, This indicates an assignment / update operation. The zero-bias compensation value is... To update the coefficients, The angular velocity measured by the inertial measurement unit. This represents the angular velocity predicted based on the motion model.

[0018] The beneficial effects of this invention are as follows: This invention establishes a whole-body state vector and fuses and estimates data from multiple inertial measurement units under a unified reference frame, effectively avoiding the drift and lag problems caused by independent attitude estimation of each part in traditional methods, and achieving consistent perception of whole-body attitude.

[0019] This invention enables the system to dynamically adjust the contribution ratio of each inertial measurement unit according to the current terrain conditions by identifying terrain features in real time and generating adaptive fusion weights. In complex terrain, it automatically enhances the weight of reliable sensors, significantly improving the robustness and accuracy of attitude estimation.

[0020] This invention employs a hierarchical nonlinear control law based on quaternion error, with the control gain adaptively adjusted according to the disturbance amplitude. This enables the switching capability between compliant control under small disturbances and high stiffness response under large disturbances, effectively enhancing the stability of the robot under different working conditions.

[0021] This invention predicts the center of gravity shift caused by upper limb movement and calculates the feedforward compensation torque, enabling the chassis to respond to upper limb movements in advance, avoiding overall imbalance caused by upper limb movement, and achieving coordinated movement between the chassis and upper limbs.

[0022] This invention features a sensor anomaly detection and zero-bias self-calibration mechanism, enabling continuous whole-body attitude estimation and stable control in the event of sensor failure or communication delay, thus ensuring the long-term operational reliability of the system. Attached Figure Description

[0023] Figure 1 This is a flowchart of the multi-sensor fusion posture control method for a wheeled humanoid robot according to the present invention; Figure 2 This is a schematic diagram of the module composition of the attitude control system of the present invention; Figure 3 This is a comparison chart of the attitude control performance under instantaneous impact disturbance of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and beneficial effects of this application clearer, the following detailed description, in conjunction with the accompanying drawings and specific embodiments, further illustrates this application. It should be understood that the specific embodiments described in this specification are merely for explaining this application and are not intended to limit it.

[0025] The multi-sensor fusion attitude control method for wheeled humanoid robots of the present invention is based on an attitude control system.

[0026] See Figure 2 The attitude control system includes a terrain recognition module, a multi-inertial measurement unit (IMU) fusion module, a whole-body control module, a center of gravity compensation module, and a fault-tolerant calibration module. The terrain recognition module acquires the terrain status in real time using chassis IMU and wheel speed information, calculates parameters such as terrain tilt angle and curvature, and generates a dynamic weight matrix. The multi-inertial measurement unit fusion module uses extended Kalman filtering to weightedly fuse the attitude data from the chassis, upper body, and head IMUs. The whole-body control module receives the fused whole-body attitude status and disturbance information, and generates control commands for the chassis, upper body, head, and arms based on quaternion errors and nonlinear adaptive control laws. The center of gravity compensation module predicts the upper limb center of gravity offset based on joint angles and calculates the feedforward compensation torque. The fault-tolerant calibration module uses residual monitoring, zero-bias estimation, and online filtering to detect the health status of the IMUs and correct drift.

[0027] Example 1 This embodiment uses the full-body posture control of a service robot platform as an application scenario. The robot adopts a humanoid design with a wheeled chassis and an upper body structure. The chassis has a differential drive structure, and the upper body includes a waist rotation joint, two arm joints, and a head pitch joint. The robot is equipped with three inertial measurement units, which are respectively installed at the center of the chassis, the upper body, and the head. At the same time, each joint is equipped with an encoder for measuring joint angles.

[0028] See Figure 1The attitude control method in this embodiment includes the following steps: Step S1: Full-body state modeling.

[0029] Establish the robot's whole-body state vector The expression for the state vector is: .

[0030] in, For the chassis attitude quaternion, The upper body posture quaternion. For head posture quaternions, The chassis angular velocity, The upper body angular velocity, The head angular velocity, These are the angle vectors for each joint.

[0031] In this embodiment, the chassis attitude quaternion It contains four components, each representing the chassis attitude relative to the world coordinate system. Upper body attitude quaternion. This represents the posture of the upper torso relative to the world coordinate system. Head posture quaternion. This represents the head's pose relative to the world coordinate system. (Joint angle vectors) It includes seven joint angles: waist rotation angle, left and right shoulder joint angle, left and right elbow joint angle, and head pitch angle.

[0032] Step S2: Multi-sensor data acquisition.

