Mechanical arm self-adaptive grabbing method based on geometric constraint and gravity optimization

Abstract

The invention discloses a mechanical arm self-adaptive grabbing method based on geometric constraint and gravity optimization, and relates to the field of robot motion planning, and the mechanical arm self-adaptive grabbing method comprises the steps that three-dimensional point cloud data of a target object is acquired, and a smooth surface area is segmented through geometric feature analysis; candidate grabbing point pairs meeting normal vector parallel constraint and friction cone constraint are screened; generating a grabbing strategy including the target clamping force and the grabbing pose based on the candidate grabbing point pair; taking the grabbing pose as a target, and generating a motion track from the initial pose of the mechanical arm to the grabbing pose and with optimized joint load by introducing a path planning algorithm of a gravitational torque cost function; and the mechanical arm is controlled to move along the movement track, and control modes are automatically switched according to the distance and tail end force feedback till grabbing is completed. The stability, the energy efficiency and the safety of the grabbing operation of the mechanical arm in the unstructured environment are improved through geometric physical constraint screening of grabbing points, gravity optimization planning of a movement track and self-adaptive variable stiffness control.

Description

Technical Field

[0001] This invention relates to the field of robot motion planning, and more specifically to an adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization. Background Technology

[0002] In unstructured robotic operations, such as warehouse sorting, logistics loading and unloading, or flexible production lines, robotic arms need to autonomously adapt to objects of different positions, shapes, and materials to achieve stable, safe, and efficient grasping.

[0003] Existing data-driven grasping methods based on deep learning typically output grasping poses directly by training neural networks. These methods heavily rely on a large amount of labeled data, and their recognition and localization are prone to failure when faced with irregularly shaped objects, reflective surfaces, or complex geometric structures outside the training set. More importantly, the grasping suggestions they generate often lack physical interpretability and fail to explicitly consider key geometric and physical constraints such as surface normals, local curvature, and frictional properties at the grasping point. Therefore, these methods lack robustness at the physical level, are prone to grasping slippage or failure, and cannot guarantee the stability of generalized grasping of unknown objects.

[0004] Furthermore, in terms of robotic arm motion trajectory planning, mainstream algorithms usually take the shortest geometric path or minimum motion time as the optimization objective, ignoring the dynamic characteristics of the robotic arm itself, especially the influence of gravity load. When grasping heavy objects, the algorithm may plan a trajectory that puts the robotic arm in a fully extended state, causing the root joint to bear huge gravitational torque, which not only increases the energy consumption and heat generation of the drive motor, but may also trigger overload protection or accelerate the wear of mechanical parts, lacking proactive optimization of joint load during the motion process.

[0005] Meanwhile, traditional robotic arm position control, such as high-gain PID control, typically maintains constant high stiffness during the grasping contact phase. This control strategy is extremely sensitive to positioning errors. According to Hooke's Law, even a small position overshoot can generate a huge instantaneous impact force under rigid contact. When grasping fragile items such as glassware and easily deformable food, this rigid impact can easily damage the target object. It lacks the ability to adaptively adjust compliance according to the contact state and cannot meet the dual requirements of rapid positioning and safe contact.

[0006] Therefore, designing an adaptive grasping method for robotic arms based on geometric constraints and gravity optimization, which can balance physical robustness, motion efficiency, and contact safety, to meet the challenges of complex and diverse grasping tasks in unstructured environments, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides an adaptive grasping method for robotic arms based on geometric constraints and gravity optimization, which aims to solve the problems of insufficient grasping stability of existing data-driven grasping methods in unstructured environments and excessive joint load caused by existing path planning methods ignoring the dynamic characteristics of robotic arms, thereby improving the stability, energy efficiency and safety of robotic arms in grasping operations in unstructured environments.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: An adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization includes the following steps: S1. Obtain the 3D point cloud data of the target object and segment the smooth surface area through geometric feature analysis; S2. On the smooth surface region, select candidate gripping point pairs that satisfy the normal vector parallel constraint and the friction cone constraint; S3. Generate a grasping strategy including target clamping force and grasping pose based on the candidate grasping point pairs; S4. Using the grasping pose as the target, a path planning algorithm with gravitational torque cost function is introduced to generate a motion trajectory from the initial pose of the robotic arm to the grasping pose with optimized joint load. S5. Control the robotic arm to move along the motion trajectory, and automatically switch the control mode according to the distance and end force feedback until the grasping is completed.

