Intelligent planning method for flexible bone removal path of manipulator for livestock head bone and meat separation

Abstract

This invention discloses an intelligent planning method for the flexible deboning path of a robotic arm in separating meat from bone in livestock heads, belonging to the field of intelligent meat slaughtering and processing technology. By analyzing the bone-meat interface features such as the location, angle, and gripping point coordinates of the separation point in the heads of livestock like pigs, cattle, and sheep, a bone-meat interface separation feature dataset is constructed to determine the flexible deboning path parameters for the robotic arm. Based on these parameters, a set of instructions for flexible deboning of different livestock heads after slaughter is constructed, and an idle motion path and a bone-meat separation and deboning path, including the robotic arm's movement path parameters, are planned. The movement path parameters are then converted into spatial motion quantities and rotation angle adjustment quantities of the robotic arm through inverse kinematics, guiding the robotic arm to achieve adaptive flexible deboning of different livestock heads. This invention solves the problems of high labor intensity and low precision in manual deboning of livestock by-products with bones, providing a solution for flexible deboning of these products.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology. More specifically, this invention relates to an intelligent planning method for the flexible deboning path of a robotic arm that separates meat from bone in the head of livestock. Background Technology

[0002] Deboned head meat from livestock such as pigs, cattle, and sheep is a popular meat product among consumers. Traditional methods of separating meat from bones in livestock faces have three main drawbacks: ① Complete reliance on manual deboning results in low production efficiency, high labor intensity, and a single workstation processing capacity of less than 30 heads per hour, making it unsuitable for large-scale production. ② The manual deboning process is dangerous and prone to burns. After braising, the surface temperature of the bone-in byproduct of the livestock head exceeds 85°C, requiring workers to debon quickly, creating a safety hazard in the high-temperature environment. ③ Manual separation involves significant errors, relies heavily on experience, and makes it difficult for workers to accurately identify the connection between bone and muscle, easily leading to bone residue remaining in the muscle tissue, affecting the product's taste. Individual differences in worker operation also contribute to raw material waste, high losses, and significant economic damage.

[0003] In summary, existing processes for separating meat from bones in livestock heads suffer from reliance on manual labor, high waste, and low efficiency, failing to meet the demands of industrialized production processes for standardized, uniform, and highly precise meat separation with minimal waste. Currently, intelligent and flexible technologies and equipment for separating meat from bones in livestock heads are lacking. Therefore, the industry urgently needs a solution for separating meat from bones that reduces worker workload, integrates multi-dimensional information sensing technology, and offers high precision, efficiency, intelligence, and standardization, filling the technological and equipment gaps in this field both domestically and internationally. Summary of the Invention

[0004] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0005] To achieve these objectives and other advantages of the present invention, a method for intelligent planning of a flexible deboning path for a robotic arm to separate meat from bones in livestock heads is provided, comprising the following steps: By analyzing the bone-meat interface features of livestock heads, a bone-meat interface separation feature dataset of livestock heads is constructed. The bone-meat interface features include at least the location of the bone-meat separation point, the separation angle, and the coordinates of the grasping point. Based on the mechanical constraints of the robot, its workspace coordinate system is determined, and the origin of the workspace coordinate system is taken as the initial position of the robot. The bone-flesh interface separation feature dataset is transformed and calculated to map the bone-flesh interface separation feature dataset to the workspace coordinate system of the robot arm, thereby obtaining the flexible bone removal path parameters of the robot arm, which include the pose of the robot arm at the grasping point. Based on the parameters of the robotic arm's flexible deboning path, a set of instructions for flexible deboning of livestock heads after slaughter is constructed. This includes planning the robotic arm's idle motion path and the bone-meat separation and deboning path. The idle motion path is from the initial position to the grasping point, and the bone-meat separation and deboning path is from the grasping point to the bone disposal point. The A* algorithm is used to calculate the movement path parameters for the idle motion path, and the preset path planning rules are used to calculate the movement path parameters for the bone-meat separation and deboning path. Based on the movement path parameters of the robotic arm, the spatial motion and angular adjustment of the robotic arm in the idle motion path and the bone separation and deboning path are calculated by inverse kinematics, and the spatial motion and angular adjustment are converted into control parameters of the robotic arm.

[0006] Preferably, the bone-muscle interface separation feature dataset includes: the coordinates of the grasping points of the mandible and maxilla, the angles at which the gripper of the robotic arm cuts into the mandible and maxilla, the position coordinates of the bone-muscle separation point of the mandible, and the position coordinates of the bone-muscle separation point of the maxilla.

[0007] Preferably, the movement path parameters of the robot in its workspace coordinate system include: the pose coordinates of the turning points in the movement trajectory of the robot from the initial position to the grasping point, the pose coordinates of the turning points in the movement trajectory of the robot from the grasping point to the bone disposal point, and the pose coordinates of the bone disposal point.

[0008] Preferably, the preset path planning rules include mandibular bone removal path planning rules; The mandibular bone removal path planning rules include: rotating from the mandibular grasping point P1 to the mandibular bone-flesh separation point O as the center, moving away from the maxilla by a first angle δ to the turning point P1', and then translating along the y-axis of the manipulator away from the maxilla by a first distance L1 to the mandibular bone removal point P1'', wherein the first angle δ is not less than 60°, and the first distance L1 is not less than the width of the mandible. Let the pose coordinates of the mandibular grasping point P1 be (x1, y1, z1, θ1), where (x1, y1, z1) are the position coordinates of the mandibular grasping point P1 in the workspace coordinate system of the robot arm, θ1 is the angle at which the gripper of the robot arm cuts into the mandible, and the position coordinates of the mandibular bone-muscle separation point O are (x1, y1, z1, θ1). o , y o , z o Given that the pose coordinates of the turning point P1' are (x1', y1', z1', θ1'), then... θ1'=θ1; Let the pose coordinates of the mandibular bone rejection point P1'' be (x1'', y1'', z1'', θ1''), then θ1''=θ0, where θ0 is the initial angle of the gripper of the robot.