[0033] Sensor data from the chassis inertial measurement unit, upper body inertial measurement unit, and head inertial measurement unit are collected separately. The chassis inertial measurement unit collects triaxial acceleration and triaxial angular velocity data of the chassis, with a sampling frequency configured at 200Hz. The upper body inertial measurement unit collects triaxial acceleration and triaxial angular velocity data of the upper torso, also with a sampling frequency configured at 200Hz. The head inertial measurement unit collects triaxial acceleration and triaxial angular velocity data of the head, with a sampling frequency configured at 200Hz. Simultaneously, joint angle data from each joint encoder is collected, with all joint encoders having a sampling frequency configured at 500Hz. In addition, the encoders of the left and right wheel motors collect the rotational speed information of the left and right wheels for calculating their speeds.

[0034] Step S3: Terrain feature identification.

[0035] The terrain recognition module calculates the terrain tilt angle based on the acceleration data from the chassis inertial measurement unit. The formula for calculating the terrain tilt angle is as follows: ,in , , These are the acceleration components of the chassis inertial measurement unit in the X, Y, and Z axes, respectively.

[0036] In this embodiment, when the robot travels on a slope with a gradient of 15 degrees, the acceleration component measured by the chassis inertial measurement unit is: =0.12g =2.54g =9.47g, the terrain slope angle is calculated. =15.02 degrees. The terrain curvature is calculated based on the speed difference between the left and right wheels. The formula for calculating terrain curvature is: ,in The speed of the right wheel. For the speed of the left wheel, This refers to the wheelbase of the left and right wheels.

[0037] In this embodiment, the wheelbase of the left and right wheels The right wheel speed is 0.45m when the robot is moving on a curve. =1.2m / s, left wheel speed =1.0m / s, the calculated terrain curvature =0.44 / m. The system performs rolling average filtering on the calculated terrain tilt angle and curvature, with a filter window length of 10 sampling points to eliminate instantaneous noise fluctuations. Adaptive terrain fusion weights are generated based on the filtered terrain tilt angle and curvature. The preset tilt angle threshold is set to 10 degrees, and the preset curvature threshold is set to 0.3 / m. When the terrain tilt angle is greater than 10 degrees or the terrain curvature is greater than 0.3 / m, the fusion weight of the upper body inertial measurement unit is increased from the baseline value of 0.3 to 0.4, the fusion weight of the head inertial measurement unit is increased from the baseline value of 0.2 to 0.3, and the fusion weight of the chassis inertial measurement unit is decreased from the baseline value of 0.5 to 0.3.

[0038] Step S4: Multi-source data fusion.

[0039] The multi-inertial measurement unit (IMU) fusion module uses terrain-adaptive fusion weights and an extended Kalman filter to weight and fuse the attitude data from each IMU. The state prediction equation of the extended Kalman filter is as follows: The observation equation is ,in For the first The state vector at time t, To control the input, For process noise, For the observation vector, To mitigate noise, the attitude quaternion is first updated based on the angular velocity data from each inertial measurement unit (IMU). Then, the gravity direction is calculated using the acceleration data from each IMU and used as an observation to correct the predicted attitude. The weighted fusion formula is as follows: ,in The fused pose matrix, , , These are the terrain-adaptive fusion weights for the chassis, upper body, and head inertial measurement units, respectively. , , The attitude matrices are obtained from the inertial measurement units of the chassis, upper body, and head, respectively. The fused attitude matrix is ​​converted into quaternion form as the whole-body attitude estimate.

[0040] Step S5: Terrain disturbance compensation.

[0041] A perturbation vector is constructed based on the terrain slope, terrain curvature, and rate of change of terrain slope. The expression for the perturbation vector is as follows: ,in For the terrain slope angle, For the curvature of the terrain, This represents the rate of change of terrain dip angle. The rate of change of terrain dip angle is obtained by differential calculation of the terrain dip angle; the calculation formula is as follows: ,in The sampling period is [value]. The terrain compensation torque is calculated based on the perturbation vector and the preset terrain compensation gain matrix. The formula for calculating the terrain compensation torque is: ,in For terrain compensation torque, This is the terrain compensation gain matrix. In this embodiment, the terrain compensation gain matrix... It is a 3×3 diagonal matrix, with diagonal elements set to 5.0, 2.0 and 1.5 respectively, corresponding to the compensation gains for terrain slope angle, terrain curvature and terrain slope angle change rate.

[0042] Step S6: Layered attitude control.

[0043] The whole-body control module calculates the quaternion attitude errors for the chassis, upper body, and head respectively. The calculation method for quaternion attitude errors is as follows: ; in For the first The target attitude quaternion for each part For the first The current attitude quaternion of each part, To represent quaternion multiplication, This represents the quaternion conjugate. The control gain is adaptively adjusted based on the quaternion attitude error and disturbance amplitude. The disturbance amplitude is the L2 norm of the disturbance vector, calculated using the following formula: .