[0009] Preferably, in step S1, segmenting the smooth surface region through geometric feature analysis includes: Calculate the surface normal vector of each point in the point cloud based on principal component analysis; Calculate the local curvature at each point; Based on a region growing algorithm, and using local curvature as the growth criterion, regions with curvature change rates less than a preset threshold are selected. Neighboring points are clustered into the same geometrically smooth surface.

[0010] Preferably, S2 includes: Within the segmented smooth surface region, search for pairs of points whose surface normal vectors are parallel and opposite to each other, wherein the angle error between the normal vectors of the two points is less than a preset angle. ; For each pair of points, a friction cone model of its contact point is established, and it is determined whether the force direction applied by the robotic arm fingertip is located within the friction cone; the constraint condition of the friction cone is: , The angle between the direction of the applied force and the normal vector of the contact point surface. The static friction coefficient is preset. Point pairs selected through parallel normal constraints and friction cone constraints are determined as candidate grab point pairs.

[0011] Preferably, S3 includes: Estimate the mass of the target object based on its point cloud features, and calculate the target clamping force by combining the gravitational acceleration and the preset safety factor. The midpoint coordinates and direction of the line connecting the two contact points of the candidate grasping point are determined as the grasping pose, which includes the position and attitude of the end effector. Preferably, in step S4, the path planning algorithm is an improved RRT* algorithm, whose incremental cost function for path expansion is:

[0012] in, This represents the state vector of the parent node in the joint space. This represents the state vector of the child node in the joint space. This is the cost term for geometric path smoothness. This is the cost term for gravitational torque. These are the weighting coefficients.

[0013] Preferably, the geometric path smoothness cost term Using weighted Euclidean distance:

[0014] Where n is the total number of joints in the robotic arm. and These are the angle values ​​of the i-th joint in the state vectors of the parent and child nodes, respectively. is the weighting coefficient for the i-th joint.

[0015] Preferably, the gravitational moment cost term Calculations based on a static gravity compensation model include: Computational robotic arm in configuration Below, the joint torque vector required to overcome the weight of its own connecting rod and the weight of the end load. :

[0016] in, Let be the gravity term vector generated by the weight of the robotic arm's own links. The Jacobian matrix of the robotic arm under the current configuration. For the generalized force vector of the end load; The gravitational torque cost term is defined as a normalized form of the sum of the absolute values ​​of the torques acting on all joints:

[0017] in, It is a vector The i-th component, It is the maximum rated torque allowed for the i-th joint.

[0018] Preferably, in step S5, the automatic switching control mode includes three stages: During the free motion phase, high-rigidity position control is used to control the robotic arm to quickly track the motion trajectory; During the contact buffer phase, when the distance d between the end effector and the nearest point on the target object surface is less than the preset safety distance... When this happens, switch to low stiffness impedance control; During the steady-state clamping phase, when the end force sensor detects the component of the contact force in the normal direction... Reaching the preset contact force threshold When necessary, switch to force-position hybrid control to maintain stable gripping.

[0019] Preferably, the low-stiffness impedance control law of the contact buffer stage is:

[0020] Where M is the inertia parameter matrix, B is the damping parameter matrix, K is the stiffness parameter matrix, and x, , These are the actual position, velocity, and acceleration of the end effector, respectively. For the desired position, The contact force detected by the end sensor; At this stage, the component values ​​of the stiffness parameter matrix K in the end-approach direction are... It is set to 5% to 10% of the stiffness value corresponding to the free motion stage.

