A limb coordination planning method and device for a humanoid robot crossing an obstacle

Abstract

The application provides a humanoid robot limb coordination planning method and device when crossing obstacles, comprising: obtaining the humanoid robot trunk side tilt angle, load distribution quantity and holding height as input parameters; calculating the load centroid lateral offset distance according to the load distribution quantity and the trunk side tilt angle, and determining the centroid height linkage quantity in combination with the holding height and the load centroid lateral offset distance; judging the load state according to the centroid height linkage quantity, and calculating the dynamic torque amplification ratio in the high-position holding working condition; adjusting the side tilt angle allowable range according to the dynamic torque amplification ratio, and determining the side tilt amplitude boundary by fusing the load distribution quantity; constructing the constraint condition according to the side tilt amplitude boundary, and generating the limb joint trajectory meeting the constraint by using the inverse kinematics solution; and evaluating the coordination by dynamically simulating the limb joint trajectory, and generating the humanoid robot limb coordination planning scheme according to the evaluation result.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and apparatus for limb coordination planning when a humanoid robot crosses obstacles. Background Technology

[0002] Humanoid robots performing obstacle-crossing tasks in complex terrain environments are a crucial test of their practical application capabilities. Research in this area directly impacts the stability and adaptability of robots in diverse scenarios such as disaster relief, industrial inspection, and domestic services. Among these, the proper coordination of torso posture is particularly important, determining whether the robot can maintain overall balance and achieve smooth movement on uneven surfaces or steps. Especially under uneven load distribution, existing posture control methods often rely on preset fixed compensation rules or simplified models assuming bilateral symmetrical loads. These methods perform reasonably well when both hands are evenly gripping an object, maintaining the center of mass projection within the supporting polygon through slight torso adjustments. However, when the load is concentrated on one arm, the simple torso tilting strategy to the opposite side reveals significant shortcomings. This tilting aims to counteract the lateral shift of the center of mass, but the torso tilting action itself introduces additional torque. This torque originates from the component decomposition of gravity and inertia on the tilted plane, and its magnitude is directly affected by the load gripping height. Gripping height, as a core variable, profoundly alters the effectiveness of tilt compensation. In situations where the load is low, such as when a rescue robot is holding a heavy object like a sandbag low on one side while crossing a muddy ditch, a moderate tilt of the torso to the opposite side is indeed effective. It helps the robot stabilize its leg movement by lowering the center of gravity height and pulling back the center of gravity projection, avoiding the risk of slippage when the support surface shrinks. However, when the load is held high, such as when an industrial inspection robot is holding a heavy sensor high on one side while crossing a factory pipeline obstacle, the same tilt angle causes the torque arm to lengthen dramatically. The long arm leverage effect of the high load amplifies the inertial torque, especially during dynamic steps. The robot's upper body tilt further raises the overall center of gravity, generating a torque opposite to the leg swing direction, causing the center of gravity to quickly deviate from the support range. This compensatory reversal phenomenon is also prominent in home service scenarios. For example, when a robot is holding a kettle high on one side while crossing a living room threshold, the tilt not only fails to correct the deviation but also exacerbates the upper body sway, forcing the stride to deform or directly causing a lateral fall. Especially during dynamic posture adjustments, the reversal of compensation effects due to varying grip heights renders the original fixed tilt strategy completely ineffective, making it difficult for the robot to maintain natural, coordinated gait in real-world tasks with frequently changing load positions. Therefore, identifying the reasonable range of torso tilt angle in real time based on the actual grip height of the load, and dynamically limiting or correcting the tilt amplitude when one side is heavier than the other, to avoid posture instability caused by amplified torque arms, has become a key challenge in limb coordination planning for humanoid robots traversing obstacles. Summary of the Invention

[0003] This invention provides a method for limb coordination planning when a humanoid robot crosses obstacles, mainly including: The humanoid robot's torso tilt angle, load distribution, and grip height are obtained as input parameters. The lateral offset distance of the load center of gravity is calculated based on the load distribution and the torso tilt angle, and the center of gravity height linkage is determined by combining the grip height and the lateral offset distance of the load center of gravity. The load state is determined based on the centroid height linkage amount, and the dynamic torque amplification factor is calculated under the high-position grip condition. Adjust the allowable range of the roll angle according to the dynamic torque amplification factor, and determine the roll amplitude boundary by integrating the load distribution; Based on the stated tilt amplitude boundary, constraints are constructed, and inverse kinematics is used to generate limb joint trajectories that satisfy the constraints. By dynamically simulating and evaluating the coordination of the limb joint trajectories, a limb coordination planning scheme for the humanoid robot is generated based on the evaluation results.

[0004] Furthermore, the smoothed humanoid robot torso tilt angle, load distribution, and filtered grip height are obtained as input parameters, including: The raw data of the humanoid robot's torso tilt angle were collected, the raw values ​​of the load mass on the left and right sides were collected, and the original values ​​of the grip height were obtained from the encoders of the shoulder and elbow joints and converted by positive kinematics. The original data of the humanoid robot's torso tilt angle, the original values ​​of the load mass on the left and right sides, and the original value of the gripping height are subjected to low-pass filtering to obtain the smoothed torso tilt angle, the filtered left and right load mass values, and the filtered gripping height. The load distribution is calculated based on the filtered load mass values ​​on the left and right sides, serving as a parameter characterizing the degree of load asymmetry. The smoothed humanoid robot torso tilt angle, the load distribution, and the filtered grip height are used as input parameters.