[0044] The preset disturbance threshold is set to 0.3. When the disturbance amplitude is less than 0.3, a low stiffness gain value and proportional gain are used. Set to 8.0, differential gain Set to 2.0. When the disturbance amplitude is greater than or equal to 0.3, a high stiffness gain value is used, with a proportional gain. Set to 15.0, differential gain Set to 5.0. The calculation formula for the control torque command of each part is as follows: The control torque command is sent to the motor driver of each joint after torque limiting processing.

[0045] Step S7: Center of gravity feedforward compensation.

[0046] The center of gravity compensation module predicts the center of gravity shift caused by upper limb movement based on joint angle data. The formula for predicting the center of gravity shift is as follows: ,in For the first The mass of the upper limb linkage For the first The length of the upper limb link For the first The joint angles of each upper limb joint are calculated. In this embodiment, the mass of each upper arm is 0.8 kg and the length is 0.25 m; the mass of each forearm is 0.5 kg and the length is 0.22 m. When the robot's arms extend forward, the center of gravity offset calculated based on the joint angles is used to predict the forward movement of the robot's center of gravity.

[0047] The feedforward compensation torque is calculated based on the rate of change of the center of gravity offset and the preset center of gravity compensation gain. The formula for calculating the feedforward compensation torque is as follows: ,in For feedforward compensation torque, To compensate for the gain at the center of gravity, This represents the rate of change of the center of gravity offset. Center of gravity compensation gain. Set to 12.0. The feedforward compensation torque is superimposed on the chassis control torque command, enabling the chassis to respond in advance to changes in the center of gravity during upper limb movements.

[0048] Step S8: Fault tolerance calibration.

[0049] The fault-tolerant calibration module performs residual monitoring on each inertial measurement unit. The criteria for residual monitoring are as follows: ,in The acceleration magnitude measured by the inertial measurement unit. It is the acceleration due to gravity. A preset residual threshold is set to 1.5 m / s². When the absolute value of the difference between the acceleration magnitude measured by an inertial measurement unit (IMU) and the gravitational acceleration exceeds 1.5 m / s², the IMU is considered to have abnormal data. For abnormal IMUs, the system reduces their fusion weight by 50% or removes them if the abnormality persists for more than 0.5 seconds. Simultaneously, the zero-bias compensation value of each IMU is periodically updated. The update formula for the zero-bias compensation value is as follows: ,in The zero bias compensation value This indicates an assignment / update operation. To update the coefficients, The angular velocity measured by the inertial measurement unit. This represents the angular velocity predicted based on the motion model. Update coefficients. Set to 0.01, and the update cycle to 100ms.

[0050] Example 2 This embodiment uses the full-body posture control of an inspection robot platform in complex terrain as an application scenario.

[0051] The robot needs to perform inspection operations in various terrain conditions within the factory area, such as ramps, gravel roads, and slippery surfaces. At the same time, its upper limbs need to perform tasks such as opening and closing doors and pressing buttons.

[0052] The attitude control method in this embodiment is the same as the steps in Embodiment 1, except that the parameter configuration is different.

[0053] In step S3, the preset tilt angle threshold is set to 8 degrees and the preset curvature threshold is set to 0.25 / m to adapt to the relatively flat terrain characteristics of the industrial plant area, but with local slopes.

[0054] In step S4, the reference fusion weight of the chassis inertial measurement unit is set to 0.4, the reference fusion weight of the upper body inertial measurement unit is set to 0.35, and the reference fusion weight of the head inertial measurement unit is set to 0.25.

[0055] In step S6, the preset disturbance threshold is set to 0.25. The proportional gain in low stiffness mode is set to 10.0, and the derivative gain is set to 2.5. The proportional gain in high stiffness mode is set to 18.0, and the derivative gain is set to 6.0.

[0056] In step S7, the center of gravity compensation gain is set to 15.0 to enhance the center of gravity compensation effect when the upper limb performs operational tasks.

[0057] Example 3 This embodiment illustrates the attitude control performance of the method of the present invention when subjected to instantaneous external disturbances.

[0058] See Figure 3 In the experiment, the robot was subjected to a lateral impact disturbance while stationary, with a peak impact force of 50N and a duration of 0.2 seconds.

[0059] When using the method of this invention, the system immediately switches the control gain to high-stiffness mode after detecting an impact disturbance. The peak value of the head attitude deviation is 1.6 degrees, and the peak value of the waist attitude deviation is 1.1 degrees. The attitude recovery time, i.e., the time required to recover from the peak attitude deviation to a deviation of less than 0.5 degrees, is 0.45 seconds. When using the traditional single inertial measurement unit control method, the peak value of the head attitude deviation is 3.8 degrees, and the peak value of the waist attitude deviation is 2.9 degrees. The attitude recovery time is 1.25 seconds.

[0060] The comparative results show that the method of the present invention reduces the peak attitude deviation by about 58% and shortens the attitude recovery time by about 64% when subjected to external disturbances.