[0009] Preferably, the preset path planning rules include maxillary bone removal path planning rules; The maxillary bone removal path planning rules include: translating from the maxillary bone grasping point P2 along the x-axis of the manipulator away from the animal's head by a second distance L2 to the turning point P2', and then translating along the y-axis of the manipulator away from the animal's head by a third distance L3 to the maxillary bone discarding point P2'', wherein the second distance L2 is not less than 0.5 times the length of the maxilla, and the third distance L3 is not less than the width of the maxilla; Let the pose coordinates of the maxillary grasping point P2 be (x2, y2, z2, θ2), where (x2, y2, z2) are the position coordinates of the maxillary grasping point P2 in the workspace coordinate system of the robot arm, θ2 is the angle at which the gripper of the robot arm cuts into the maxilla, and the pose coordinates of the turning point P2' be (x2', y2', z2', θ2'). θ2'=θ2; Let the pose coordinates of the maxillary bone rejection point P2'' be (x2'', y2'', z2'', θ2''), then θ1''=θ0, where θ0 is the initial angle of the gripper of the robot.

[0010] Preferably, the inverse kinematics solution includes: for a circular trajectory, using an interpolation equation based on the center and angular velocity of the circular arc to obtain the coordinates of the interpolation points on the circular trajectory; for a straight trajectory, using linear interpolation to calculate the coordinates of the interpolation points on the straight trajectory, calculating the spatial motion and rotation adjustment based on the current pose coordinates and the target pose coordinates, and converting the spatial motion and rotation adjustment into control parameters for the manipulator; Preferably, the robotic arm includes a three-dimensional motion module with servo motors, and the conversion of the spatial motion quantities into control parameters for the robotic arm includes: based on the single-pulse displacement δ of the three-dimensional motion module and a preset movement speed. Calculate the number of pulses N and frequency f required for the servo motor. in,( , , The spatial motion quantity is obtained by calculating the difference between the target position coordinates and the current position coordinates of the three-dimensional moving module.

[0011] Preferably, the robotic arm includes a rotary cylinder that drives the gripper to rotate, and converting the rotation angle adjustment amount into control parameters for the robotic arm includes: converting the rotation angle adjustment amount of the robotic arm into a target rotation angle for the rotary cylinder. Then, combining the rotational inertia J of the rotary cylinder with the target angular acceleration... and load torque T L The minimum output torque T required for the rotary cylinder can be calculated using the following formula. m : T m =J +T L Therefore, based on the minimum output torque T m The structural parameters of the rotary cylinder are matched with the corresponding flow control valve opening parameter K. v With the working air pressure parameter P, where: K v =C In the formula, C is the valve port flow coefficient, A is the effective area of ​​the cylinder piston, r is the equivalent radius of the output shaft, and the working air pressure parameter P is determined according to the air supply pressure and safety factor of the rotary cylinder. Based on the target rotation angle Given the preset rotary cylinder motion time parameters, the velocity curve of the rotary motion is planned, and the target angular acceleration is calculated from it. ; Based on the target rotation angle Based on the corresponding gripper pose and a pre-established gripper-load interaction mechanical model, the load torque T is calculated. L ; The gripper-load interaction mechanical model was established through the following methods: testing the peeling force, adhesion force, elastic modulus, and plastic deformation parameters of the muscle tissue at the bone-meat interface of the livestock head under different working conditions, and establishing a load characteristic database; measuring the friction coefficient of the gripper clamping surface, the effective clamping area, the equivalent radius of the output shaft, and the transmission efficiency; establishing the relationship between the load torque, gripper clamping force, and instantaneous peeling force at the bone-meat interface based on rigid body dynamics and contact mechanics; installing a torque sensor at the end of the manipulator to collect the actual load torque, and iteratively calibrating the model parameters using the least squares method to control the model prediction error within a preset range.

[0012] Preferably, the bone-meat interface separation feature dataset is obtained through a deep learning-based method, specifically including: collecting samples categorized by breed, weight, and degree of cooking, covering different livestock breeds and processing conditions; using an instance segmentation model to complete pixel-level segmentation of the bone-meat interface and detection of deboning sites, and improving generalization ability through transfer learning; using a clustering algorithm to cluster samples of the same breed and specification to establish a dataset of deboning path parameters and control parameters, forming a basic deboning template; and using an incremental learning mechanism to back-feed the path parameters with the highest net meat yield in actual operation to the database, iteratively optimizing the model accuracy.

[0013] Preferably, the free path planning of the airborne motion path is implemented using the A* algorithm, which specifically includes: pre-constructing a grid map of the robot's workspace and marking fixed obstacles; using the robot's current real-time pose as the starting point and the next grasping point as the ending point, using the A* algorithm to search for the global shortest collision-free path and generate a sequence of key points for the path.

[0014] The present invention has at least the following beneficial effects: The method provided by this invention helps to automate the separation of meat and bone in livestock heads. By acquiring the three-dimensional coordinate data of the meat-bone interface and performing coordinate transformation, visual information can be directly linked to the motion control of the robotic arm, thereby reducing reliance on human experience and enabling the separation process to be carried out continuously and automatically.

[0015] Based on pre-defined path planning rules, particularly for the different motion trajectories of the mandible and maxilla, the robotic arm's movements are made more consistent with the anatomical structure of the livestock's head. This method guides the actuator along a reasonable path for separation operations, helping to reduce unnecessary damage to muscle tissue and improve the integrity and efficiency of bone-meat separation.

[0016] By using inverse kinematics, the planned path is transformed into spatial motion quantities and angular adjustment quantities for each axis, and further converted into specific control parameters for servo motors and rotary cylinders, such as pulse count, frequency, and valve opening, thus achieving precise digital control of the robot's end effector position, speed, and torque. This helps improve the repeatability and stability of motion execution.

