A speed coordination method and system for a biped robot with cross-modal credibility constraints

Abstract

The application discloses a biped robot cross-mode reliability constraint speed coordination method and system, relates to the field of motion control technology, and calculates the comprehensive reliability and the time variation rate through environment perception data and target motion data, carries out segmented gain processing on the time variation rate, and obtains original guide parameter; the original guide parameter is verified based on the virtual energy pool mechanism, virtual mass and virtual damping coefficient are obtained, a shadow centroid model is constructed, the target motion data is filtered, and the first speed parameter is obtained; the current pose and motor parameters of the biped robot are acquired, the safety envelope is obtained through the contraction of the comprehensive reliability, the first speed parameter is verified, the second speed parameter and the continuous boundary condition are obtained; the safety envelope, the continuous boundary condition and the second speed parameter are calculated through constraint coupling and inverse dynamics optimization, and the foot end contact force and the joint torque instruction are obtained.

Description

Technical Field

[0001] This invention relates to the field of motion control technology, and in particular to a speed coordination method and system for cross-modal reliability constraints of bipedal robots. Background Technology

[0002] Bipedal robots have become a hot topic in robotics research in recent years due to their human-like locomotion and ability to adapt to complex terrain.

[0003] Traditional mobile robots typically employ a hard braking strategy of zeroing speed when sensing signal loss, or directly multiply the confidence level by the desired speed as the command output. While this approach is acceptable for four-wheeled robots, for bipedal robots with high degrees of freedom and underactuated characteristics, a sudden jump in the overall confidence level can generate infinite acceleration in the speed command, causing the center of gravity to overturn momentarily beyond its limit, directly triggering instability in divergent motion components and even leading to a fall. In extreme scenarios such as high-speed movement or strong winds, air resistance significantly impacts the posture stability and motor output of bipedal robots. However, existing methods often ignore these aerodynamic forces as unmodelable random disturbances, failing to fully utilize their potential value as environmental auxiliary resources. This results in motors operating under unnecessarily high loads, leading to high energy consumption and susceptibility to overheating. Furthermore, existing control architectures lack active cutoff and fuse protection capabilities, allowing abnormal signals to be amplified and propagated along the link to the underlying actuators, resulting in insufficient overall survivability. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies have difficulty in solving the problem of smooth and continuous speed commands when sensing sudden changes. Abnormal signals are amplified and transmitted to the underlying actuators at each level along the link, and there is a lack of active interception and circuit breaker protection capabilities.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, a velocity coordination method for cross-modal reliability constraints of a bipedal robot includes the following steps: Step S1: Calculate the comprehensive confidence level and time change rate using environmental perception data and target motion data; perform piecewise gain processing on the time change rate to obtain the original admittance parameter. Step S2: Verify the original admittance parameters based on the virtual energy pool mechanism to obtain the virtual mass and virtual damping coefficient, construct the shadow centroid model, filter the target motion data, and obtain the first velocity parameter; Step S3: Obtain the current pose and motor parameters of the bipedal robot, shrink the sensor using the comprehensive confidence level to obtain a safety envelope, and verify the first velocity parameter to obtain the second velocity parameter and continuous boundary conditions. Step S4: Through constraint coupling and inverse dynamics optimization, the safety envelope, continuous boundary conditions and second velocity parameters are calculated to obtain the foot contact force and joint torque command.

[0006] Preferably, step S1 includes the following sub-steps: Step S11: Obtain environmental perception data and target motion data of the bipedal robot, calculate the Mahalanobis distance between the visual observation position and the radar observation position to obtain the spatial consistency confidence, use the extended Kalman filter to recursively calculate the target motion data to obtain the predicted speed, and compare it with the current measured target speed to obtain the temporal coherence confidence. Step S12: The spatial consistency confidence and temporal coherence confidence are fused through a joint probability model to obtain a normalized comprehensive confidence. The normalized comprehensive confidence is then subjected to temporal difference to extract the smoothed confidence change rate. Step S13: Based on the confidence change rate, construct a piecewise mapping function to map the confidence change rate to the original admittance parameters, which include the original virtual mass and the original virtual damping coefficient.

[0007] Preferably, the piecewise mapping function includes a first gain function and a second gain function, and the processing logic is as follows: When the rate of change of the credibility When the value is 0, the first gain calculation is triggered. The confidence change rate is processed by the first gain function to obtain the corresponding original virtual mass and original virtual damping coefficient. When the rate of change of the credibility When the value is 0, the second gain calculation is triggered. The confidence change rate is calculated using the second gain function to obtain the corresponding original virtual mass and original virtual damping coefficient.

[0008] Preferably, step S2 includes the following sub-steps: Step S21: Construct a virtual energy pool mechanism. Calculate the transient spurious power based on the current speed, following deviation, and original virtual mass of the bipedal robot. Perform discrete integration on the transient spurious power to obtain the remaining energy state of the virtual energy pool. Step S22: Compare the remaining energy state with a preset safe energy threshold. When the remaining energy state is greater than the safe energy threshold, the energy is determined to be sufficient, and the original admittance parameter is used as the target admittance parameter. The target admittance parameter includes the target virtual mass and the target virtual damping coefficient. When the remaining energy state is less than or equal to the safe energy threshold, it is determined that the passivity has been destroyed, triggering the cutoff mechanism. The original virtual mass is corrected to the safe mass constant as the target virtual mass, and the original virtual damping coefficient is subjected to over-limit dynamic gain as the target virtual damping coefficient. Step S23: Construct a shadow centroid model using the target admittance parameters, and perform differential calculations on the target motion data to obtain the expected centroid velocity, expected centroid acceleration, and expected centroid jerk, which are then used as the first velocity parameter.

[0009] Preferably, step S3 includes the following sub-steps: Step S31: Obtain the current pose, motor parameters and relative airflow velocity of the bipedal robot; calculate the estimated aerodynamic drag value at the current moment based on the relative airflow velocity; substitute the estimated aerodynamic drag value into the motor power boundary model for correction to obtain the corrected power boundary; and simultaneously fuse the geometric step size boundary to obtain the initial motion boundary. Step S32: Use the comprehensive confidence level as a shrinkage scaling factor to process the initial motion boundary and generate a safe envelope; Step S33: Obtain the contact state switching critical point and high dynamic deceleration range of the continuous phase variable, calculate the inertial impact force of the first velocity parameter, superimpose the inertial impact force and the estimated aerodynamic drag to obtain the net inertial residual force, process the net inertial residual force, and output the second velocity parameter and continuous boundary conditions.