[0021] Preferably, step S5 further includes a slip detection and suppression step: During the steady-state clamping phase, the tangential force component signal output by the end force sensor is monitored in real time, and high-frequency filtering is performed to extract the micro-vibration signal characteristics. If the characteristic amplitude of the micro-flicker signal exceeds a preset threshold, it is determined to be a precursor to slippage, triggering a slippage suppression strategy to increase the target clamping force until the micro-flicker signal disappears; the slippage suppression strategy is expressed as follows:

[0022] in, The target clamping force at the current moment, The preset force gain coefficient, The estimated slip rate is calculated until the micro-vibration signal disappears.

[0023] As can be seen from the above technical solution, compared with the prior art, the technical solution of the present invention has the following beneficial effects: 1. This method utilizes the local curvature of point clouds for geometric surface segmentation and selects grasping points based on normal vector parallel constraints and physical friction cone models. This allows the grasping plan to be independent of pre-trained models or semantic labels of specific objects. It can derive grasping poses that meet mechanical stability conditions from the shape features of the object itself, thereby effectively dealing with the diversity of objects in unstructured environments and improving the physical reliability and success rate of grasping actions.

[0024] 2. In the specific path planning, by introducing the gravity torque cost function and optimizing the RRT* algorithm, the generated trajectory not only avoids obstacles, but also prioritizes the robotic arm configuration with lower joint torque load. This prompts the robotic arm to automatically adopt a more labor-saving and shorter lever arm posture when moving, especially when grasping heavy objects. This helps to extend the service life of key drive components and improves the reliability and continuity of the system when performing heavy grasping tasks.

[0025] 3. Based on the mechanism of automatically switching control modes according to distance and force feedback, when approaching an object, it switches from high-rigidity position control to low-rigidity impedance control, actively reducing the equivalent stiffness at the moment of contact. It can smoothly absorb the contact impact caused by positioning error. Combined with the slip detection and suppression function in the steady-state clamping stage, it can ensure stable gripping while reducing the risk of damage or slippage of the target object due to rigid contact or improper gripping force, thus improving the overall safety of gripping operations. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0027] Figure 1 A flowchart of an adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the Monte Carlo simulation experiment results provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figure 1As shown, this embodiment provides an adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization, including the following steps: S1. Obtain the 3D point cloud data of the target object and segment the smooth surface area through geometric feature analysis; S2. On the smooth surface region, select candidate gripping point pairs that satisfy the normal vector parallel constraint and the friction cone constraint; S3. Generate a grasping strategy including target clamping force and grasping pose based on the candidate grasping point pairs; S4. Using the grasping pose as the target, a path planning algorithm with gravitational torque cost function is introduced to generate a motion trajectory from the initial pose of the robotic arm to the grasping pose with optimized joint load. S5. Control the robotic arm to move along the motion trajectory, and automatically switch the control mode according to the distance and end force feedback until the grasping is completed.

[0030] This method improves the stability of the robotic arm in grasping unknown or irregularly shaped objects by selecting grasping points based on point cloud geometric features and physical friction constraints; optimizes the joint load and energy consumption of the robotic arm's motion trajectory by introducing a gravitational torque cost function in path planning; and achieves compliant and safe contact with fragile objects by automatically switching stiffness control modes based on distance and force feedback, thereby comprehensively improving the success rate and safety of grasping operations in unstructured environments.

[0031] The following provides a further explanation of each step and related features in the above method; In this embodiment, S1, the three-dimensional point cloud data of the target object is acquired, and the smooth surface region is segmented by geometric feature analysis. In the specific implementation, a depth camera is used to acquire RGB-D images of the target scene at a frequency of 30Hz. Through intrinsic parameter calibration and coordinate transformation, a raw 3D point cloud with millimeter-level precision is generated. To improve data quality, a statistical outlier removal filter is then used for processing: the average distance to its K=50 nearest neighbors is calculated for each point. If the distance follows a Gaussian distribution, all noise points with a mean distance exceeding 3 times the standard deviation are removed. This step can effectively filter out discrete points caused by sensor noise and ambient light interference.