[0061] In summary, the embodiments disclosed herein have at least the following technical effects: This invention achieves the estimation of chassis, upper body, and head posture in a unified reference frame by establishing a whole-body state vector and a multi-inertial measurement unit fusion mechanism, effectively avoiding posture drift and lag issues. Through terrain-adaptive gain adjustment, the invention automatically adjusts the control gain according to the disturbance amplitude, enhancing stability in complex environments. Through center of gravity prediction and feedforward compensation, this invention achieves dynamic coupling and coordination between the chassis and upper limbs. This invention features sensor anomaly detection and zero-drift self-calibration mechanisms, ensuring the long-term operational stability of the system.

[0062] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A multi-sensor fusion attitude control method for a wheeled humanoid robot, characterized in that, include: Establish a state vector containing quaternions of chassis, upper body, and head postures, angular velocities, and joint angles; Acceleration and angular velocity data of each inertial measurement unit, as well as angle data of the joint encoder; The terrain tilt angle is calculated based on the acceleration data of the chassis inertial measurement unit in the inertial measurement unit, the terrain curvature is calculated based on the wheel speed difference, and terrain adaptive fusion weights are generated. Based on the fusion weights, an extended Kalman filter is used to weight and fuse the pose data to obtain the whole body pose estimate. Based on the terrain slope angle, the terrain curvature and its rate of change, a disturbance vector is constructed, and the terrain compensation torque is calculated. The quaternion attitude errors of the chassis, upper body, and head are calculated, and the control gain is adaptively adjusted based on the disturbance amplitude. The control torque command is generated in combination with the terrain compensation torque. Based on the joint angle, the upper limb center of gravity offset is predicted, and the feedforward compensation torque is calculated and superimposed onto the chassis control command. The inertial measurement unit is subjected to residual monitoring and weight reduction processing, and the zero bias compensation value is periodically updated.

2. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The formula for calculating the terrain slope angle is: ； The formula for calculating the curvature of the terrain is: ； in, The terrain slope angle, , , These represent the acceleration components of the chassis inertial measurement unit along three axes. The curvature of the terrain, The speed of the right wheel. For the speed of the left wheel, This refers to the wheelbase of the left and right wheels.

3. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The terrain adaptive fusion weights are generated as follows: When the terrain tilt angle is greater than a preset tilt angle threshold or the terrain curvature is greater than a preset curvature threshold, the fusion weight of the upper body and head inertial measurement units is increased, and the fusion weight of the chassis inertial measurement unit is decreased. When the terrain tilt angle is less than or equal to the preset tilt angle threshold and the terrain curvature is less than or equal to the preset curvature threshold, each inertial measurement unit adopts a preset reference fusion weight.

4. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The formula for the weighted fusion is: ； in, The fused pose matrix, , , The terrain adaptive fusion weights are respectively for the chassis, upper body, and head inertial measurement units. , , These are the attitude matrices measured by the corresponding inertial measurement units.

5. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The disturbance vector is: ； The formula for calculating the terrain compensation moment is: ； in, Let be the disturbance vector. The terrain slope angle, The curvature of the terrain, The rate of change of terrain slope. For the terrain compensation torque, This is the terrain compensation gain matrix.

6. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The formula for calculating the control torque command is as follows: ； in, For the first Control torque command for the part, The value can be the chassis, upper body, or head. The quaternion attitude error is... The rate of change of the quaternion attitude error, For proportional gain, The differential gain is the percentage gain, and both the proportional gain and the differential gain are disturbance amplitudes. The function.

7. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The adaptive adjustment method of the control gain is as follows: When the disturbance amplitude is less than a preset disturbance threshold, a low stiffness gain value is used; When the disturbance amplitude is greater than or equal to the preset disturbance threshold, a high stiffness gain value is adopted.

8. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The formula for predicting the center of gravity offset is: ； The formula for calculating the feedforward compensation torque is as follows: ； in, This is the centroid offset. For the first The mass of the upper limb linkage For the first The length of the upper limb link For the first The angle of an upper limb joint, The feedforward compensation torque, To compensate for the gain at the center of gravity, The rate of change of the center of gravity offset.

9. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to any one of claims 1 to 8, characterized in that, The determination method for the residual monitoring is as follows: When the absolute value of the difference between the acceleration magnitude measured by the inertial measurement unit and the gravitational acceleration is greater than the preset residual threshold, the inertial measurement unit is determined to be abnormal, and it is downweighted or removed.

10. The multi-sensor fusion attitude control method for a wheeled humanoid robot according to claim 1, characterized in that, The update formula for the zero-bias compensation value is: ； in, This indicates an assignment / update operation. The zero-bias compensation value is... To update the coefficients, The angular velocity measured by the inertial measurement unit. This represents the angular velocity predicted based on the motion model.

Images