[0010] Preferably, the processing logic for processing the net inertial residual force and outputting the second velocity parameter and continuous boundary conditions is as follows: Extract the unit direction vector of the current acceleration of the bipedal robot, and perform a dot product operation between the estimated aerodynamic drag value and the unit direction vector to obtain the projection scalar of the aerodynamic drag in the acceleration direction; Subtract the projection scalar from the inertial impact force scalar value of the first velocity parameter to obtain the net inertial residual force; The net inertial residual force is compared with a preset force threshold: When the net inertial residual force is less than or equal to the force threshold, it is determined that the current inertial impact is within the safe range, and the first velocity parameter is output as the second velocity parameter. When the net inertial residual force is greater than the force threshold, it is determined that the current inertial impact exceeds the safe range, and an extension process is performed. The extension process includes: Calculate the difference between the net inertial residual force and the force threshold, and divide the difference by the force threshold to obtain the over-limit ratio coefficient; obtain the reference gait execution cycle of the bipedal robot, and calculate the safe gait execution cycle, wherein the safe gait execution cycle = reference gait execution cycle. Excess ratio coefficient The continuous phase variables are corrected by the safe gait execution cycle to obtain the current action execution rate; Calculate the initial jerk value corresponding to the net inertial residual force, compare the initial jerk value with the preset upper limit value of jerk, and if it exceeds the limit, forcibly truncate and correct the initial jerk value to the upper limit value of jerk. Trajectory reconstruction is performed based on the current action execution rate, and a second velocity parameter and continuous boundary conditions are output. The continuous boundary conditions are used to limit the foot velocity to zero at the critical point of contact state switching.

[0011] Preferably, step S4 includes the following sub-steps: Step S41: Construct a constrained optimization model based on the second velocity parameter, safety envelope, and continuous boundary conditions; Step S42: Obtain the joint space mass matrix and bias force matrix of the bipedal robot, substitute the joint space mass matrix and bias force matrix into the constraint optimization model, and perform inverse operation to obtain the foot contact force and joint acceleration. Step S43: Based on the foot contact force and joint acceleration, and combined with the actuator selection matrix and contact Jacobian matrix of the bipedal robot, the joint torque command is calculated, and the foot contact force and the joint torque command are sent to the joint actuator of the bipedal robot for execution.

[0012] Preferably, step S42 further includes the following sub-steps: Step S421: Obtain the current actual joint data of the bipedal robot. The actual joint data includes joint position and joint velocity. Calculate the joint space mass matrix based on the multibody rigid body dynamics algorithm, calculate the bias force matrix, and solve for the contact Jacobian matrix and the corresponding time derivative matrix corresponding to the contact point of the bipedal robot's foot. Step S422: Using the joint space mass matrix, bias force matrix, and contact Jacobian matrix, construct the dynamic equation of the bipedal robot; using the contact Jacobian matrix and time derivative matrix, construct the motion mapping equation that transforms from the joint acceleration space to the foot end Cartesian space; and input the dynamic equation and the motion mapping equation as constraints into the constraint optimization model. Step S423: Set the foot contact force and joint acceleration as optimization decision variables, start the quadratic programming solver, perform convergence calculation on the constrained optimization model, output the optimal solution vector, and decompose the foot contact force and joint acceleration from the optimal solution vector.

[0013] Preferably, the process of sending the foot contact force and the joint torque command to the joint actuator of the bipedal robot is as follows: A force-position composite control loop is constructed inside the underlying servo controller. The force-position composite control loop includes a feedforward channel and a compensation feedback channel. The joint torque command is input into the feedforward channel. The foot contact force is mapped into a joint impedance compensation torque by using the contact Jacobian matrix transpose of the bipedal robot and input into the compensation feedback channel. The actual joint angles and actual joint velocities fed back in real time by the encoders of each joint of the bipedal robot are obtained. The motion deviation is calculated based on the target joint angles and target joint velocities. The motion deviation is calculated using a PD regulator to obtain the closed-loop stabilizing torque. The feedforward torque, joint impedance compensation torque and closed-loop stabilizing torque are summed to generate a comprehensive torque control quantity. The comprehensive torque control quantity is discretized and packaged according to the underlying servo control cycle to obtain a torque control data packet. The torque control data packet is synchronously sent to the end effector of each physical joint of the bipedal robot through a real-time industrial communication bus. The end effector is analyzed and amplified by the current loop inside the end effector to drive the motor actuator of the corresponding joint to output the comprehensive torque control quantity, thereby completing the bipedal robot's motion posture and contact force distribution.