[0005] Furthermore, the step of calculating the lateral offset distance of the load center of gravity based on the load distribution and the torso tilt angle, and determining the center of gravity height linkage amount in conjunction with the grip height and the lateral offset distance of the load center of gravity, includes: The coordinates of the lateral application point of the resultant load force are determined based on the load distribution and the lateral coordinates of the wrist end position. The lateral spacing of the load is calculated based on the coordinates of the lateral action point and the position of the spinal axis. Geometric projection is then performed using the tilt angle of the humanoid robot's torso to obtain the lateral offset distance of the load's center of mass. The vertical deviation is calculated based on the difference between the grip height and the preset standard torso height, and the center of gravity height linkage is calculated based on the vertical deviation and the lateral offset distance.

[0006] Furthermore, the step of determining the load state based on the centroid height linkage and calculating the dynamic torque amplification factor under high-position grip conditions includes: The center of gravity height linkage amount is compared with a preset threshold. If the center of gravity height linkage amount exceeds the threshold, it is determined to be a high-position grip condition. For high-grip working conditions, the vertical distance is calculated based on the grip height and the preset lumbar joint reference position, and the torque arm length is calculated by combining the torso tilt angle and the vertical distance. The dynamic torque amplification factor is determined based on the ratio of the torque arm length to the standard reference value.

[0007] Furthermore, the step of adjusting the allowable range of the roll angle based on the dynamic torque amplification factor and determining the roll amplitude boundary by integrating the load distribution includes: The preset upper and lower limits of the roll angle are adjusted according to the reciprocal of the dynamic torque amplification factor to obtain the adjusted allowable upper and lower limits of the roll angle. The offset correction amount is calculated based on the load distribution, and the upper and lower limits of the allowable roll angle are asymmetrically corrected. The offset correction amount is then reduced towards the side with heavier load to obtain the roll amplitude boundary.

[0008] Furthermore, the step of constructing constraint conditions based on the tilt amplitude boundary and generating limb joint trajectories that satisfy the constraints using inverse kinematics includes: Based on the tilt amplitude boundary, construct the inequality constraint condition for the torso posture, and take the upper and lower limits of the boundary as the maximum and minimum allowable values ​​of the tilt angle. A Jacobi matrix of the lower limb kinematic chain is established. The gradient projection method is used to solve the inverse kinematics of the Jacobi matrix. In each iteration cycle, the joint angle correction is obtained by pseudo-inverse operation of the Jacobi matrix and projected into the feasible constraint interval to obtain the hip joint compensation angle, knee joint compensation angle and ankle joint compensation angle. The limb joint trajectories are formed by arranging them in the time sequence of the obstacle crossing action.

[0009] Furthermore, the step of dynamically simulating and evaluating the coordination of the limb joint trajectories, and generating a limb coordination planning scheme for the humanoid robot based on the evaluation results, includes: Based on the limb joint trajectory, kinematic simulation is performed to calculate the trunk tilt angle and the center of mass projection position, and the minimum distance from the center of mass projection point to the edge of the support area is extracted as the stability margin value. The system detects whether the real-time joint angular velocity exceeds the preset joint velocity upper limit. If the stability margin value is lower than the preset stability margin threshold or the preset joint angular velocity exceeds the joint velocity upper limit, it is determined that the coordination requirements are not met. If the stability margin value is higher than or equal to the stability margin threshold and the joint angular velocity does not exceed the joint velocity upper limit, then the coordination requirement is determined to be met. If the coordination requirements are not met, the torque arm length parameter is adjusted, the dynamic torque amplification factor and the tilt amplitude boundary are updated, and the limb joint trajectory is regenerated. If the coordination requirements are met, the limb joint trajectory is encapsulated into a joint instruction sequence to generate a coordination planning scheme.

[0010] This invention provides a limb coordination planning device for a humanoid robot when crossing obstacles, mainly comprising: The data acquisition module is used to acquire the humanoid robot's torso tilt angle, load distribution, and grip height as input parameters. The load center of gravity calculation module is used to calculate the lateral offset distance of the load center of gravity based on the load distribution and the torso tilt angle, and to determine the center of gravity height linkage amount by combining the grip height and the lateral offset distance of the load center of gravity. The working condition judgment module is used to judge the load state based on the center of gravity height linkage amount and calculate the dynamic torque amplification factor under the high-position grip working condition. The roll boundary determination module is used to adjust the allowable range of the roll angle according to the dynamic torque amplification factor and determine the roll amplitude boundary by integrating the load distribution. The trajectory generation module is used to construct constraints based on the tilt amplitude boundary and generate limb joint trajectories that satisfy the constraints using inverse kinematics. The coordination planning module is used to evaluate the coordination by dynamically simulating the trajectory of the limb joints, and generate a limb coordination planning scheme for the humanoid robot based on the evaluation results.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method and apparatus for limb coordination planning in humanoid robots crossing obstacles. It proposes a systematic solution to the comprehensive impact of load distribution, torso tilt angle, and grip height on robot stability and coordination in obstacle-crossing scenarios. The invention obtains the torso tilt angle, load distribution, and grip height; calculates the lateral offset distance of the load center of mass and the linkage amount of the center of mass height; accurately determines the load state; calculates the dynamic torque amplification factor under high-grip conditions; dynamically adjusts the allowable range of tilt angle; integrates the load distribution to construct tilt amplitude boundary constraints; uses inverse kinematics to generate limb joint trajectories that satisfy the constraints; and finally evaluates coordination through dynamic simulation to generate the optimal limb coordination planning scheme. The core innovation of this invention lies in combining load state with tilt constraints, ensuring the stability and coordination of the robot's limbs when crossing obstacles, significantly improving adaptability and motion smoothness in complex environments, and providing technical assurance for the safe operation of humanoid robots under dynamic loads. Attached Figure Description

[0012] Figure 1 This is a flowchart of a limb coordination planning method for a humanoid robot crossing obstacles according to the present invention.