[0017] The path planning rules and control parameter calculation process involved in this method allow for adaptive settings by adjusting key parameters (such as rotation angle, translation distance, and movement speed) based on different livestock species or individual sizes. This flexible planning capability enables the system to cope with differences in the objects being handled within a certain range, improving the applicability and practicality of the equipment.

[0018] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the hardware configuration upon which the method described in this invention relies; Figure 2 This is a flowchart of the process by which the robotic arm performs flexible bone removal according to the present invention; Figure 3 A flowchart illustrating the intelligent planning of a flexible bone removal path in the control system described in this invention; Figure 4 This is a schematic diagram of the mandibular bone removal path and the maxillary bone removal path described in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0021] This embodiment addresses the scenario of separating meat from bones in multiple parts of a livestock's head after half-cooking. Taking the planning process of separating meat from bones in a pig's head as an example, it elaborates in detail the specific implementation process of the intelligent human-like deboning method and control strategy for multiple parts of a livestock's head.

[0022] like Figure 1 As shown, the robotic arm in this embodiment of the invention relies on the following hardware configuration: Conveying mechanism 1, such as a conveyor belt; The clamping mechanism 2 is disposed on the conveying mechanism 1 and is used to clamp the head of the livestock after braising and make the extension direction of the jawbone of the livestock head consistent with the moving direction of the conveying mechanism, so that the head of the livestock moves with the conveying mechanism. The three-axis truss robotic arm 3 spans the conveying mechanism. The three-axis truss robotic arm is provided with two sets of x-axis moving modules along the moving path of the conveying mechanism. Each x-axis moving module is provided with a y-axis moving module along the width of the conveying mechanism, and each y-axis moving module is provided with a z-axis moving module along the vertical direction. The mandibular flexible bone removal end actuator 4 is located at the lower end of the z-axis moving module upstream of the moving path of the conveying mechanism 1; The maxillary flexible bone removal end actuator 5 is located at the lower end of the z-axis moving module downstream of the moving path of the conveying mechanism 1; The control system (host computer) is communicatively connected to the conveying mechanism 1, the three-axis gantry robotic arm 3, the mandibular flexible bone removal end actuator 4, and the maxillary flexible bone removal end actuator 4, respectively.

[0023] Here, the x-axis, y-axis, and z-axis movement modules are all driven and controlled by servo motors. The mandibular flexible bone removal end actuator includes a rotary cylinder, a gripper cylinder located at the output end of the rotary cylinder, and a gripper located at the output end of the gripper cylinder. The rotary cylinder is used to adjust the gripping angle of the gripper, and the gripper cylinder is used to drive the gripper to perform gripping or releasing actions. The control system controls the conveying mechanism, the three-axis gantry robot arm, the mandibular flexible bone removal end actuator, and the maxillary flexible bone removal end actuator to perform bone removal operations on the mandible and maxilla according to the following path planning method.

[0024] In this embodiment, the robotic arm is equipped with a three-axis motion module for both the mandible and maxilla. The motion origin of each of these two three-axis motion modules is defined. The motion origin is the mechanical zero point of the end effector, that is, the spatial position corresponding to the geometric center of the gripper when the three-dimensional motion module is reset to the initial position of the stroke limit in the X, Y, and Z axes. This point serves as the absolute coordinate origin (0,0,0) of the end effector's workspace coordinate system. The coordinates of all key bone removal points (such as the mandibular gripping point P1, the maxillary gripping point P2, and the mandibular bone-muscle separation point O) and the trajectory parameters of the end effector are all absolute spatial coordinates relative to the aforementioned workspace origin.

[0025] like Figure 2 As shown in this embodiment, the intelligent planning method for the flexible deboning path of a robotic arm to separate meat from bones in livestock heads includes the following steps: S1. By analyzing the bone-meat interface features of livestock heads, construct a bone-meat interface separation feature dataset of livestock heads. The bone-meat interface features include at least the location of the bone-meat separation point, the separation angle, and the coordinates of the grasping point. S2. Based on the mechanical constraints of the robot's actuator, determine its workspace coordinate system, with the origin of the workspace coordinate system serving as the robot's initial position. S3. Perform coordinate system transformation calculation on the bone-flesh interface separation feature dataset to map the bone-flesh interface separation feature dataset to the workspace coordinate system of the actuator, and obtain the flexible bone removal path parameters of the manipulator, which include the pose of the manipulator at the grasping point. S4. Based on the flexible deboning path parameters of the robotic arm, construct a set of instructions for flexible deboning of livestock heads after slaughter, including planning the idle motion path of the robotic arm and the bone and meat separation and deboning path. The idle motion path is from the initial position to the grasping point, and the bone and meat separation and deboning path is from the grasping point to the discarding point. For the idle motion path, the A* algorithm is used to calculate the movement path parameters, and for the bone and meat separation and deboning path, the preset path planning rules are used to calculate the movement path parameters. This embodiment employs a two-tiered path planning mechanism: a fixed operational path and a dynamic idle path. The bone-and-flesh separation and deboning path (i.e., the path from the grasping point to the bone disposal point) strictly follows the mandibular / maxillary bone removal path rules (a combination of circular rotation and linear translation). This path is determined by the animal's anatomical structure and must be consistently executed to ensure the integrity of bone-and-flesh separation and avoid bone residue. The idle motion path (i.e., the path from any current position of the robotic arm, such as the previous bone disposal point, standby position, or fault reset position, to the next grasping point) uses free path planning, aiming for the shortest time, no collisions, and low energy consumption, with no mandatory requirements on the path shape. The planning logic for the two paths is clearly distinguished and processed separately.

[0026] S5. Based on the movement path parameters of the robotic arm, calculate the spatial motion and angular adjustment of the actuator in the idle motion path and the bone separation and dismantling path through inverse kinematics, and convert the spatial motion and angular adjustment into control parameters of the actuator.