[0014] Secondly, a velocity coordination system for cross-modal reliability constraints of a bipedal robot includes a mapping module, an energy verification module, a safety envelope module, and a control constraint module. The mapping module calculates the comprehensive confidence level and time change rate by using environmental perception data and target motion data, and performs piecewise gain processing on the time change rate to obtain the original admittance parameter. The energy verification module verifies the original admittance parameters based on the virtual energy pool mechanism, obtains the virtual mass and virtual damping coefficient, constructs a shadow centroid model, and filters the target motion data to obtain the first velocity parameter. The safety envelope module acquires the current pose and motor parameters of the bipedal robot, shrinks it using the comprehensive confidence level to obtain a safety envelope, and verifies the first velocity parameter to obtain the second velocity parameter and continuous boundary conditions. The control constraint module calculates the safety envelope, continuous boundary conditions, and second velocity parameters through constraint coupling and inverse dynamics optimization to obtain foot contact force and joint torque commands. The beneficial effects of this invention are as follows: This invention maps the rate of change of confidence to virtual mass and damping coefficient through a piecewise gain function, and utilizes the inherent low-pass filtering characteristics of physical inertia to smoothly absorb sudden changes in commands, achieving a continuous C2 transition of the center of mass velocity, acceleration, and jerk throughout the entire process. The piecewise function provides the virtual mass with millisecond-level surges to counteract sudden disturbances when the system environment deteriorates, and linear release to avoid forward instability when the environment recovers. When sensor hardware failure causes high-frequency oscillations in admittance parameters, automatic fuse protection is provided, establishing an energy isolation zone between the information domain and the physical domain. The introduction of the aerodynamic drag model transforms environmental wind force from a random disturbance into a controllable auxiliary resource, reducing motor load in headwinds and anticipating braking difficulties in tailwinds. The confidence-driven safety envelope dynamic scaling actively reserves emergency braking margins for the control system, achieving real-time coupling between perceived mass and motion limits. The time-domain stretching processing uses a time-for-space strategy to spread the acceleration peak within the safety red line, combined with hard jerk cutoff and zero-speed landing constraints at the feet. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the steps of a speed coordination method for cross-modal reliability constraints of a bipedal robot, as provided in one embodiment of the present invention; Figure 2 This is a basic flowchart illustrating a speed coordination system for cross-modal reliability constraints of a bipedal robot, as provided in one embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example 1, referring to Figure 1 A velocity coordination method for cross-modal reliability constraints of a bipedal robot is provided, comprising the following steps: Step S1: Calculate the comprehensive confidence level and time change rate using environmental perception data and target motion data, and perform piecewise gain processing on the time change rate to obtain the original admittance parameters; Step S2: Verify the original admittance parameters based on the virtual energy pool mechanism to obtain the virtual mass and virtual damping coefficient, construct the shadow centroid model, filter the target motion data, and obtain the first velocity parameter; Step S3: Obtain the current pose and motor parameters of the bipedal robot, shrink the pose using the comprehensive confidence level to obtain the safety envelope, verify the first velocity parameter, and obtain the second velocity parameter and continuous boundary conditions. Step S4: Through constraint coupling and inverse dynamics optimization, the safety envelope, continuous boundary conditions and second velocity parameters are calculated to obtain the foot contact force and joint torque command.

[0018] Step S1 includes the following sub-steps: Step S11: Obtain environmental perception data and target motion data of the bipedal robot, calculate the Mahalanobis distance between the visual observation position and the radar observation position to obtain the spatial consistency confidence, use the extended Kalman filter to recursively calculate the target motion data to obtain the predicted speed, and compare it with the current measured target speed to obtain the temporal coherence confidence. Existing technologies typically use a fixed Euclidean distance threshold; for example, a difference greater than 1 meter is considered a failure. This embodiment introduces Mahalanobis distance and covariance matrix, enabling it to adapt to the blind zones and error characteristics of different sensors in different directions. For example, radar ranging may be accurate but angle-dependent, while visual ranging may be accurate but ranging may be inaccurate, thus affecting spatial consistency confidence. The calculation expression is: ; in, The visual observation position vector, This is the radar observation position vector. The spatial observation covariance matrix for joint calibration of the two sensors is used to represent the noise distribution characteristics of each sensor. The above formula is used to smoothly map the error from 0 to positive infinity to a confidence level between (0,1]. When the two sensors diverge in their respective directions, the confidence level will drop rapidly. However, when they have slight divergence in their original calibration error, a tolerance level will be generated to avoid false alarms. The robot's current measured target velocity is extracted and compared with the predicted velocity obtained by recursion using extended Kalman filtering to calculate the time coherence confidence score. The calculation expression is: ; in, The target speed is the one currently measured. To predict speed, This is the continuity attenuation coefficient, and it is a positive number. , The maximum safe speed step error threshold for the center of mass of the bipedal robot; Step S12: The spatial consistency confidence and temporal coherence confidence are fused through a joint probability model to obtain the normalized comprehensive confidence. The normalized comprehensive confidence is then subjected to temporal difference to extract the smoothed confidence change rate. In this embodiment, the spatial consistency confidence score and the temporal coherence confidence score are multiplied to obtain the normalized overall confidence score. The range of values ​​is Regarding overall credibility The original rate of change was calculated using the first-order backward difference. Subsequently, the smoothed confidence change rate was extracted using a dynamic filtering function. : ; ; Suppose the sensor detects a disturbance with an extremely similar shape, spatial confidence level =0.9, but the interfering object appeared at an extremely fast speed that violated the laws of physics, and the temporal continuity was... =0.1. If a weighted summation is used, the total confidence level is still 0.5, meaning the bipedal robot will still twitch and take a step due to the incorrect target. The multiplicative model in this embodiment yields... =0.09, directly triggering a danger alarm to avoid the risk of being misled by erroneous data; Step S13: Based on the confidence change rate, construct a piecewise mapping function to map the confidence change rate to the original admittance parameters, which include the original virtual mass and the original virtual damping coefficient.

[0019] This embodiment makes a prediction based on the positive or negative sign of the confidence change rate. When the environment shows a deteriorating trend, the defense mechanism will intervene in advance without waiting for the overall confidence to completely drop to the bottom before reacting. At the same time, the original virtual mass output in this step directly constitutes the core input data for calculating the transient power change of the virtual energy pool. Through high-dimensional admittance parameter mapping, abnormal commands are eliminated by using virtual physical properties, realizing a continuous and smooth transition from normal following to safe hovering.

[0020] The piecewise mapping function includes a first gain function and a second gain function, and the processing logic is as follows: When the rate of change in credibility When the value is 0, the first gain calculation is triggered. The confidence change rate is processed by the first gain function to obtain the corresponding original virtual mass and original virtual damping coefficient. When the rate of change in credibility When the value is 0, the second gain calculation is triggered. The confidence change rate is calculated using the second gain function to obtain the corresponding original virtual mass and original virtual damping coefficient.