[0032] Furthermore, the smooth surface region is segmented through geometric feature analysis, including: Calculate the surface normal vector of each point in the point cloud based on principal component analysis; Calculate the local curvature at each point; Based on a region growing algorithm, and using local curvature as the growth criterion, regions with curvature change rates less than a preset threshold are selected. Neighboring points are clustered into the same geometrically smooth surface; Specifically, a spherical neighborhood with a radius of 5 times the average spacing of the point cloud is constructed centered on this point. The covariance matrix of all points within the neighborhood is calculated, and the unit eigenvector corresponding to the smallest eigenvalue is the surface normal vector of that point. Subsequently, the local curvature is calculated using the eigenvalues ​​of the same covariance matrix. A value close to 0 indicates that the local area is planar, while a larger value indicates higher curvature. Finally, curvature-based region growing and segmentation is performed: starting from the point with the minimum curvature as the seed point, the curvature difference between this point and all its neighboring points is less than a preset threshold. The points are merged into the same region and iteratively grown until they can no longer be expanded, thereby accurately segmenting the object point cloud into multiple continuous geometrically smooth surfaces. This process is based on underlying geometric properties and does not rely on prior knowledge of the object, which is the primary key to achieving physically stable grasping of any unknown object.

[0033] In this embodiment S2, on the smooth surface region, candidate grasping point pairs that satisfy the normal vector parallel constraint and the friction cone constraint are selected; including: Within the segmented smooth surface region, search for pairs of points whose surface normal vectors are parallel and opposite to each other, wherein the angle error between the normal vectors of the two points is less than a preset angle. The preset angle here This constraint ensures that the two fingertips of the gripper can apply force in a direction approximately perpendicular to the contact surface, forming the optimal force transmission direction, which is the geometric basis for satisfying the force closure condition. For each pair of points, a friction cone model of its contact point is established, and it is determined whether the force direction applied by the robotic arm fingertip is located within the friction cone; the constraint condition of the friction cone is: , The angle between the direction of the applied force and the normal vector of the contact point surface. The static friction coefficient is preset; at this point, The value is not fixed, but is estimated based on the visual texture roughness of the object's surface through a lightweight lookup table. For example, 0.15 is used for smooth metal surfaces and 0.8 is used for rough rubber surfaces, so as to dynamically adapt to different materials. Point pairs selected through parallel normal constraints and friction cone constraints are determined as candidate grasping point pairs. The dual constraint screening mechanism in this step eliminates those options that are spatially reachable but physically unstable from a massive number of possible contact combinations. The final output candidate grasping point pairs have an extremely high probability of successful grasping.

[0034] In this embodiment S3, a grasping strategy including target clamping force and grasping pose is generated based on the candidate grasping point pairs; including: Estimate the mass of the target object based on its point cloud features, and calculate the target clamping force by combining the gravitational acceleration and the preset safety factor. The midpoint coordinates and direction of the line connecting the two contact points of the candidate grasping point are determined as the grasping pose, which includes the position and attitude of the end effector. When generating a specific grasping strategy, it is necessary to determine the target clamping force. This embodiment estimates the mass of an object using point clouds. First, calculate the 3D convex hull of the segmented object points or use octree voxelization to obtain an approximate volume V. Then, based on the task scenario, preset a typical material density ρ. The target clamping force is given by the formula Calculate, where g is the acceleration due to gravity. To ensure a reliable grip even in the presence of vibration or acceleration, a safety margin is provided. Subsequently, the grasping pose is determined based on the selected candidate grasping point pairs, and the grasping position is defined as the midpoint of the line connecting the two points. The grasping posture is determined by two vectors: one is the direction of the normalized line vector as the approach direction of the gripper, and the other is the normal of the gripper obtained by cross product of the vector with the normal vector of one of the points and then normalized. Thus, the position and posture homogeneous transformation matrix of the end effector at the grasping moment is uniquely determined. This step transforms the abstract grasping point into a complete grasping instruction that the robotic arm controller can directly execute, achieving a closed-loop process from perception to planning.