[0013] Figure 2 This is a schematic diagram of the structure of a limb coordination planning device for a humanoid robot crossing obstacles according to the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0015] like Figures 1-2 This embodiment of a method and apparatus for limb coordination planning when a humanoid robot crosses obstacles may specifically include: Step S101: Obtain the humanoid robot's torso tilt angle, load distribution, and grip height as input parameters.

[0016] Raw data of the torso's tilt angle relative to the direction of gravity is collected by an inertial measurement unit installed on the torso of the humanoid robot. Raw values ​​of the left and right load masses are read from six-dimensional force sensors at the wrists of both arms. The joint angles of the shoulder and elbow joints are obtained from encoders and converted using forward kinematics to obtain the raw height of the gripping point in the torso coordinate system. The raw tilt angle data, the raw values ​​of the left and right load masses, and the raw height are input into a low-pass Butterworth filter to remove high-frequency noise components, resulting in smoothed torso tilt angles, filtered left and right load masses, and filtered gripping heights. The absolute value of the difference between the filtered left and right load masses is calculated and divided by the sum of the two values ​​to obtain a load distribution value characterizing the degree of load asymmetry between the left and right sides. The filtered gripping height, the smoothed torso tilt angle, and the load distribution value are used as input parameters for subsequent torso posture coordination.

[0017] In one embodiment, an inertial measurement unit installed in the chest cavity of the humanoid robot's torso includes a three-axis accelerometer and a three-axis gyroscope. By measuring the components of gravitational acceleration along each axis of the torso coordinate system, it calculates the raw data of the torso's tilt angle relative to the world coordinate system in the vertical direction. This tilt angle reflects the degree of tilt of the torso in the coronal plane.

[0018] Specifically, the six-dimensional force sensors are installed at the left and right wrist joints respectively to measure the three-dimensional force and three-dimensional torque borne by each wrist in real time. The vertical force component is extracted and divided by the gravitational acceleration value to obtain the original values ​​of the load mass on the left and right sides. The two values ​​represent the mass of the object currently held by the left and right arms respectively.

[0019] It should be noted that the acquisition of the gripping point height depends on the forward kinematics transformation process. The current rotation angle value of each joint is read from the rotary encoders of the shoulder and elbow joints. Based on the link length parameters and joint axis directions of the humanoid robot arm, the homogeneous transformation matrix between the coordinate systems of each joint is calculated sequentially according to the De Navet-Hatenberg parametric method. After multiplying the transformation matrices together, the component of the end position vector in the vertical direction of the torso coordinate system is extracted, which is the original value of the gripping point height.

[0020] In one possible implementation, the low-pass Butterworth filter adopts a second-order structure, and its cutoff frequency is set according to the walking gait period of the humanoid robot. It filters out high-frequency interference components introduced by mechanical vibration and electrical noise during the sensor sampling process, and outputs the smoothed torso tilt angle, the filtered left load mass value, the filtered right load mass value, and the filtered grip height.

[0021] For example, the load distribution is calculated by taking the filtered left load mass value and the filtered right load mass value as inputs, taking the absolute value of the difference between the two and dividing it by the sum of the two. The resulting ratio is between zero and one. The closer the value is to one, the more concentrated the load is on one arm. The closer the value is to zero, the more balanced the load is on both sides. This load distribution, together with the filtered grip height and the smoothed torso tilt angle, constitutes the input parameters for subsequent torso posture coordination.

[0022] Step S102: Calculate the lateral offset distance of the load center of gravity based on the load distribution amount and the torso tilt angle, and determine the center of gravity height linkage amount by combining the grip height and the lateral offset distance of the load center of gravity.

[0023] Based on the filtered left and right load mass values, lateral coordinate components are extracted from the wrist distal positions obtained through forward kinematics conversion. A weighted average is calculated using the left and right load mass values ​​as weights for the corresponding wrist lateral coordinates to obtain the lateral coordinates of the resultant load force. The horizontal distance between the lateral coordinates of the resultant load force and the lateral zero point of the humanoid robot's spinal axis is calculated as the lateral load spacing. The smoothed torso tilt angle is then used to geometrically project the lateral load spacing. Multiplying the cosine of the torso tilt angle by the lateral load spacing yields the lateral offset distance of the load's center of mass relative to the spinal axis. The vertical deviation of the grip point is obtained by subtracting the standard torso height (from the hip joint center to the shoulder joint center) from the filtered grip height. Multiplying this vertical deviation by the lateral offset distance and dividing by the standard torso height yields the change in center of mass height caused by the grip height.

[0024] In one embodiment, the lateral coordinates of the resultant load force represent the equivalent position of the load held by both arms in the lateral direction of the torso coordinate system. This coordinate is not a simple arithmetic average of the left and right wrist positions, but a weighted calculation with the load mass on each side as the weight, so that the lateral coordinate of the wrist on the side with the heavier load contributes more to the final coordinates of the point of application.

[0025] Specifically, when a humanoid robot holds a larger object with its left hand and a smaller object with its right hand, the coordinates of the lateral point of application of the resultant load force are biased towards the left wrist. The degree of offset is related to the ratio of the mass of the load on both sides. This weighting method is consistent with the calculation principle of the point of application of the resultant force in mechanics.

[0026] It should be noted that the humanoid robot's spinal axis is located at the horizontal zero point in the torso coordinate system. As a reference for judging the direction and amount of load offset, the horizontal distance between the coordinates of the horizontal point of application of the resultant load force and this zero point is the horizontal distance of the load. When this distance is positive, it means that the load is biased to the right side of the torso, and when it is negative, it means that the load is biased to the left side of the torso.