[0021] In this embodiment, when the rate of change in credibility When the first gain function is triggered, it is determined that the system's environmental perception state has deteriorated, possibly due to a sudden occlusion of the target. To suppress potential erroneous actions by the robot, the first gain function is used to amplify the confidence rate of change, and the original virtual mass and original virtual damping coefficient are calculated. The expression for the first gain function is as follows: ; ; in, The original virtual quality at the current moment. The original virtual damping coefficient calculated at the current moment. and These represent the basic physical mass and basic damping constant of the bipedal robot. and This is a fallback proportional gain coefficient, applied when the environment is continuously deteriorating, i.e., the rate of change approaches 0, but the absolute value has extremely low reliability, providing high virtual inertia. and For the surge differential gain coefficient, This is the curvature constant factor, used to adjust the steepness of the mass surge curve. The rate of change in credibility after smoothing; Suppose a bipedal robot is following its owner when a car suddenly drives by and blocks the owner's view. At the instant the robot is obscured, the rate of change in its credibility drops dramatically, affecting the robot's virtual quality. The speed increases instantly. At this point, even if the sensing sensor sends an incorrect speed command to the underlying layer because it loses the target, the robot will maintain its position without violent twitching due to its current virtual inertia being infinite, thus avoiding the tipping of its center of gravity and falling. When the rate of change in credibility When the value is 0, the calculation of the second gain function is triggered: At this point, the system's environmental perception state is determined to be stable or recovering. Obstacles may have moved away, and the target may become clear again. To prevent the robot from experiencing forward instability, a low-gain calculation is performed using a second gain function to obtain the corresponding original virtual mass and original virtual damping coefficient. The expression for the second gain function is: ; ; in, The overall credibility after normalization at the current moment. and For low gain, the coefficient varies slowly. The curvature constant factor; When the car drove away and the owner's image became clear again, the rate of change in credibility... When the weight changes from negative to positive, the system must not instantly reduce the virtual mass back to the normal 55kg. During the period of target loss, the robot's internal controller has accumulated a significant position tracking error. If the weight were to decrease instantly, the robot would experience lurching due to the enormous compensation torque, causing it to fall face-down. After triggering the second gain function, a linear formula is used... The robot's virtual weight will gradually fade away as the overall credibility increases, guiding the robot to slowly restart and eventually return to its normal following speed.

[0022] To verify the effectiveness of the piecewise mapping function under extreme occlusion conditions, a dynamic following test was conducted on a bipedal robot, with the robot's basic mass set. =55kg, Basic damping =150Ns / m. At t=1.0s, an obstruction was artificially applied, causing the target to suddenly disappear. At t=1.5s, the obstruction was removed. The real-time mapping data of the original admittance parameters are shown in Table 1. Table 1: Original Admittance Parameter Mapping Data Table ; As shown in Table 1, the sudden occlusion of the target at t=1.0s leads to... When the value is less than 0, the first gain function is triggered, and the virtual mass instantly increases from 56.5 kg to 268.4 kg. The virtual inertia causes the robot to become physically inert, forcibly absorbing the sudden change in command caused by the loss of the target. However, when the target reappears at t=1.5s, although... It exhibits a positive jump of +15.0, but triggers the second gain function, causing the virtual mass to gradually drop back to 61.0 kg, thus preventing the robot from lunging forward violently.

[0023] Step S2 includes the following sub-steps: Step S21: Construct a virtual energy pool mechanism. Calculate the transient spurious power based on the current speed, following deviation, and original virtual mass of the bipedal robot. Perform discrete integration on the transient spurious power to obtain the remaining energy state of the virtual energy pool. Define speed following deviation Calculate the transient spurious power based on the difference between the current velocity and the target velocity of the bipedal robot. : ; Remaining energy state of the virtual energy pool Perform discrete integral updates: ; in, The time derivative of the original virtual mass. This represents the original virtual damping coefficient. For discrete control periods, The baseline energy recovery rate set for the system's underlying layer is 5W in this embodiment to prevent the energy pool from naturally depleting due to long-term normal operation. The purpose of this embodiment is to address the additional power generated by drastic changes in mass. In this embodiment, a virtual energy reservoir is created to record and quantify in real time whether the changes in the original admittance parameters are consuming kinetic energy normally. Step S22: Compare the remaining energy state with the preset safe energy threshold: When the remaining energy state is greater than the safe energy threshold, the energy is determined to be sufficient, and the original admittance parameters are used as the target admittance parameters. The target admittance parameters include the target virtual mass and the target virtual damping coefficient. When the remaining energy state is less than or equal to the safe energy threshold, it is determined that the passivity has been destroyed, triggering the cutoff mechanism. The original virtual mass is corrected to the safe mass constant as the target virtual mass, and the original virtual damping coefficient is subjected to over-limit dynamic gain, which is then used as the target virtual damping coefficient. The safe energy threshold is set to 10J. The remaining energy state is compared with the safe energy threshold to obtain the target admittance parameter. Assuming that the radar sensor has a hardware failure and continuously outputs a target jump signal of ±180 degrees within 0.1 seconds, the original virtual mass will oscillate at high frequency. The oscillation of the original virtual mass will inject false energy into the control system. At this time, the virtual energy pool will quickly fall below the safe energy threshold. The system will immediately trigger the circuit breaker protection and force the virtual mass to be set to the safe mass constant, i.e. the basic physical mass. At the same time, the speed following deviation is used to superimpose the over-limit dynamic gain on the damping coefficient.

[0024] Step S23: Construct a shadow centroid model using the target admittance parameters, and perform differential calculations on the target motion data to obtain the expected velocity, expected acceleration, and expected jerk of the centroid, which are then used as the first velocity parameter.

[0025] In addition to the actual physical control of the bipedal robot, this embodiment constructs a shadow centroid model using a virtual target mass and a virtual target damping coefficient. This creates a virtual damping traction mechanism between the target velocity and the robot's actual velocity, preventing the robot from directly responding to step commands. The continuous-domain differential equation of the shadow centroid model is: ; in, The current measured target speed. The current expected velocity of the shadow centroid, i.e., the filtered velocity. The current expected acceleration of the shadow's center of mass. For the target virtual mass, the left side of the above equation represents the virtual inertial force, and the right side represents the virtual damping force pulling towards the target velocity, with units of N (Newtons). Within the discrete control cycle of the bipedal robot, the above differential equation is calculated using the explicit Euler method. First, the target admittance parameter for the current cycle is extracted, and then the expected acceleration of the shadow's center of mass is calculated. : ; Specifically, the expected acceleration of the centroid is integrated to obtain the updated expected velocity of the centroid. The expected acceleration of the centroid in the current cycle is subjected to a first-order backward difference, and the derivative is obtained to obtain the expected jerk of the centroid. The expected velocity, expected acceleration, and expected jerk of the centroid calculated in the current control cycle are used as the first velocity parameters and output to the safety envelope module.