[0035] In this embodiment S4, taking the grasping pose as the target, a path planning algorithm with gravitational torque cost function is introduced to generate a motion trajectory from the initial pose of the robotic arm to the grasping pose with optimized joint load. The path planning algorithm is an improved RRT* algorithm, and its incremental cost function for path expansion is:

[0036] in, This represents the state vector of the parent node in the joint space. This represents the state vector of the child node in the joint space. This is the cost term for geometric path smoothness. This is the cost term for gravitational torque. These are the weighting coefficients.

[0037] Furthermore, the geometric path smoothness cost term Using weighted Euclidean distance:

[0038] Where n is the total number of joints in the robotic arm. and These are the angle values ​​of the i-th joint in the state vectors of the parent and child nodes, respectively. is the weighting coefficient for the i-th joint, which is proportional to the maximum distance from that joint to the end effector.

[0039] Furthermore, the gravitational torque cost term Calculations based on a static gravity compensation model include: Computational robotic arm in configuration Below, the joint torque vector required to overcome the weight of its own connecting rod and the weight of the end load. :

[0040] in, Let be the gravity term vector generated by the weight of the robotic arm's own links. The Jacobian matrix of the robotic arm under the current configuration. For the generalized force vector of the end load; The gravitational torque cost term is defined as a normalized form of the sum of the absolute values ​​of the torques acting on all joints:

[0041] in, It is a vector The i-th component, It is the maximum rated torque allowed for the i-th joint.

[0042] When generating a specific grasping strategy, it is necessary to determine the target clamping force. This embodiment estimates the mass of an object using point clouds. First, calculate the 3D convex hull of the segmented object points or use octree voxelization to obtain an approximate volume V. Then, based on the task scenario, preset a typical material density ρ. The target clamping force is given by the formula Calculate, where g is the acceleration due to gravity. To ensure a reliable grip even in the presence of vibration or acceleration, a safety margin is provided. This step employs an improved RRT* algorithm for trajectory planning, which introduces a gravitational torque cost term into the cost function. Each time the algorithm expands, it generates a new node. At that time, in addition to calculating the geometric distance cost In addition, it will calculate the gravitational torque cost under this posture in real time; specifically, based on the gravity term vector in the robotic arm's dynamics model. And the current Jacobian matrix Combined with end load gravity Through formula Calculate the torque required at each joint to overcome gravity; Furthermore, to make this cost term dimensionless and comparable to the geometric cost, it is normalized, and the total cost is... Weighting coefficients in It can be calibrated experimentally to balance path length and energy consumption; The path planning algorithm here enables the planner to actively avoid strenuous postures that cause the robotic arm to fully extend and joint torque to be huge when searching for a path. Instead, it tends to choose energy-saving configurations with the elbow tucked in and the center of gravity closer to the base, thereby optimizing motion energy consumption at the planning level.

[0043] In this embodiment, S5, the robotic arm is controlled to move along the motion trajectory, and the control mode is automatically switched according to the distance and end force feedback until the grasping is completed. The automatic switching control mode includes three stages: During the free motion phase, high-rigidity position control is used to control the robotic arm to quickly track the motion trajectory; During the contact buffer phase, when the distance d between the end effector and the nearest point on the target object surface is less than the preset safety distance... When this happens, switch to low stiffness impedance control; During the steady-state clamping phase, when the end force sensor detects the component of the contact force in the normal direction... Reaching the preset contact force threshold When necessary, switch to force-position hybrid control to maintain stable gripping.

[0044] Furthermore, the low-stiffness impedance control law for the contact buffer stage is as follows:

[0045] Where M is the inertia parameter matrix, B is the damping parameter matrix, K is the stiffness parameter matrix, and x, , These are the actual position, velocity, and acceleration of the end effector, respectively. For the desired position, The contact force detected by the end sensor; At this stage, the component values ​​of the stiffness parameter matrix K in the end-approach direction are... It is set to 5% to 10% of the stiffness value corresponding to the free motion stage.