[0027] Specifically, the bone-meat interface features of the livestock head can be acquired using a 3D vision system. This system can include a structured light scanner or a binocular camera, mounted above the conveyor mechanism directly facing the livestock's head. The point cloud data acquired by the scanner is filtered and fitted to obtain the coordinates of the gripping points of the mandible and maxilla, the fitted curve of the bone-meat separation interface (the fitted curve of the bone-meat separation interface can be obtained by fitting the 3D point cloud data with a polynomial or spline curve, and the fitting order can be selected as 3rd or 4th order), the angle at which the grippers cut into and grip the mandible and / or maxilla at the bone-meat separation interface (i.e., the separation angle), the length and width of the mandible and maxilla respectively, and the position coordinates of the bone-meat separation points of the mandible and maxilla, etc. Figure 4 The mandibular axis of rotation indicated in the middle is the mandibular separation point, and the end point of the maxillary suture is the maxillary bone-muscle separation point.

[0028] After analyzing the bone-meat interface features of livestock heads, the data related to the deboning task performed by the robotic arm can be extracted to form a bone-meat interface separation feature dataset. For example, the coordinates of the gripping points of the mandible and maxilla, the angles at which the robotic arm's grippers cut into the mandible and maxilla, the length and width of the mandible and maxilla, and the position coordinates of the bone-meat separation points of the mandible and maxilla.

[0029] The workspace coordinate system uses the mechanical zero point as the origin of motion, with the x-axis along the conveying direction, the y-axis along the width direction, and the z-axis along the vertical direction. Mechanical constraints include the travel limits of each axis, and the upper limits of velocity and acceleration, which determine the boundaries of the coordinate system. Coordinate transformation is achieved through a homogeneous transformation matrix, unifying the point cloud in the visual coordinate system to the robot's base coordinate system. The transformed data is the parameter for the robot's flexible bone removal path.

[0030] Furthermore, the acquisition of the bone-meat interface separation feature dataset can also be achieved using deep learning-based methods. By analyzing the bone-meat interface features such as the location of the bone-meat separation point, separation angle, and grasping point coordinates in the head of livestock such as pigs, cattle, and sheep, a bone-meat interface separation feature dataset can be constructed. Specifically, this includes: collecting samples categorized by breed, weight, and degree of braising, covering mainstream breeds such as pigs and cattle, including different weight ranges and braising processes; labeling the bone-meat interface features and key site coordinates for each sample; using instance segmentation models such as Mask R-CNN to complete pixel-level segmentation of the bone-meat interface and detection of deboning sites; improving generalization ability through transfer learning; initializing the backbone network and detection branch parameters using models pre-trained on general datasets such as COCO; fine-tuning the model using self-collected and labeled livestock and poultry head samples; using the K-means clustering algorithm to cluster samples of the same breed and specification to form basic deboning templates for pigs and cattle; iteratively optimizing the deep learning model and basic templates through an incremental learning mechanism; and feeding back the bone-meat interface parameters with the highest net meat yield and lowest loss in actual production line operation to the database as new training data to re-optimize model parameters and iteratively improve model accuracy.

[0031] Next, based on the bone-meat interface separation feature dataset, the flexible deboning path parameters of the robotic arm are determined. Based on these parameters, a flexible deboning instruction set for the heads of different livestock after slaughter is constructed, and the idle motion path of the robotic arm and the bone-meat separation and deboning path are planned. Specifically, in this embodiment, the path planning adopts a two-tiered architecture of "fixed operation path + dynamic idle path," which ensures both the quality of bone-meat separation and the optimal movement of the robotic arm from any position to the target point. Level 1: Bone and Meat Separation Operation Path: This refers to the path from the grasping point to the bone disposal point, strictly following the mandibular / maxillary bone removal path rules (a combination of circular rotation and straight-line translation). This path is determined by the animal's anatomical structure and must be consistently executed to ensure the integrity of bone and meat separation and avoid bone residue.

[0032] For example, the path planning rules for the bone-meat separation operation are set as follows: the mandibular grasping point rotates at least 60° around the mandibular bone-meat separation point along an arc trajectory to the transition point, and then translates at least the width of the mandibular bone along the negative y-axis to the bone disposal point; the maxillary grasping point translates at least 0.5 times the length of the maxillary bone along the positive x-axis to the transition point, and then translates at least the width of the maxillary bone along the positive y-axis to the bone disposal point.

[0033] Second level: Idle movement path: This refers to the path taken by the robotic arm from any current pose (such as the previous bone discard point, standby position, or fault reset position) to the next grasping point. It is implemented using the A* global path planning algorithm. Before conducting path search, the entire workspace of the robot arm needs to be converted into a grid map. This is the basis for the successful execution of the A* algorithm. The grid map will clearly mark fixed obstacles such as equipment racks, conveyor lines, and guardrails, and clearly mark impassable areas to avoid collisions during the movement of the robot arm and define a safe range for subsequent path search. After the grid map is constructed and obstacles are marked, the robot's current pose information is obtained in real time. This real-time pose is used as the starting point for path search, and the next grasping point to be reached is set as the end point for path search. This gives the algorithm a clear starting and target position, ensuring that the search direction matches the actual operation requirements.

[0034] Based on a determined starting point and destination, the A* algorithm is launched to conduct a global path search. The algorithm combines a preset cost function and comprehensively considers factors such as path length, movement energy consumption, and obstacle avoidance requirements. It filters feasible paths one by one in the grid map, continuously compares and optimizes, and finally selects the optimal path that satisfies the shortest movement distance and avoids obstacles throughout the entire process.

[0035] After the algorithm completes the search, it will decompose the optimal path into a continuous sequence of path key points. These key points contain precise pose coordinate information, which can directly provide the execution basis for the movement of the robot arm, allowing the robot arm to move smoothly and efficiently from the current position to the gripping point according to this sequence, thus completing the path execution of the idle motion path.