[0026] Step S3 includes the following sub-steps: Step S31: Obtain the current pose, motor parameters and relative airflow velocity of the bipedal robot; calculate the estimated aerodynamic drag value at the current moment based on the relative airflow velocity; substitute the estimated aerodynamic drag value into the motor power boundary model for correction to obtain the corrected power boundary; and simultaneously fuse the geometric step size boundary to obtain the initial motion boundary. Step S32: Use the overall confidence level as a shrinkage scaling factor to process the initial motion boundary and generate a safe envelope; Step S33: Obtain the contact state switching critical point and high dynamic deceleration range of the continuous phase variable, calculate the inertial impact force of the first velocity parameter, and obtain the net inertial residual force by calculating the estimated value of the inertial impact force and aerodynamic drag. Process the net inertial residual force and output the second velocity parameter and continuous boundary conditions.

[0027] In this embodiment, traditional control methods usually treat air resistance as a disturbance and ignore it or simply compensate for it. However, this invention is the first to incorporate aerodynamic resistance as an environmental auxiliary braking resource into the motor power boundary model. At the same time, it combines comprehensive reliability to dynamically scale the motion boundary and uses time-domain stretching technology to achieve safe and smooth deceleration under high dynamic conditions, ensuring the stability and compliance of the bipedal robot under extreme working conditions. Obtain the relative airflow velocity of the bipedal robot The relative airflow velocity is the vector difference between the robot's center-of-mass velocity and the ambient wind speed. The estimated aerodynamic drag value at the current moment is calculated based on the fluid dynamics equations. : ; in, air density ( ), The drag coefficient of the bipedal robot is dimensionless and calibrated by CFD simulation. The effective projected area of ​​the robot in the windward direction ( ), Let be the vector velocity of the robot relative to the air (m / s). The rated maximum driving force envelope of each drive motor of the bipedal robot is obtained, calculated from the peak torque and reduction ratio of the motors. The estimated aerodynamic drag is vector-superimposed with the motor envelope to obtain the corrected power boundary. Simultaneously, based on the geometric kinematic model of the robot's legs, the maximum geometric step length boundary of the center of mass space is mapped. The intersection of the corrected power boundary and the geometric step length boundary is taken to obtain the initial motion boundary. The corrected power boundary shows that, in the headwind direction, aerodynamic drag actually provides the robot with additional braking torque, making its equivalent braking capacity greater than the nominal value of the pure motor. In the tailwind direction, the braking torque is canceled out by the wind force, and the actual braking capacity decreases. Assuming the robot stops abruptly in a headwind of 10 m / s, the direction of aerodynamic drag is opposite to the velocity. At this point, the upper limit of the corrected power boundary in the deceleration direction is significantly higher than the motor limit in the windless state, which is equivalent to providing an additional brake pad for the control system. The controller can utilize this braking force to reduce the motor torque output, thereby reducing energy consumption and heat generation. The normalized comprehensive confidence is extracted and used as a scaling factor to perform an inward compression mapping on the initial motion boundary, generating the final safety envelope. The desired acceleration (the desired acceleration of the shadow centroid model) is extracted from the first velocity parameter, and the projected scalar of aerodynamic drag in the acceleration direction is calculated. If the robot stops suddenly in a headwind, at this time When the wind resistance is greater than 0, the direction of wind resistance is opposite to the direction of acceleration, and the projection is positive. Therefore, the net inertial residual force will be less than the pure inertial force, reducing the required motor output and avoiding unnecessary energy consumption and motor overload. Conversely, during a sudden stop with the wind... When the net residual force is less than 0, the net inertial residual force increases, triggering safety measures. When the net residual force exceeds the upper limit of the safety envelope, forcibly executing the deceleration command within the original cycle will generate acceleration spikes, with the acceleration approaching infinity, easily leading to robot imbalance. This embodiment extends the action time to reduce the peak acceleration.

[0028] The logic for processing the net inertial residual force and outputting the second velocity parameter and continuous boundary conditions is as follows: Extract the unit direction vector of the current acceleration of the bipedal robot, and perform a dot product operation between the estimated aerodynamic drag value and the unit direction vector to obtain the projection scalar of aerodynamic drag in the acceleration direction; Subtract the projected scalar from the scalar value of the inertial impact force of the first velocity parameter to obtain the net inertial residual force. The net inertial residual force is compared with a preset force threshold: When the net inertial residual force is less than or equal to the force threshold, it is determined that the current inertial impact is within the safe range, and the first velocity parameter is output as the second velocity parameter. When the net inertial residual force exceeds the force threshold, the current inertial impact is determined to be outside the safe range, and an extension process is executed. The extension process includes: Calculate the difference between the net inertial residual force and the force threshold, and divide the difference by the force threshold to obtain the over-limit ratio coefficient; obtain the reference gait execution cycle of the bipedal robot, calculate the safe gait execution cycle, and the safe gait execution cycle = reference gait execution cycle. Excess ratio coefficient The current action execution rate is obtained by correcting the continuous phase variables through the safe gait execution cycle; Calculate the initial jerk value corresponding to the net inertial residual force, compare the initial jerk value with the preset upper limit of jerk, and if it exceeds the limit, forcibly truncate and correct the initial jerk value to the upper limit of jerk. Trajectory reconstruction is performed based on the current action execution rate, and a second velocity parameter and continuous boundary conditions are output. The continuous boundary conditions are used to limit the foot velocity to zero at the critical point of contact state switching.

[0029] In this embodiment, a force threshold of 350N is preset. The maximum static friction force is derived based on the typical ground friction coefficient μ=0.6, and the maximum permissible horizontal impact force without tipping over is derived based on the robot's current supporting polygon area and ZMP stability margin. If the force exceeds the threshold, the robot will inevitably slip or tip over. The baseline gait execution cycle is 0.8s, which is the natural periodic frequency of the inverted pendulum. Based on the bipedal robot's leg length L=0.8m, the periodicity formula of the linear inverted pendulum model is used. This represents the energy-efficient time it takes for the robot to take a step without extreme interference, with a maximum jerk value of 30. The design incorporates harmonic reducers for bipedal robot joints to prevent wear caused by sudden braking.

[0030] Step S4 includes the following sub-steps: Step S41: Construct a constrained optimization model based on the second velocity parameter, safety envelope, and continuous boundary conditions; Step S42: Obtain the joint space mass matrix and bias force matrix of the bipedal robot, substitute the joint space mass matrix and bias force matrix into the constraint optimization model, and perform inverse operation to obtain the foot contact force and joint acceleration. Step S43: Based on the foot contact force and joint acceleration, combined with the actuator selection matrix and contact Jacobian matrix of the bipedal robot, the joint torque command is calculated, and the foot contact force and joint torque command are sent to the joint actuator of the bipedal robot for execution.