[0027] In one possible implementation, the geometric projection process takes into account the effect of trunk tilt on the lateral offset distance. When the trunk is in a vertical posture, the tilt angle is zero, and the load lateral spacing is equal to the lateral offset distance. When the trunk tilts to one side, the actual lateral offset distance of the load relative to the spinal axis is the load lateral spacing multiplied by the cosine of the tilt angle. This projection relationship reflects the effective lateral component in the tilt plane.

[0028] For example, the center of mass height linkage describes the coupled effect of the grip height change on the overall center of mass position. The calculation is based on the fact that the higher the grip point is than the standard torso height, the more significant the change in center of mass position caused by the lateral offset. The product of the vertical deviation and the lateral offset distance divided by the standard torso height characterizes the amplification or reduction effect of the height factor on the lateral migration of the center of mass. This linkage is particularly critical when a humanoid robot grips a heavy object on one side at a high position to cross obstacles.

[0029] Step S103: Determine the load state based on the centroid height linkage amount, and calculate the dynamic torque amplification factor under the high-position grip condition.

[0030] The center of gravity height linkage amount is compared with a preset height linkage threshold. If the center of gravity height linkage amount exceeds the height linkage threshold, the current load state is determined to be a high-position grip condition; if the center of gravity height linkage amount does not exceed the height linkage threshold, the current load state is determined to be a low-position grip condition. For the low-position grip condition, the lateral offset distance is directly output as the center of gravity lateral offset compensation amount. For the high-position grip condition, based on the filtered grip height and the preset vertical reference position of the lumbar joint, the vertical distance from the grip point to the lumbar joint is calculated. The vertical distance is multiplied by the sine of the filtered trunk tilt angle to obtain the torque arm length of the load force on the lumbar joint. The dynamic torque amplification factor is obtained based on the ratio of the torque arm length to the preset upright posture standard torque arm reference value.

[0031] In one embodiment, the height linkage threshold is set based on the relationship between the humanoid robot's torso structure and load stability. When the gripping point is below the shoulder joint and close to the waist height, the center of mass height linkage is usually small. At this time, the load's influence on the torso posture is mainly reflected in the lateral displacement, which is a low-position gripping condition.

[0032] Specifically, in the low-position gripping condition, the height of the object held by the humanoid robot's single arm is close to the level of the hip joint. The torque arm generated by the load is relatively short. The torso can be tilted to the opposite side to realize the centroid projection returning to the inside of the supporting polygon. At this time, the lateral offset distance is directly used as the compensation reference, and there is no need to consider the torque amplification effect.

[0033] It should be noted that when the center of gravity height linkage exceeds the preset threshold, it indicates that the grip point has risen to the vicinity of the shoulder joint or higher, and the torque effect of the load on the lumbar spine joint is significantly enhanced. At this time, the process of high grip condition is entered. As the key hinge point connecting the torso and lower limbs, the lumbar spine joint bears all the gravitational torque from the upper body and the load.

[0034] In one possible implementation, the torque arm length is calculated based on the lever principle. The vertical distance from the grip point to the lumbar joint represents the vertical distance between the line of action of the force and the joint axis of rotation. When the torso remains upright, this vertical distance is the torque arm length. When the torso tilts to the side, the torque arm length changes with the tilt angle. Specifically, it is the vertical distance multiplied by the sine of the tilt angle. The larger the tilt angle, the longer the projection component of the torque arm in the tilt direction, and the greater the lateral overturning moment generated by the load on the lumbar joint. This geometric relationship explains the mechanical mechanism that the torso tilting in the high grip position exacerbates the risk of instability.

[0035] For example, the reference value of the standard torque arm in the upright posture corresponds to the reference distance from the standard grip height to the lumbar joint of the humanoid robot in an unloaded and vertical state. This reference value is used as the normalized denominator, making the dynamic torque amplification factor a dimensionless proportional coefficient. When the factor is greater than one, it indicates that the torque effect in the current high grip state exceeds the standard working condition. The larger the factor value, the more obvious the torque amplification effect during torso posture adjustment.

[0036] Understandably, the introduction of dynamic torque amplification factor enables humanoid robots to quantitatively assess the additional instability risk brought about by high loads when crossing obstacles, thereby providing a basis for the dynamic limitation of torso tilt angle.

[0037] Step S104: Adjust the allowable range of the roll angle according to the dynamic torque amplification factor, and determine the roll amplitude boundary by integrating the load distribution.

[0038] The preset upper and lower limits of the torso tilt angle are adjusted according to the dynamic torque amplification factor. The reciprocal of the dynamic torque amplification factor is multiplied by the upper and lower limits of the torso tilt angle, respectively, to obtain the gain-adjusted upper and lower limits of the allowable tilt angle. Asymmetric corrections are then applied to the upper and lower limits of the allowable tilt angle based on the load distribution. Half the difference between the upper and lower limits of the allowable tilt angle is multiplied by the load distribution as an offset correction amount. This offset correction amount is then applied to the load-biased side of both the upper and lower limits of the allowable tilt angle, resulting in the tilt amplitude boundaries.

[0039] In one embodiment, the upper limit and lower limit of the torso tilt angle reference are preset angle values ​​based on the physical limits of the humanoid robot's torso structure and the range of motion of its joints. The upper limit corresponds to the maximum allowable angle of the torso tilting to the right, and the lower limit corresponds to the maximum allowable angle of the torso tilting to the left.

[0040] Specifically, when the dynamic torque amplification factor is greater than one, its reciprocal is less than one. After multiplying by the upper and lower limits of the reference, the allowable upper and lower limits of the tilt angle both shrink towards zero. This means that the allowable range of trunk tilt in the high grip state is compressed, and the larger the dynamic torque amplification factor, the narrower the allowable range.