[0046] Furthermore, S5 also includes slip detection and suppression steps: During the steady-state clamping phase, the tangential force component signal output by the end force sensor is monitored in real time, and high-frequency filtering is performed to extract the micro-vibration signal characteristics. If the characteristic amplitude of the micro-flicker signal exceeds a preset threshold, it is determined to be a precursor to slippage, triggering a slippage suppression strategy to increase the target clamping force until the micro-flicker signal disappears; the slippage suppression strategy is expressed as follows:

[0047] in, The target clamping force at the current moment, The preset force gain coefficient, The estimated slip rate is calculated until the micro-vibration signal disappears.

[0048] When the robotic arm executes its trajectory, the control system automatically switches between three stages based on the distance to the target and the contact force information. The free movement stage employs high-rigidity position control. To ensure rapid and accurate tracking of the planned trajectory, when the shortest distance d between the end point and the target object surface is less than the preset safety distance. When the thickness reaches 20mm, the system immediately switches to the contact buffer stage; During the contact buffer phase, the impedance control law is activated. And reduce the components of the stiffness matrix K in the approach direction. This reduces the impact force to 5%-10% during the free motion phase, while increasing damping B to suppress oscillations. This active softening strategy allows the contact impact force to be reduced from 50N to 1N when the visual positioning has a 1-2mm error that causes the robotic arm to overshoot slightly, according to Hooke's Law. This enables smooth and non-damaging contact with fragile items such as eggs and glass cups. After entering the steady-state clamping stage, when the six-dimensional force sensor detects the normal contact force... Exceeding the threshold This indicates that reliable contact has been achieved, the system has locked the position ring, and is maintaining the target clamping force. Meanwhile, the slip detection mechanism is activated: the tangential force signal from the force sensor is subjected to a 200Hz high-pass filter to extract the high-frequency micro-vibration component; if its energy exceeds the threshold continuously within a 0.5-second window, slip is determined to have occurred, triggering the suppression strategy, dynamically increasing the clamping force until the slip stops. This series of state-based intelligent switching and compensation ensures the robustness and safety of the entire process from contact to gripping.

[0049] In specific implementation verification, such as Figure 2 The Monte Carlo simulation results shown provide crucial data support for the effectiveness of this method. The simulation, through hundreds of random sampling experiments, statistically analyzed key physical indicators under two control strategies, including... Figure 2 Contact force distribution in a and Figure 2The damage rate statistics in b show that when in contact with fragile objects, the instantaneous contact force peak generated by traditional high-stiffness PID control has a wide distribution, and its maximum value often exceeds 5N, resulting in an average damage rate of more than 15.5%. However, by using the staged variable stiffness control of this invention, the contact force is effectively suppressed to a low force range below 2N, the peak value is significantly reduced, and the estimated damage rate can be reduced to less than 1%. This result not only quantifies the breakthrough improvement in contact safety of the robotic arm adaptive grasping method based on geometric constraints and gravity optimization in this embodiment, but also verifies the overall effectiveness of the technical path of ensuring the stability of the grasping point through geometric constraint screening, reducing joint load through gravity optimization planning, and finally achieving safe contact through variable stiffness compliant control. It can be widely applied to the reliable operation of robotic arms in unstructured, high-safety-requirement scenarios.

[0050] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0051] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization, characterized in that, Includes the following steps: S1. Obtain the 3D point cloud data of the target object and segment the smooth surface area through geometric feature analysis; S2. On the smooth surface region, select candidate gripping point pairs that satisfy the normal vector parallel constraint and the friction cone constraint; S3. Generate a grasping strategy including target clamping force and grasping pose based on the candidate grasping point pairs; S4. Using the grasping pose as the target, a path planning algorithm with gravitational torque cost function is introduced to generate a motion trajectory from the initial pose of the robotic arm to the grasping pose with optimized joint load. S5. Control the robotic arm to move along the motion trajectory, and automatically switch the control mode according to the distance and end force feedback until the grasping is completed.

2. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 1, characterized in that, In step S1, segmenting the smooth surface region through geometric feature analysis includes: Calculate the surface normal vector of each point in the point cloud based on principal component analysis; Calculate the local curvature at each point; Based on a region growing algorithm, and using local curvature as the growth criterion, regions with curvature change rates less than a preset threshold are selected. Neighboring points are clustered into the same geometrically smooth surface.

3. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 1, characterized in that, S2 includes: Within the segmented smooth surface region, search for pairs of points whose surface normal vectors are parallel and opposite to each other, wherein the angle error between the normal vectors of the two points is less than a preset angle. ; For each pair of points, a friction cone model of its contact point is established, and it is determined whether the force direction applied by the robotic arm fingertip is located within the friction cone; the constraint condition of the friction cone is: , The angle between the direction of the applied force and the normal vector of the contact point surface. The static friction coefficient is preset. Point pairs selected through parallel normal constraints and friction cone constraints are determined as candidate grab point pairs.

4. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 1, characterized in that, S3 includes: Estimate the mass of the target object based on its point cloud features, and calculate the target clamping force by combining the gravitational acceleration and the preset safety factor. The midpoint coordinates and direction of the line connecting the two contact points of the candidate grasping point are determined as the grasping pose, which includes the position and attitude of the end effector.

5. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 1, characterized in that, In S4, the path planning algorithm is an improved RRT* algorithm, and its incremental cost function for path expansion is: in, This represents the state vector of the parent node in the joint space. This represents the state vector of the child node in the joint space. This is the cost term for geometric path smoothness. This is the cost term for gravitational torque. These are the weighting coefficients.

6. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 5, characterized in that, The geometric path smoothness cost term Using weighted Euclidean distance: Where n is the total number of joints in the robotic arm. and These are the angle values ​​of the i-th joint in the state vectors of the parent and child nodes, respectively. is the weighting coefficient for the i-th joint.

7. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 5, characterized in that, The gravitational torque cost term Calculations based on a static gravity compensation model include: Computational robotic arm in configuration Below, the joint torque vector required to overcome the weight of its own connecting rod and the weight of the end load. : in, Let be the gravity term vector generated by the weight of the robotic arm's own links. The Jacobian matrix of the robotic arm under the current configuration. For the generalized force vector of the end load; The gravitational torque cost term is defined as a normalized form of the sum of the absolute values ​​of the torques acting on all joints: in, It is a vector The i-th component, It is the maximum rated torque allowed for the i-th joint.

8. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 1, characterized in that, In S5, the automatic switching control mode includes three stages: During the free motion phase, high-rigidity position control is used to control the robotic arm to quickly track the motion trajectory; During the contact buffer phase, when the distance d between the end effector and the nearest point on the target object surface is less than the preset safety distance... When this happens, switch to low stiffness impedance control; During the steady-state clamping phase, when the end force sensor detects the component of the contact force in the normal direction... Reaching the preset contact force threshold When necessary, switch to force-position hybrid control to maintain stable gripping.

9. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 8, characterized in that, The low-stiffness impedance control law for the contact buffer stage is as follows: Where M is the inertia parameter matrix, B is the damping parameter matrix, K is the stiffness parameter matrix, and x, , These are the actual position, velocity, and acceleration of the end effector, respectively. For the desired position, The contact force detected by the end sensor; At this stage, the component values ​​of the stiffness parameter matrix K in the end-approach direction are... It is set to 5% to 10% of the stiffness value corresponding to the free motion stage.

10. The adaptive grasping method for a robotic arm based on geometric constraints and gravity optimization according to claim 1, characterized in that, S5 further includes a slip detection and suppression step: During the steady-state clamping phase, the tangential force component signal output by the end force sensor is monitored in real time, and high-frequency filtering is performed to extract the micro-vibration signal characteristics. If the characteristic amplitude of the micro-flicker signal exceeds a preset threshold, it is determined to be a precursor to slippage, triggering a slippage suppression strategy to increase the target clamping force until the micro-flicker signal disappears; the slippage suppression strategy is expressed as follows: in, The target clamping force at the current moment, The preset force gain coefficient, The estimated slip rate is calculated until the micro-vibration signal disappears.

Images