[0036] The movement path parameters also include the pose coordinates of each turning point. Inverse kinematics is used to calculate the coordinates of each interpolation point by employing central angle interpolation for circular segments and linear interpolation for straight segments, thereby determining the spatial motion and end-effector angle adjustment for each axis. The planned path is discretized into interpolation points and transformed into actuator control parameters through inverse kinematics. This mechanism ensures that regardless of the robot's current position, it can reach the work start point in the shortest time without collision, while also avoiding the risk of equipment collisions.

[0037] Finally, the movement path parameters of the robotic arm are transformed into the spatial motion and rotation adjustment of the robotic arm joints through inverse kinematics, guiding the actuator to achieve adaptive flexible deboning of the head for different animal species. The robotic arm can reach any target coordinate using the following process: First, the target pose coordinates are subtracted from the current pose coordinates fed back in real time by the encoder of the robotic arm's movement module to obtain the spatial motion (Δx, Δy, Δz) and rotation adjustment Δθ; second, according to the above two-level planning logic, the idle motion path and the bone separation and deboning path are processed respectively; then, the coordinates of all interpolation points on the trajectory are transformed into the single-axis displacement of the three-dimensional movement module and the rotation angle of the rotary cylinder through inverse kinematics; next, according to the parameter calculation formula in the subsequent steps, the displacement and angle are transformed into the number of servo motor pulses N, frequency f, and the opening degree Kv and working air pressure P of the flow control valve of the rotary cylinder; finally, the servo system and pneumatic system execute commands, and the encoder and displacement sensor provide real-time feedback on the pose deviation and dynamically correct it to achieve precise positioning.

[0038] When converting spatial motion quantities into servo motor control parameters, a closed-loop stepper motor or AC servo motor can be selected. Its single-pulse displacement δ can be 0.01 mm, and the preset moving speed v can be 50 mm / s. The pulse count and frequency are calculated based on the motion quantity. When converting angular adjustment quantities into rotary cylinder control parameters, a rotary cylinder with position feedback can be selected. The target rotation angle is obtained through inverse kinematics, and the load torque is estimated using a gripper-load mechanical model based on the gripper shape, meat elasticity parameters, and friction coefficient. The cylinder output torque is calculated based on the moment of inertia, target angular acceleration, and load torque, and then combined with structural parameters such as the cylinder piston area and output shaft radius to match the valve opening and working air pressure. This system can achieve continuous operation in pig head deboning experiments, with a path tracking error of less than ±1 mm and repeatability meeting the separation process requirements. This method can improve the automation and consistency of deboning operations and reduce the intensity of manual intervention.

[0039] In another embodiment, the bone-meat interface separation feature dataset includes: the coordinates of the gripping points of the mandible and maxilla, the angles at which the grippers of the actuator cut into the mandible and maxilla, the position coordinates of the bone-meat separation point of the mandible, and the position coordinates of the bone-meat separation point of the maxilla.

[0040] The specific composition of the bone-muscle interface separation feature dataset includes several key geometric parameters. Specifically, the coordinates of the mandibular grasping point can be set to the center point 20mm to 30mm from the anterior edge of the mandible, and the coordinates of the maxillary grasping point can be set to 15mm to 25mm from the anterior edge of the mandible. The angle at which the gripper cuts into the mandible can be set to the angle between the tangent of the fitted curve of the bone-muscle interface at the mandibular grasping point and the horizontal plane; the angle at which it cuts into the maxilla can be set to the angle between the tangent of the fitted curve of the bone-muscle interface at the maxillary grasping point and the horizontal plane. The location coordinates of the mandibular bone-muscle separation point can be defined as the midpoint of the line connecting the two mandibular glenoid fossae, and the location coordinates of the maxillary bone-muscle separation point can be defined as the junction of the maxillary midline and the anterior mandible.

[0041] The acquisition of the aforementioned data can be accomplished using a vision measurement system. This system may include a line laser 3D scanner, mounted above the conveyor mechanism, directly facing the center of the livestock head gripper. The scanner scans the passing pig's head, acquiring high-density point cloud data of its surface. Point cloud processing software can segment the data, identify the contours of the mandible and maxilla, and extract the coordinates of the aforementioned gripping and separation points using feature recognition algorithms. The gripper's entry angle is calculated by the path planning module based on the skeletal geometry and a preset entry strategy, which aims to make the gripper cut along the tangent direction of the bone surface to reduce resistance. The fitted curve is generated using the least squares method to ensure that the curve is smooth and closely approximates the actual bone-meat interface.

[0042] The acquired coordinate and angle data serve as the initial input for path planning. The calculated coordinates of the grasping point, separation point, etc., are all referenced to the scanner coordinate system. This data is then transmitted to the control system for coordinate system transformation and subsequent trajectory planning. Through data collection and analysis of different batches of pig head samples, this data structure can stably provide the geometric constraints required for path planning, providing the robotic arm with an accurate spatial positioning reference, thereby supporting it in completing bone-meat separation actions that conform to anatomical features.

[0043] In another embodiment, the movement path parameters of the actuator in its workspace coordinate system include: the pose coordinates of the turning points in the movement trajectory of the actuator from the initial position to the grasping point, the pose coordinates of the turning points in the movement trajectory of the actuator from the grasping point to the bone disposal point, and the pose coordinates of the bone disposal point.

[0044] The movement path parameters are specifically composed of a series of key point pose coordinates. The actuator moves from the initial position to the gripping point, and there can be one or more turning points in its trajectory. For example, a transition point can be set for obstacle avoidance. The pose coordinates of this point can be preset according to the robot's workspace and the approximate position of the livestock's head.