[0041] It should be noted that the asymmetric correction process makes differentiated adjustments to the roll boundary based on the load distribution. When the load is biased to the right, the allowable upper limit of the roll angle shrinkage is greater than the allowable lower limit shrinkage, and vice versa. The magnitude of the offset correction is determined by the product of the load distribution and the allowable range width. This asymmetric shrinkage causes the roll amplitude boundary to be biased to the side opposite to the load bias direction.

[0042] Understandably, the tilt amplitude boundary, as a rigid constraint on torso posture adjustment, limits the effective range of torso tilt angle for humanoid robots during obstacle crossing.

[0043] Step S105: Construct constraint conditions based on the tilt amplitude boundary, and use inverse kinematics to generate limb joint trajectories that satisfy the constraints.

[0044] Based on the tilt amplitude boundary, inequality constraints for the torso posture are constructed. The upper and lower limits of the tilt amplitude boundary are used as the maximum and minimum allowable values ​​for the torso tilt angle in the inverse kinematics solution process, respectively, forming a feasible constraint interval for the joint angles. A Jacobian matrix for the humanoid robot's lower limb kinematics chain is established. This matrix describes the mapping relationship between the angular velocities of the hip, knee, and ankle joints and the positional velocity of the end-foot. Positional velocity transfer relationships from the hip joint through the knee joint to the ankle joint are constructed sequentially according to the rotation axis direction and link length, resulting in the lower limb Jacobian matrix. The gradient projection method is used to perform inverse kinematics solution on the lower limb Jacobian matrix. In each iteration cycle, the joint angle correction is obtained through pseudo-inverse operation of the Jacobian matrix. The joint angle correction is projected onto the feasible constraint interval. If the corrected joint angle exceeds the tilt amplitude boundary, it is restricted to the boundary value, resulting in the hip joint compensation angle, knee joint compensation angle, and ankle joint compensation angle. The hip joint compensation angle, the knee joint compensation angle, and the ankle joint compensation angle are arranged in the time sequence of the obstacle crossing action to obtain the limb joint trajectory that satisfies the lateral tilt amplitude boundary constraint.

[0045] In one embodiment, the tilt amplitude boundary is introduced into the inverse kinematics solution process as a hard constraint for trunk posture adjustment. The upper and lower limits of this boundary define the maximum allowable angles of the trunk tilting to the left and right in the coronal plane, respectively. During the inverse kinematics solution process, the trunk tilt angle caused by any combination of joint angles must not exceed this range.

[0046] Specifically, the establishment of feasible constraint intervals is based on the kinematic coupling relationship between the humanoid robot's torso and lower limbs. The change in the adduction and abduction angle of the hip joint of the lower limb directly affects the degree of lateral tilt of the torso relative to the supporting leg. The flexion and extension angles of the knee and ankle joints indirectly affect the torso posture by changing the effective length of the supporting leg. The combination of the angles of the three joints together determines the lateral tilt state of the torso in space.

[0047] It should be noted that the Jacobian matrix of the lower limb is a core tool for describing the velocity mapping relationship between joint space and Cartesian space. For the lower limb of a humanoid robot, each column of this matrix corresponds to the cross product of the rotation axis direction of a joint and the position vector from that joint to the end foot. The number of rows in the matrix corresponds to the number of degrees of freedom of the end foot in Cartesian space, and the number of columns corresponds to the number of joints involved in the motion. When the hip joint rotates around its rotation axis, the end of the foot will generate a corresponding linear velocity. The direction and magnitude of this linear velocity are determined by the rotation axis direction of the hip joint, the link length, and the current joint angle. The contributions of the knee and ankle joints are superimposed in a similar way to form the complete Jacobian matrix of the lower limb.

[0048] In one possible implementation, the Jacobian matrix is ​​constructed starting from the hip joint and extending downwards along the kinetic chain. The rotational axes of the hip joint typically include flexion-extension, adduction-abduction, and internal / external rotation axes. Among these, the adduction-abduction axis has the most direct effect on trunk tilt. The knee joint mainly provides flexion-extension degrees of freedom, while the ankle joint includes plantar flexion-dorsiflexion and inversion-eversion degrees of freedom. The spatial orientation of each joint's rotational axis in the current posture, together with the length parameters of each link, constitutes the element values ​​of the Jacobian matrix.

[0049] For example, the gradient projection method is an iterative solution method for constrained optimization problems. Its core idea is that the search direction generated in each iteration is projected into the feasible region defined by the constraints. In the inverse kinematics solution, the objective function is the error vector between the current position and the desired position of the end-foot. The gradient direction corresponds to the joint angle adjustment direction where the error decreases the fastest. When the unconstrained gradient direction causes the joint angle to exceed the feasible constraint interval, the gradient projection method projects this direction onto the boundary tangent plane of the feasible region, so that the adjusted joint angle always meets the limit of the tilt amplitude boundary. In the specific implementation process, the joint angle correction amount is first calculated by the pseudo-inverse operation of the Jacobian matrix. The pseudo-inverse operation gives the joint angle change amount that minimizes the end-foot position error in the least squares sense. Then, it is checked whether the correction amount superimposed on the current joint angle will cause the trunk tilt angle to exceed the feasible constraint interval. If it does not exceed the limit, the correction amount is directly used. If it exceeds the limit, the correction amount is truncated along the constraint boundary. The truncated correction amount ensures that the joint angle is exactly at the boundary value and does not exceed the limit.