[0031] In this embodiment, a constrained optimization model is constructed based on the second velocity parameter, the safety envelope, and continuous boundary conditions: ; in, , is the second velocity parameter Calculate the target center of mass acceleration. This is the joint torque vector. The contact force vector at the foot end. To access the Jacobian matrix, For joint velocity vectors, , and This is the corresponding weight matrix; Step S42 also includes the following sub-steps: Step S421: Obtain the current actual joint data of the bipedal robot. The actual joint data includes joint position and joint velocity. Calculate the joint space mass matrix based on the multibody rigid body dynamics algorithm, calculate the bias force matrix, and solve the contact Jacobian matrix and the corresponding time derivative matrix corresponding to the contact point of the bipedal robot's foot. Obtain the current actual joint data of the bipedal robot, including joint positions. and joint velocity The joint space mass matrix is ​​calculated based on the multibody rigid body dynamics algorithm. Let n be the number of robot degrees of freedom, and calculate the bias force matrix. It includes Coriolis force, centrifugal force, and gravity terms, and solves the contact Jacobian matrix corresponding to the contact points at the feet of the bipedal robot. , m is the contact constraint dimension and its time derivative matrix. ; Step S422: Using the joint space mass matrix, bias force matrix, and contact Jacobian matrix, construct the dynamic equation of the bipedal robot. Using the contact Jacobian matrix and time derivative matrix, construct the motion mapping equation that transforms from the joint acceleration space to the foot end Cartesian space. Use the dynamic equation and motion mapping equation as constraints and input them into the constraint optimization model. The dynamic equations of the bipedal robot are constructed using the joint space mass matrix, bias force matrix, and contact Jacobian matrix: ; Using the contact Jacobian matrix and the time derivative matrix, a motion mapping equation is constructed to transform the joint acceleration space into the foot Cartesian space: ; in Let be the Cartesian acceleration vector of the foot contact point. The dynamic equation and the motion mapping equation are used as constraints and input into the constraint optimization model.

[0032] Step S423: Set the foot contact force and joint acceleration as optimization decision variables, start the quadratic programming solver, perform convergence calculation on the constrained optimization model, output the optimal solution vector, and decompose the foot contact force and joint acceleration from the optimal solution vector.

[0033] This step ensures that the final output foot contact force and joint acceleration are globally optimal, representing a unique solution that satisfies all physical constraints. When the bipedal robot is in single-leg support mode, the number of rows in the contact Jacobian matrix is ​​m=6, representing three forces plus three torque constraints, with friction cone constraints ensuring that the foot does not slip. When in two-leg support mode, m=12, and the optimizer needs to redistribute the load between the two sets of contact forces to ensure that the total resultant force is consistent with the desired acceleration. By embedding the dynamic equations and motion mapping equations as equality constraints into the optimization model, this embodiment avoids the constraint neglect problem of traditional inverse dynamics methods. The efficient convergence characteristics of the quadratic programming solver ensure that the optimization can be completed within each control cycle (approximately 1ms), meeting real-time requirements. Based on the contact force at the foot With joint acceleration Combining the actuator selection matrix S and the contact Jacobian matrix of the bipedal robot The joint torque command is calculated as follows: ; Contact force at the foot Joint torque command The actuator selection matrix S is sent to the joint actuators of the bipedal robot for execution. The purpose of the actuator selection matrix S is to exclude the torque calculation of passive joints (such as spring suspension joints or undriven redundant joints), ensuring that torque commands are only issued to actively controlled joints to avoid erroneous driving. The actuator selection matrix S is used to select active joints and is usually a diagonal matrix, with active joints corresponding to 1s and passive joints corresponding to 0s.

[0034] Taking a bipedal robot in a single-leg support switching and emergency stop condition as an example, the airborne quadratic programming solver (QPSolver) jointly solves the dynamic equation, friction cone and continuous boundary conditions in each control cycle (1ms). The output optimal solution vector decomposition results are shown in Table 2. Table 2: Results of the Constrained Optimization Model ; The process of sending foot contact force and joint torque commands to the joint actuators of the bipedal robot is as follows: A force-position composite control loop is constructed inside the underlying servo controller. The force-position composite control loop includes a feedforward channel and a compensation feedback channel. The joint torque command is input into the feedforward channel. The contact Jacobian matrix transpose of the bipedal robot is used to map the foot contact force into the joint impedance compensation torque and input it into the compensation feedback channel. The actual joint angles and velocities of each joint in the bipedal robot are obtained in real time from the encoders. The motion deviation is calculated based on the target joint angles and velocities. The motion deviation is calculated using a PD regulator to obtain the closed-loop stabilizing torque. The feedforward torque, joint impedance compensation torque, and closed-loop stabilizing torque are summed to generate a comprehensive torque control quantity. The comprehensive torque control quantity is discretized and packaged according to the underlying servo control cycle to obtain torque control data packets. Through the real-time industrial communication bus, the torque control data packets are synchronously sent to the end effector servo drives of each physical joint of the bipedal robot. The end effector servo drives the motor actuators of the corresponding joints to output comprehensive torque control quantities, thereby completing the bipedal robot's motion posture and contact force distribution.

[0035] The control hardware architecture of this embodiment includes a host computer (running the WBC algorithm) and multiple distributed end-effector servo drives (mounted to each physical joint). A force-position composite control loop is constructed within each end-effector servo drive, ultimately generating a comprehensive torque control quantity. The formula for calculating the comprehensive torque control quantity is as follows: ; in, This refers to the feedforward joint torque issued by the WBC, i.e., the aforementioned joint torque command. To map the foot contact force to the impedance compensation torque in joint space using the contact Jacobian matrix transpose, and The proportional-differential gain parameters for the position loop and velocity loop of the underlying driver. and For the desired joint angle and velocity; and The actual joint angles and velocities are collected. Due to the nonlinearity of collisions when the bipedal robot contacts the ground, single position control can lead to mechanical breakage, and force control without anchor points can lead to posture collapse. In this embodiment, the composite circuit provides the main explosive force and compliance to cope with strong wind disturbances and inertial impacts. The feedback channel uses a PD regulator for closed-loop stabilization to fine-tune the steady-state error of gear backlash. The calculated comprehensive torque control quantity is packaged into a PDO process data object and sent to the end effector of each joint through the EtherCAT real-time industrial Ethernet bus and DC distributed clock synchronization mode. Then, inside the driver, the magnetic field orientation control algorithm is used to analyze and amplify the current, converting it into three-phase current of the motor, which drives the motor actuator of the corresponding joint to output the comprehensive torque control quantity.