[0045] The trajectory turning points from the grab point to the bone disposal point, such as P1' in the mandibular bone disposal path and P2' in the maxillary bone disposal path, can be calculated according to the path planning rules below. The pose coordinates of the bone disposal points, such as P1'' in the mandibular bone disposal point and P2'' in the maxillary bone disposal point, can be set above the collection container located at the edge of the workspace.

[0046] The generation of these trajectory parameters relies on the path planning module in the control system. This module can run on an industrial programmable logic controller or an industrial computer. The planning module receives the bone-muscle interface coordinate data after coordinate system transformation and calls the preset mandibular and maxillary bone removal path rules. Based on the geometric relationships and constraints in the rules, the module automatically calculates the precise pose coordinates (including x, y, z, θ) of turning points such as P1' and P2' using coordinate rotation and translation formulas. The coordinates of the bone disposal point are calibrated and stored during system initialization based on the actual installation position of the collection container. During the calculation process, all coordinate values ​​are based on the workspace coordinate system of the actuator, which is established through calibration during the installation and commissioning of the robot arm.

[0047] The calculated sequence of pose coordinates (grasping point, turning point, and bone removal point) constitutes the complete motion path parameters. At the hardware level, the actuator can be a Cartesian coordinate robot module driven by servo motors. The x, y, and z linear modules can be mounted on corresponding guide rails on the frame, jointly driving the end effector's rotary cylinder and gripper movement. The control system uses these discrete pose coordinates as target points and generates smooth, continuous motion commands for each servo axis through inverse kinematics and interpolation algorithms. Turning points in the trajectory provide clear waypoints for the robot, enabling it to approach the target with a reasonable posture, perform the bone removal action, and ultimately transport the bone to the removal point for release. This parameter definition method provides a clear and calculable path basis for the robot's automated motion.

[0048] In another embodiment, such as Figure 2 As shown, the path planning rules include the mandibular osteotomy path planning rules; The mandibular bone removal path planning rule includes: rotating from the mandibular grasping point P1 to the mandibular bone-muscle separation point O as the center, rotating by a first angle δ in a direction away from the maxilla to the turning point P1', and then translating along the y-axis of the actuator in a direction away from the maxilla to the mandibular bone removal point P1'', wherein the first angle δ is not less than 60°, and the first distance L1 is not less than the width of the mandible; Let the pose coordinates of the mandibular grasping point P1 be (x1, y1, z1, θ1), where (x1, y1, z1) are the position coordinates of the mandibular grasping point P1 in the workspace coordinate system of the actuator, θ1 is the angle at which the gripper of the actuator cuts into the mandible, and the position coordinates of the mandibular bone-muscle separation point O are (x1, y1, z1, θ1). o , y o , z o Given that the pose coordinates of the turning point P1' are (x1', y1', z1', θ1'), then... (1) θ1'=θ1; Let the pose coordinates of the mandibular bone rejection point P1'' be (x1'', y1'', z1'', θ1''), then (2) θ1''=θ0, where θ0 is the initial angle of the gripper of the actuator.

[0049] The mandibular bone removal path planning rules specifically describe the continuous motion process of the actuator from grasping the bone to discarding it. The first rotation angle δ is a key parameter, specifically set to 60° to ensure effective separation of the mandible from surrounding tissues. The first translation distance L1 is another key parameter, its specific value set based on the width of the pig mandible calculated by the visual measurement system, to ensure the bone is completely removed from the working area. The gripper ingress angle θ1 of the actuator is set as the angle between the tangent of the fitted curve of the bone-meat separation interface at the mandibular grasping point and the horizontal plane. This angle aims to allow the gripper to wedge into the bone-meat separation interface with optimal mechanical conditions.

[0050] The implementation of the above path rules relies on a series of coordinate transformation calculations. First, based on the coordinates (x1, y1, z1) of the mandibular grasping point P1 and the coordinates (x, y1, z1) of the mandibular bone-muscle separation point O obtained by the 3D vision system... o , y o , z o The coordinates (x1', y1', z1') of the turning point P1' are calculated by substituting these coordinates into the rotation formula. This rotation calculation is performed in the path planning module of the control system, which can run on an industrial computer. Subsequently, based on the calculated coordinates of P1' and the selected translation distance L1, the coordinates (x1'', y1'', z1'') of the bone disposal point P1'' are calculated using the translation formula. Throughout the entire movement, the gripper's attitude angle remains constant at the entry angle θ1 during the rotation phase and returns to the initial safety angle θ0 after translating to the bone disposal point, ensuring safe release of the bone.

[0051] The path of the mandibular debonding actuator from its initial position to the grasping point can also be based on another planning rule, such as: from the initial position P d0 (x) d0 , y d0 , z d0 , θ d0 Translate along the x-axis to a position P with the same x-axis coordinates as the grab point P1. 1a (x1, y) d0 , z d0 , θ d0 Then, translate along the y-axis to a position above the grab point P1. 1b (x1, y1, z) d0 , θ d0 Finally, it descends along the z-axis to P1, which is 1-2 mm above the gripping point P1. 1c (x1, y1, z1+1 (or 2), θ) d0 Then adjust the gripper attitude angle from the initial angle θ. d0 To θ1, the gripper eventually cuts into the bone-flesh interface along the z-axis to reach the gripping point P1.

[0052] In another embodiment, such as Figure 3 As shown, the path planning rules include the maxillary bone removal path planning rules; The maxillary bone removal path planning rule includes: translating a second distance L2 from the maxillary grasping point P2 away from the animal's head along the x-axis of the actuator to the turning point P2', and then translating a third distance L3 away from the animal's head along the y-axis of the actuator to the maxillary bone discarding point P2'', wherein the second distance L2 is not less than 0.5 times the length of the maxilla, and the third distance L3 is not less than the width of the maxilla; Let the pose coordinates of the maxillary grasping point P2 be (x2, y2, z2, θ2), where (x2, y2, z2) are the position coordinates of the maxillary grasping point P2 in the workspace coordinate system of the actuator, θ2 is the angle at which the gripper of the actuator cuts into the maxilla, and the pose coordinates of the turning point P2' be (x2', y2', z2', θ2'). (3) θ2'=θ2; Let the pose coordinates of the maxillary bone rejection point P2'' be (x2'', y2'', z2'', θ2''), then (4) θ1''=θ0, where θ0 is the initial angle of the gripper of the actuator.