[0050] Understandably, the projection processing of constraint boundaries essentially involves finding a compromise direction in joint space that reduces end-effector position error without violating tilt constraints. When a humanoid robot experiences high-level unilateral loads while traversing obstacles, the tilt amplitude boundary is dynamically compressed. In this case, the constraint processing mechanism of the gradient projection method will trigger boundary truncation operations more frequently, forcing the lower limb joints to adopt a more conservative angle adjustment scheme. Furthermore, the hip joint compensation angle, knee joint compensation angle, and ankle joint compensation angle are the joint angle increments obtained after constraint projection processing within a single iteration cycle. The accumulated compensation angles across multiple iteration cycles form the evolution trajectory of the joint angles, which describes the complete joint motion process from the initial posture to the final posture of the traversal action.

[0051] Preferably, the temporal arrangement of the limb joint trajectories follows the phase division of the obstacle crossing action, including the single-leg standing phase of the supporting leg, the swinging leg crossing phase, and the double-leg transition phase. The compensation angle sequence in each phase has different variation characteristics. The single-leg standing phase of the supporting leg is most sensitive to the lateral tilt constraint. At this time, the planning of the limb joint trajectory directly determines whether the humanoid robot can maintain overall balance when the support surface is reduced.

[0052] Step S106: The coordination is evaluated by dynamically simulating the trajectory of the limb joints, and a limb coordination planning scheme for the humanoid robot is generated based on the evaluation results.

[0053] Based on the limb joint trajectories, a kinematic simulation is performed on the dynamic process of the humanoid robot traversing obstacles. The torso tilt angle and center of mass projection position are calculated frame by frame. The minimum distance from the center of mass projection point to the boundary of the convex hull formed by the contact point of the supporting foot is extracted as the stability margin value. Simultaneously, it is detected whether the angular velocity of each joint exceeds the preset joint velocity upper limit of 2.0 radians per second, resulting in a trajectory coordination evaluation result. If the stability margin value in the trajectory coordination evaluation result is lower than the preset stability margin threshold of 0.05 meters, or the joint angular velocity exceeds the joint velocity upper limit, the coordination requirement is determined not to be met. The process then backtracks to the vertical distance estimation stage, increasing the estimated vertical distance value by a preset adjustment step of 0.01 meters. The dynamic torque amplification factor and tilt amplitude boundary are recalculated, and inverse kinematics is performed again to obtain the updated limb joint trajectory, which is then returned to the evaluation. If the stability margin value is higher than or equal to the stability margin threshold and the joint angular velocity does not exceed the joint velocity upper limit, the coordination requirement is determined to be met. The hip joint compensation angle, knee joint compensation angle, and ankle joint compensation angle in the limb joint trajectory are encapsulated into a joint command sequence in chronological order. By integrating the dynamic change curve of the trunk tilt angle with the time series of the joint angles of each lower limb based on the joint command sequence, a limb coordination planning scheme for the humanoid robot to cross obstacles is obtained.

[0054] In one embodiment, the kinematic simulation process uses a discrete-time stepping method to solve the trajectory of the limb joints frame by frame. Each frame corresponds to a time sampling point in the action of crossing the obstacle. At this time point, the spatial position of the humanoid robot's overall center of mass is calculated based on the current angle values ​​of the hip joint, knee joint and ankle joint, and the center of mass is projected onto the horizontal ground along the direction of gravity to obtain the coordinates of the center of mass projection point.

[0055] Specifically, the support area is formed by the outer contour of the support foot currently in contact with the ground. When the humanoid robot is in a single-leg support phase, the support area degenerates into the contact area of ​​a single foot. When it is in a two-leg support phase, the support area expands into the convex hull of the contact area of ​​both feet. The stability margin is defined as the shortest Euclidean distance from the centroid projection point to the edge of the support area. The larger this distance is, the higher the static stability of the current posture.

[0056] It should be noted that the detection of joint angular velocity is carried out synchronously in the kinematic simulation. The angular velocity required for each joint to change from the current frame angle value to the next frame angle value is obtained by dividing the angle increment by the inter-frame time interval. The preset upper limit of joint velocity reflects the physical response capability of the joint servo driver. Exceeding this upper limit means that the joint cannot complete the expected angle change within the specified time.

[0057] In one possible implementation, the trajectory coordination evaluation result contains two types of information: the minimum stability margin value of each frame throughout the entire span cycle and the maximum joint angular velocity of each frame. When the minimum stability margin value is lower than the stability margin threshold, it indicates that there is a risk frame where the centroid projection is close to or exceeds the edge of the support area. When the maximum joint angular velocity exceeds the upper limit of the joint velocity, it indicates that there is a risk frame where the joint movement is too fast. If either condition is triggered, it is determined that the coordination requirements are not met.

[0058] For example, the core of the backtracking adjustment mechanism is to indirectly affect the subsequent calculation of the tilt amplitude boundary by modifying the estimated value of the torque arm length. When the trajectory does not meet the coordination requirements, it means that the current tilt constraint is too loose, resulting in excessive centroid shift or excessive joint movement. At this time, increasing the estimated value of the vertical distance with the preset adjustment step size is equivalent to artificially amplifying the influence weight of the high-position grip effect. The amplified dynamic torque magnification factor makes the tilt amplitude boundary further narrow. The narrowed boundary forces the joint angle adjustment to be more conservative in the next round of inverse kinematics solution, thereby producing a smoother limb joint trajectory.