[0036] Example 2, refer to Figure 2 A velocity coordination system with cross-modal reliability constraints for bipedal robots is provided, including a mapping module, an energy verification module, a safety envelope module, and a control constraint module. The mapping module calculates the comprehensive confidence level and time change rate by combining environmental perception data and target motion data, and performs piecewise gain processing on the time change rate to obtain the original admittance parameters. The energy verification module verifies the original admittance parameters based on the virtual energy pool mechanism, obtains the virtual mass and virtual damping coefficient, constructs a shadow centroid model, filters the target motion data, and obtains the first velocity parameter. The safety envelope module acquires the current pose and motor parameters of the bipedal robot, shrinks it through comprehensive confidence to obtain the safety envelope, and verifies the first velocity parameter to obtain the second velocity parameter and continuous boundary conditions. The control constraint module calculates the safety envelope, continuous boundary conditions, and second velocity parameters through constraint coupling and inverse dynamics optimization, thereby obtaining foot contact force and joint torque commands.

[0037] The mapping module uses Mahalanobis distance instead of Euclidean distance and utilizes the jointly calibrated covariance matrix to adaptively address the anisotropic noise of heterogeneous sensors. This overcomes the shortcomings of traditional fixed threshold schemes, such as high false alarm rates and poor environmental adaptability. When the confidence level in either spatial or temporal dimension approaches zero, the overall confidence level instantly drops to zero, constructing a virtual energy pool and monitoring the injected virtual power in real time as admittance parameters change. When passivity is compromised, the module automatically triggers a cutoff fuse protection. This mechanism establishes an energy isolation zone between the information and physical domains to prevent sensor hardware failures. The safety envelope module incorporates aerodynamic drag into the power boundary calculation, transforming ambient wind force from interference into an auxiliary braking resource. In headwinds, additional braking force reduces motor load, and in tailwinds, it anticipates braking difficulties and triggers protection in advance. The reliability-driven dynamic scaling of the safety envelope enables real-time coupling of perceived quality and motion limits. Net inertial residual force is processed in the time domain to extend the motion cycle, reducing peak acceleration to a safe range. Combined with hard jerk cutoff and zero-velocity foot landing constraints, the risk of high-frequency resonance is eliminated. The control constraint module uses dynamic equations and friction cones as constraints to solve for the globally optimal foot contact force and joint torque online. The force-position composite control loop integrates three channels: feedforward fast response, impedance compliant interaction, and feedback stabilization adjustment. Synchronously transmitted via a real-time industrial bus, it ensures precise and synchronized execution of multi-joint torques, effectively avoiding overall machine vibration caused by phase differences.

[0038] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0039] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0040] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A velocity coordination method for cross-modal reliability constraints of a bipedal robot, characterized in that, Includes the following steps: Step S1: Calculate the comprehensive confidence level and time change rate using environmental perception data and target motion data; perform piecewise gain processing on the time change rate to obtain the original admittance parameter. Step S2: Verify the original admittance parameters based on the virtual energy pool mechanism to obtain the virtual mass and virtual damping coefficient, construct the shadow centroid model, filter the target motion data, and obtain the first velocity parameter; Step S3: Obtain the current pose and motor parameters of the bipedal robot, shrink the sensor using the comprehensive confidence level to obtain a safety envelope, and verify the first velocity parameter to obtain the second velocity parameter and continuous boundary conditions. Step S4: Through constraint coupling and inverse dynamics optimization, the safety envelope, continuous boundary conditions and second velocity parameters are calculated to obtain the foot contact force and joint torque command.

2. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11: Obtain environmental perception data and target motion data of the bipedal robot, calculate the Mahalanobis distance between the visual observation position and the radar observation position to obtain the spatial consistency confidence, use the extended Kalman filter to recursively calculate the target motion data to obtain the predicted speed, and compare it with the current measured target speed to obtain the temporal coherence confidence. Step S12: The spatial consistency confidence and temporal coherence confidence are fused through a joint probability model to obtain a normalized comprehensive confidence. The normalized comprehensive confidence is then subjected to temporal difference to extract the smoothed confidence change rate. Step S13: Based on the confidence change rate, construct a piecewise mapping function to map the confidence change rate to the original admittance parameters, which include the original virtual mass and the original virtual damping coefficient.

3. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 2, characterized in that, The piecewise mapping function includes a first gain function and a second gain function, and the processing logic is as follows: When the rate of change of the credibility When the value is 0, the first gain calculation is triggered. The confidence change rate is processed by the first gain function to obtain the corresponding original virtual mass and original virtual damping coefficient. When the rate of change of the credibility When the value is 0, the second gain calculation is triggered. The confidence change rate is calculated using the second gain function to obtain the corresponding original virtual mass and original virtual damping coefficient.

4. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 3, characterized in that, Step S2 includes the following sub-steps: Step S21: Construct a virtual energy pool mechanism. Calculate the transient spurious power based on the current speed, following deviation, and original virtual mass of the bipedal robot. Perform discrete integration on the transient spurious power to obtain the remaining energy state of the virtual energy pool. Step S22: Compare the remaining energy state with a preset safe energy threshold. When the remaining energy state is greater than the safe energy threshold, the energy is determined to be sufficient, and the original admittance parameter is used as the target admittance parameter. The target admittance parameter includes the target virtual mass and the target virtual damping coefficient. When the remaining energy state is less than or equal to the safe energy threshold, it is determined that the passivity has been destroyed, triggering the cutoff mechanism. The original virtual mass is corrected to the safe mass constant as the target virtual mass, and the original virtual damping coefficient is subjected to over-limit dynamic gain as the target virtual damping coefficient. Step S23: Construct a shadow centroid model using the target admittance parameters, and perform differential calculations on the target motion data to obtain the expected centroid velocity, expected centroid acceleration, and expected centroid jerk, which are then used as the first velocity parameter.

5. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 4, characterized in that, Step S3 includes the following sub-steps: Step S31: Obtain the current pose, motor parameters and relative airflow velocity of the bipedal robot; calculate the estimated aerodynamic drag value at the current moment based on the relative airflow velocity; substitute the estimated aerodynamic drag value into the motor power boundary model for correction to obtain the corrected power boundary; and simultaneously fuse the geometric step size boundary to obtain the initial motion boundary. Step S32: Use the comprehensive confidence level as a shrinkage scaling factor to process the initial motion boundary and generate a safe envelope; Step S33: Obtain the contact state switching critical point and high dynamic deceleration range of the continuous phase variable, calculate the inertial impact force of the first velocity parameter, superimpose the inertial impact force and the estimated aerodynamic drag to obtain the net inertial residual force, process the net inertial residual force, and output the second velocity parameter and continuous boundary conditions.

6. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 5, characterized in that, The processing logic for the net inertial residual force, outputting the second velocity parameter and continuous boundary conditions is as follows: Extract the unit direction vector of the current acceleration of the bipedal robot, and perform a dot product operation between the estimated aerodynamic drag value and the unit direction vector to obtain the projection scalar of the aerodynamic drag in the acceleration direction; Subtract the projection scalar from the inertial impact force scalar value of the first velocity parameter to obtain the net inertial residual force; The net inertial residual force is compared with a preset force threshold: When the net inertial residual force is less than or equal to the force threshold, it is determined that the current inertial impact is within the safe range, and the first velocity parameter is output as the second velocity parameter. When the net inertial residual force is greater than the force threshold, it is determined that the current inertial impact exceeds the safe range, and an extension process is performed. The extension process includes: Calculate the difference between the net inertial residual force and the force threshold, and divide the difference by the force threshold to obtain the over-limit ratio coefficient; obtain the reference gait execution cycle of the bipedal robot, and calculate the safe gait execution cycle, wherein the safe gait execution cycle = reference gait execution cycle. Excess ratio coefficient The continuous phase variables are corrected by the safe gait execution cycle to obtain the current action execution rate; Calculate the initial jerk value corresponding to the net inertial residual force, compare the initial jerk value with the preset upper limit value of jerk, and if it exceeds the limit, forcibly truncate and correct the initial jerk value to the upper limit value of jerk. Trajectory reconstruction is performed based on the current action execution rate, and a second velocity parameter and continuous boundary conditions are output. The continuous boundary conditions are used to limit the foot velocity to zero at the critical point of contact state switching.

7. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 6, characterized in that, Step S4 includes the following sub-steps: Step S41: Construct a constrained optimization model based on the second velocity parameter, safety envelope, and continuous boundary conditions; Step S42: Obtain the joint space mass matrix and bias force matrix of the bipedal robot, substitute the joint space mass matrix and bias force matrix into the constraint optimization model, and perform inverse operation to obtain the foot contact force and joint acceleration. Step S43: Based on the foot contact force and joint acceleration, and combined with the actuator selection matrix and contact Jacobian matrix of the bipedal robot, the joint torque command is calculated, and the foot contact force and the joint torque command are sent to the joint actuator of the bipedal robot for execution.

8. The speed coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 7, characterized in that, Step S42 further includes the following sub-steps: Step S421: Obtain the current actual joint data of the bipedal robot. The actual joint data includes joint position and joint velocity. Calculate the joint space mass matrix based on the multibody rigid body dynamics algorithm, calculate the bias force matrix, and solve for the contact Jacobian matrix and the corresponding time derivative matrix corresponding to the contact point of the bipedal robot's foot. Step S422: Using the joint space mass matrix, bias force matrix, and contact Jacobian matrix, construct the dynamic equation of the bipedal robot; using the contact Jacobian matrix and time derivative matrix, construct the motion mapping equation that transforms from the joint acceleration space to the foot end Cartesian space; and input the dynamic equation and the motion mapping equation as constraints into the constraint optimization model. Step S423: Set the foot contact force and joint acceleration as optimization decision variables, start the quadratic programming solver, perform convergence calculation on the constrained optimization model, output the optimal solution vector, and decompose the foot contact force and joint acceleration from the optimal solution vector.

9. The velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in claim 8, characterized in that, The process of sending the foot contact force and the joint torque command to the joint actuator of the bipedal robot is as follows: A force-position composite control loop is constructed inside the underlying servo controller. The force-position composite control loop includes a feedforward channel and a compensation feedback channel. The joint torque command is input into the feedforward channel to obtain the feedforward torque. The foot contact force is mapped into a joint impedance compensation torque by using the contact Jacobian matrix transpose of the bipedal robot and input into the compensation feedback channel. The actual joint angles and actual joint velocities fed back in real time by the encoders of each joint of the bipedal robot are obtained. The motion deviation is calculated based on the target joint angles and target joint velocities. The motion deviation is calculated using a PD regulator to obtain the closed-loop stabilizing torque. The feedforward torque, joint impedance compensation torque, and closed-loop stabilizing torque are summed to generate a comprehensive torque control quantity. The comprehensive torque control quantity is discretized and packaged according to the underlying servo control cycle to obtain a torque control data packet. The torque control data packet is synchronously sent to the end effector of each physical joint of the bipedal robot through a real-time industrial communication bus. The end effector is analyzed and amplified by the current loop inside the end effector to drive the motor actuator of the corresponding joint to output the comprehensive torque control quantity, thereby completing the bipedal robot's motion posture and contact force distribution.

10. A velocity coordination system for cross-modal reliability constraints of a bipedal robot, applied in a velocity coordination method for cross-modal reliability constraints of a bipedal robot as described in any one of claims 1-9, characterized in that, It includes a mapping module, an energy verification module, a safety envelope module, and a control constraint module; The mapping module calculates the comprehensive confidence level and the rate of change of time by using environmental perception data and target motion data, and performs piecewise gain processing on the rate of change of time to obtain the original admittance parameter. The energy verification module verifies the original admittance parameters based on the virtual energy pool mechanism, obtains the virtual mass and virtual damping coefficient, constructs a shadow centroid model, and filters the target motion data to obtain the first velocity parameter. The safety envelope module acquires the current pose and motor parameters of the bipedal robot, shrinks it using the comprehensive confidence level to obtain a safety envelope, and verifies the first velocity parameter to obtain the second velocity parameter and continuous boundary conditions. The control constraint module calculates the safety envelope, continuous boundary conditions, and second velocity parameters through constraint coupling and inverse dynamics optimization to obtain foot contact force and joint torque commands.

Images