[0053] The maxillary bone removal path planning rule defines a movement strategy of first horizontal retreating and then lateral translation. The second translation distance L2 can be set to 0.5 times the length of the maxilla. For example, for a pig's maxilla with a length of approximately 300 mm, L2 can be specifically set to 150 mm to ensure that the maxilla is pulled away from the head sufficiently to sever the main connective tissue. The third translation distance L3 can be set to 80 mm, which must be greater than the width of the maxilla to ensure that the bone can be completely removed and detached from the gripper. The width of the maxilla can be calculated using a visual measurement system. The angle θ2 at which the gripper cuts into the maxilla can be set to the angle between the tangent of the fitted curve of the bone-meat separation interface at the maxillary gripping point and the horizontal plane.

[0054] The control system performs path planning calculations based on the aforementioned rules. First, the coordinates (x2, y2, z2) of the maxillary grasping point P2 obtained by the vision system, along with the selected L2 value, are substituted into the translation formula to directly calculate the coordinates (x2', y2', z2') of the turning point P2'. At this point, the gripper's attitude angle θ2' remains the same as the entry angle θ2. Subsequently, using the calculated P2' coordinates as a reference, and combining them with the selected L3 value, the translation formula is substituted again to calculate the coordinates (x2'', y2'', z2'') of the final bone release point P2''. Upon reaching the bone release point, the gripper's attitude angle is adjusted from θ2' to the initial safety angle θ0, preparing for bone release. The entire calculation process generates a polyline path sequence consisting of the three points P2, P2', and P2''.

[0055] The path of the maxillary bone dislocation actuator from its initial position to the grasping point can also be based on another planning rule, such as: from the initial position P u0 (x) u0 , y u0 , z u0 , θ u0 Translate along the y-axis to a position P with the same y-axis coordinates as the grab point P2. 2a (x) u0 , y2, z u0 , θ u0 Then translate along the x-axis to above the grab point P2. 2b (x2, y2, z) d0 , θ d0 Finally, it descends along the z-axis to point P1~2mm above the gripping point P2. 2c (x2, y2, z2+1 (or 2), θ) d0 Then adjust the gripper attitude angle from the initial angle θ. u0 To θ2, the gripper finally cuts into the bone-flesh interface along the z-axis to reach the gripping point P2.

[0056] Furthermore, for the maxilla, during bone removal, the bone can be clamped first, and then the grippers can be dynamically adjusted to rotate back and forth at a certain angle (e.g., 15°) around the point of separation of the maxilla from the bone, causing the maxilla to break along the bone joint. After breaking, the maxilla can be removed along the bone removal path. The position of the farthest point of the rotation path of the maxilla can be obtained using a calculation method that is basically the same as the pose coordinate of the turning point P1' in the bone removal path of the mandible, except that the point of separation of the mandible from the bone O is changed to the point I along the bone joint of the maxilla.

[0057] As for the gripper cylinders of the maxillary / mandibular bone removal end actuator, they are controlled in real time based on force feedback sensors installed on the grippers. The start and stop of the gripper cylinders are controlled according to a pre-set gripping force threshold. During the path from the initial position to the grasping point, the gripper cylinders remain relaxed; during the path from the grasping point to the bone removal point, the gripper cylinders remain clamped; and during the path from the bone removal point back to the initial position, the gripper cylinders remain relaxed.

[0058] In another embodiment, the inverse kinematics solution includes: for a circular trajectory, using an interpolation equation based on the center and angular velocity of the circular arc to obtain the coordinates of the interpolation point on the circular trajectory; for a straight trajectory, using a linear interpolation method to calculate the coordinates of the interpolation point on the straight trajectory, calculating the spatial motion and rotation adjustment based on the current pose coordinates and the target pose coordinates, and converting the spatial motion and rotation adjustment into control parameters of the actuator.

[0059] Based on the pose parameters of the mandibular grasping point P1(x1, y1, z1, θ1) and the maxillary grasping point P2(x2, y2, z2, θ2), combined with the Cartesian translation and rotation motion of the end effector, inverse kinematics and trajectory interpolation are performed. The initial pose of the end effector is set as the system zero point P0(x0, y0, z0, θ0) (calibrated during initialization), and the target pose is P... n (x n ,y n ,z n ,θ n (n=1 corresponds to the mandible, n=2 corresponds to the maxilla), then the interpolation equation for the circular arc trajectory of mandibular osteotomy is: (5) Among them, angular velocity , The preset moving speed and interpolation period Δt = 0.1s ensure a smooth trajectory without impact.

[0060] For the remaining straight line segments, linear interpolation is used, and the formula for the coordinates of the interpolation points is: (6) Among them, exercise time . In actual operation, the corresponding interpolation algorithm is automatically invoked based on the trajectory type (circular arc or straight line) and parameters output by the path planning. For example, for the mandibular bone removal path, the controller first calls the circular arc interpolation algorithm to generate a dense point sequence from P1 to P1'; upon reaching P1', it immediately switches to the straight line interpolation algorithm to generate a straight line point sequence from P1' to P1''. This segmented interpolation method discretizes the complex continuous trajectory into tiny program segments that the controller can execute, enabling the robot to accurately and smoothly reproduce the planned bone removal path, thereby ensuring the accuracy and consistency of the separation action.