[0059] Understandably, the backtracking adjustment forms a closed-loop iterative process. Each iteration includes three stages: boundary update, inverse kinematics solution, and simulation evaluation. The iteration terminates when the trajectory coordination evaluation result meets the coordination requirements or the number of iterations reaches a preset upper limit. In the scenario of a humanoid robot crossing obstacles in a factory pipeline, this iterative process typically converges within a few rounds. The converged limb joint trajectory balances the completion of the crossing action with posture stability. Furthermore, when the trajectory coordination evaluation result meets the coordination requirements, the hip joint compensation angle, knee joint compensation angle, and ankle joint compensation angle are extracted and arranged according to the time sampling order of the crossing action to form a joint command sequence. This sequence records the target angle values ​​of each joint at fixed time intervals and can be directly used to drive the joint servo controller.

[0060] Preferably, the final integration of the limb coordination planning scheme will synchronize the joint command sequence with the dynamic change curve of the trunk tilt angle in time. The trunk tilt curve describes the evolution trajectory of the tilt angle of the trunk relative to the vertical direction over time during the crossing cycle. The combination of the two forms a complete whole-body posture description, enabling the humanoid robot to coordinate the upper and lower limb movements when performing crossing actions, and maintain overall balance even under unilateral load conditions.

[0061] This invention provides a limb coordination planning device for a humanoid robot when crossing obstacles, mainly comprising: The data acquisition module is used to acquire the humanoid robot's torso tilt angle, load distribution, and grip height as input parameters. The load center of gravity calculation module is used to calculate the lateral offset distance of the load center of gravity based on the load distribution and the torso tilt angle, and to determine the center of gravity height linkage amount by combining the grip height and the lateral offset distance of the load center of gravity. The working condition judgment module is used to judge the load state based on the center of gravity height linkage amount and calculate the dynamic torque amplification factor under the high-position grip working condition. The roll boundary determination module is used to adjust the allowable range of the roll angle according to the dynamic torque amplification factor and determine the roll amplitude boundary by integrating the load distribution. The trajectory generation module is used to construct constraints based on the tilt amplitude boundary and generate limb joint trajectories that satisfy the constraints using inverse kinematics. The coordination planning module is used to evaluate the coordination by dynamically simulating the trajectory of the limb joints, and generate a limb coordination planning scheme for the humanoid robot based on the evaluation results.

[0062] If the technical solution of this application involves personal information, the product using this solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If sensitive personal information is involved, the user's separate consent has been obtained before processing, and the "express consent" requirement is met. For example, a clear sign is placed at the collection device such as a camera to inform the user that they have entered the collection area, and the user's voluntary entry is considered as consent; or the processing device clearly indicates the processing rules and obtains authorization through pop-up windows or by asking the user to upload information themselves. The personal information processing rules include the processor, the purpose of processing, the processing method, and the types of personal information.

[0063] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for limb coordination planning when a humanoid robot crosses obstacles, characterized in that, include: The humanoid robot's torso tilt angle, load distribution, and grip height are obtained as input parameters. The lateral offset distance of the load center of gravity is calculated based on the load distribution and the torso tilt angle, and the center of gravity height linkage is determined by combining the grip height and the lateral offset distance of the load center of gravity. The load state is determined based on the centroid height linkage amount, and the dynamic torque amplification factor is calculated when the gripping condition is in a high position. Adjust the allowable range of the roll angle according to the dynamic torque amplification factor, and determine the roll amplitude boundary by integrating the load distribution; Based on the stated tilt amplitude boundary, constraints are constructed, and inverse kinematics is used to generate limb joint trajectories that satisfy the constraints. By dynamically simulating and evaluating the coordination of the limb joint trajectories, a limb coordination planning scheme for the humanoid robot is generated based on the evaluation results.

2. The limb coordination planning method for a humanoid robot crossing obstacles as described in claim 1, characterized in that, The process of obtaining the smoothed humanoid robot torso tilt angle, load distribution, and filtered grip height as input parameters includes: The raw data of the humanoid robot's torso tilt angle were collected, the raw values ​​of the load mass on the left and right sides were collected, and the original values ​​of the grip height were obtained from the encoders of the shoulder and elbow joints and converted by positive kinematics. The original data of the humanoid robot's torso tilt angle, the original values ​​of the load mass on the left and right sides, and the original value of the grip height are subjected to low-pass filtering to obtain the smoothed torso tilt angle, the filtered values ​​of the load mass on the left and right sides, and the filtered grip height. The load distribution is calculated based on the filtered load mass values ​​on the left and right sides, serving as a parameter characterizing the degree of load asymmetry. The smoothed humanoid robot torso tilt angle, the load distribution, and the filtered grip height are used as input parameters.

3. The limb coordination planning method for a humanoid robot crossing obstacles as described in claim 1, characterized in that, The step of calculating the lateral offset distance of the load center of gravity based on the load distribution and the torso tilt angle, and determining the center of gravity height linkage amount by combining the grip height and the lateral offset distance of the load center of gravity, includes: The coordinates of the lateral application point of the resultant load force are determined based on the load distribution and the lateral coordinates of the wrist end position. The lateral spacing of the load is calculated based on the coordinates of the lateral action point and the position of the spinal axis. Geometric projection is then performed using the tilt angle of the humanoid robot's torso to obtain the lateral offset distance of the load's center of mass. The vertical deviation is calculated based on the difference between the grip height and the preset standard torso height, and the center of gravity height linkage is calculated based on the vertical deviation and the lateral offset distance.

4. The limb coordination planning method for a humanoid robot crossing obstacles as described in claim 1, characterized in that, The step of determining the load state based on the centroid height linkage and calculating the dynamic torque amplification factor under high-position grip conditions includes: The center of gravity height linkage amount is compared with a preset threshold. If the center of gravity height linkage amount exceeds the threshold, it is determined to be a high-position grip condition. For high-grip working conditions, the vertical distance is calculated based on the grip height and the preset lumbar joint reference position, and the torque arm length is calculated by combining the torso tilt angle and the vertical distance. The dynamic torque amplification factor is determined based on the ratio of the torque arm length to the standard reference value.

5. The limb coordination planning method for a humanoid robot crossing obstacles as described in claim 1, characterized in that, The step of adjusting the allowable range of the roll angle based on the dynamic torque amplification factor and determining the roll amplitude boundary by integrating the load distribution includes: The preset upper and lower limits of the roll angle are adjusted according to the reciprocal of the dynamic torque amplification factor to obtain the adjusted allowable upper and lower limits of the roll angle. The offset correction amount is calculated based on the load distribution, and the upper and lower limits of the allowable roll angle are asymmetrically corrected. The offset correction amount is then reduced towards the side with heavier load to obtain the roll amplitude boundary.

6. The limb coordination planning method for a humanoid robot crossing obstacles as described in claim 1, characterized in that, The step of constructing constraints based on the tilt amplitude boundary and generating limb joint trajectories that satisfy the constraints using inverse kinematics includes: Based on the tilt amplitude boundary, construct the inequality constraint condition for the torso posture, and take the upper and lower limits of the boundary as the maximum and minimum allowable values ​​of the tilt angle. A Jacobi matrix of the lower limb kinematic chain is established. The gradient projection method is used to solve the inverse kinematics of the Jacobi matrix. In each iteration cycle, the joint angle correction is obtained by pseudo-inverse operation of the Jacobi matrix and projected into the feasible constraint interval to obtain the hip joint compensation angle, knee joint compensation angle and ankle joint compensation angle. The limb joint trajectories are formed by arranging them in the time sequence of the obstacle crossing action.

7. The limb coordination planning method for a humanoid robot crossing obstacles as described in claim 1, characterized in that, The process of dynamically simulating and evaluating the coordination of the limb joint trajectories, and generating a limb coordination planning scheme for the humanoid robot based on the evaluation results, includes: Based on the limb joint trajectory, kinematic simulation is performed to calculate the trunk tilt angle and the center of mass projection position, and the minimum distance from the center of mass projection point to the edge of the support area is extracted as the stability margin value. The system detects whether the real-time joint angular velocity exceeds the preset upper limit of the joint velocity. If the stability margin value is lower than the preset stability margin threshold or the preset joint angular velocity exceeds the upper limit of the joint velocity, it is determined that the coordination requirements are not met. If the stability margin value is higher than or equal to the stability margin threshold and the joint angular velocity does not exceed the joint velocity upper limit, then the coordination requirement is determined to be met. If the coordination requirements are not met, the torque arm length parameter is adjusted, the dynamic torque amplification factor and the tilt amplitude boundary are updated, and the limb joint trajectory is regenerated. If the coordination requirements are met, the limb joint trajectory is encapsulated into a joint instruction sequence to generate a coordination planning scheme.

8. A limb coordination planning device for a humanoid robot crossing obstacles, characterized in that, The device includes: The data acquisition module is used to acquire the humanoid robot's torso tilt angle, load distribution, and grip height as input parameters. The load center of gravity calculation module is used to calculate the lateral offset distance of the load center of gravity based on the load distribution and the torso tilt angle, and to determine the center of gravity height linkage amount by combining the grip height and the lateral offset distance of the load center of gravity. The working condition judgment module is used to judge the load state based on the center of gravity height linkage amount and calculate the dynamic torque amplification factor under the high-position grip working condition. The roll boundary determination module is used to adjust the allowable range of the roll angle according to the dynamic torque amplification factor and determine the roll amplitude boundary by integrating the load distribution. The trajectory generation module is used to construct constraints based on the tilt amplitude boundary and generate limb joint trajectories that satisfy the constraints using inverse kinematics. The coordination planning module is used to evaluate the coordination by dynamically simulating the trajectory of the limb joints, and generate a limb coordination planning scheme for the humanoid robot based on the evaluation results.

9. The limb coordination planning device for a humanoid robot crossing obstacles as described in claim 8, characterized in that, The process of obtaining the smoothed humanoid robot torso tilt angle, load distribution, and filtered grip height as input parameters includes: The raw data of the humanoid robot's torso tilt angle were collected, the raw values ​​of the load mass on the left and right sides were collected, and the original values ​​of the grip height were obtained from the encoders of the shoulder and elbow joints and converted by positive kinematics. The original data of the humanoid robot's torso tilt angle, the original values ​​of the load mass on the left and right sides, and the original value of the grip height are subjected to low-pass filtering to obtain the smoothed torso tilt angle, the filtered values ​​of the load mass on the left and right sides, and the filtered grip height. The load distribution is calculated based on the filtered load mass values ​​on the left and right sides, serving as a parameter characterizing the degree of load asymmetry. The smoothed humanoid robot torso tilt angle, the load distribution, and the filtered grip height are used as input parameters.

10. The limb coordination planning device for a humanoid robot crossing obstacles as described in claim 8, characterized in that, The step of calculating the lateral offset distance of the load center of gravity based on the load distribution and the torso tilt angle, and determining the center of gravity height linkage amount by combining the grip height and the lateral offset distance of the load center of gravity, includes: The coordinates of the lateral application point of the resultant load force are determined based on the load distribution and the lateral coordinates of the wrist end position. The lateral spacing of the load is calculated based on the coordinates of the lateral action point and the position of the spinal axis. Geometric projection is then performed using the tilt angle of the humanoid robot's torso to obtain the lateral offset distance of the load's center of mass. The vertical deviation is calculated based on the difference between the grip height and the preset standard torso height, and the center of gravity height linkage is calculated based on the vertical deviation and the lateral offset distance.

Images