[0061] In another embodiment, the actuator includes a three-dimensional motion module with a servo motor, and converting the spatial motion into control parameters for the actuator includes: based on the single-pulse displacement δ of the three-dimensional motion module and a preset movement speed. Calculate the number of pulses N and frequency f required for the servo motor. (7) in,( , , The spatial motion quantity is obtained by calculating the difference between the target position coordinates and the current position coordinates of the three-dimensional moving module.

[0062] Converting spatial motion quantities into servo motor control parameters involves setting several core parameters. The single-pulse displacement δ of the 3D movement module is a key parameter; its specific value can be determined based on the accuracy level of the selected module, for example, it can be set to 0.01mm. The preset movement speed v can be selected based on the rhythm and stability requirements of the actual bone removal operation, for example, it can be set to 50mm / s. The servo motor required for this calculation can be an AC servo motor or a closed-loop stepper motor. The 3D movement module itself can be a standardized ball screw linear module or a synchronous belt linear module. The module frame can be made of aluminum alloy profiles, the ball screw can be made of hardened alloy steel, and the guide rail can be made of high-carbon chromium bearing steel.

[0063] In system assembly, the three-dimensional motion modules are mounted on the rigid main frame of the robot. Specifically, the x-axis module is mounted on the top crossbeam along the conveying direction, the y-axis module is mounted perpendicular to the conveying direction on the x-axis module's moving slide, and the z-axis module is mounted vertically on the y-axis module's moving slide. Servo motors are directly connected to the lead screws or synchronous pulleys of each module via couplings. During operation, the motion controller calculates the spatial motion quantities (Δx, Δy, Δz) output from the inverse kinematics solution—that is, the differences between the target position and the current position on each coordinate axis—combining the δ value pre-stored in the controller parameter table and the v value set for the current action segment, using the formula. The calculated pulse number N determines the total angle the motor needs to rotate, and the frequency f determines the motor's rotation speed. The controller then sends instructions containing the pulse number and frequency to the corresponding servo drivers.

[0064] The value of the single-pulse displacement δ is typically determined by the encoder resolution of the linear module and servo motor used, and is explicitly given in the product manual. The preset movement speed v is determined based on process experiments to ensure sufficient efficiency without tearing or damaging muscle tissue during bone-fetal separation. During the debugging phase, the robot arm can be tested under no-load and load conditions (such as gripping simulated bones). The deviation between the actual motion trajectory and the theoretical trajectory is measured using a laser tracker to verify the accuracy of the pulse control parameters. This method transforms abstract spatial displacement into digital pulse commands that the servo drive system can precisely execute, achieving digital closed-loop control of the robot arm's end effector position and speed, providing a foundation for the stable and repeatable completion of the flexible bone removal path.

[0065] In another embodiment, the actuator includes a rotary cylinder that drives the gripper to rotate, and converting the rotation angle adjustment amount into control parameters of the actuator includes: converting the rotation angle adjustment amount of the actuator into a target rotation angle of the rotary cylinder. Then, combining the rotational inertia J of the rotary cylinder with the target angular acceleration... and load torque T L The minimum output torque T required for the rotary cylinder can be calculated using the following formula. m : T m =J +T L (8) Therefore, based on the minimum output torque T m The structural parameters of the rotary cylinder are matched with the corresponding flow control valve opening parameter K. v With the working air pressure parameter P, where: K v =C (9) In the formula, C is the valve port flow coefficient, A is the effective area of ​​the cylinder piston, r is the equivalent radius of the output shaft, and the working air pressure parameter P is determined according to the air supply pressure and safety factor of the rotary cylinder. Based on the target rotation angle Given the preset rotary cylinder motion time parameters, the velocity curve of the rotary motion is planned, and the target angular acceleration is calculated from it. ; Based on the target rotation angle Based on the corresponding gripper pose and a pre-established gripper-load interaction mechanical model, the load torque T is calculated. L .

[0066] The actuator for the above calculations can be a vane-type or rack-and-pinion rotary cylinder, which is directly fixed to the lower end of the Z-axis moving module via a mounting flange. The cylinder's output shaft is connected to the gripper's support shaft via a coupling. The flow control valve used for control can be a proportional valve or a servo valve, which is connected to the cylinder's inlet via a pipeline and installed on a valve island near the cylinder. The calculation task is undertaken by the main controller (such as a PLC or industrial computer). The workflow is as follows: The controller first determines the joint space angle adjustment obtained from the path planning as the target angle Δθ that the rotary cylinder needs to rotate. Then, it calls the pre-stored rotary cylinder moment of inertia J and, based on the motion time allocated for the current action, plans an S-shaped velocity curve, from which the target angular acceleration α is obtained by differentiation. Simultaneously, based on the contact state between the gripper and the bones and muscles at the current and target angles, the load torque T is calculated using a pre-set gripper-load interaction mechanics model (this model is based on soft tissue elasticity and friction coefficient). L Substituting into formula T m =J•α+T L Calculate the minimum required output torque.

[0067] To obtain the minimum output torque T m Then, combining the structural parameters of the selected rotary cylinder (effective piston area A, equivalent output shaft radius r, these parameters are obtained from the product manual) and the determined working air pressure P, substitute them into formula K. v =C Calculate the required valve opening parameter K. v The K vThe value is sent as an instruction to the actuator of the flow control valve to adjust the air flow, thereby precisely controlling the output torque and movement of the cylinder. During the system commissioning phase, the accuracy of the load torque model and the entire control parameter calculation chain can be verified and calibrated by measuring the deviation between the actual rotation angle and the target rotation angle through rotation tests on standard weights or simulated loads. This method transforms the gripper's attitude angle change requirements into quantified drive parameters for the pneumatic actuator, which helps to achieve stable and precise orientation of the end gripper under load variation conditions.

[0068] The process of establishing the pre-set gripper-load interaction mechanical model is as follows: Basic mechanical property testing: Using a universal testing machine, the peel force and adhesion of the bone-meat interface of livestock head under different working conditions were tested, as well as the elastic modulus and plastic deformation parameters of muscle tissue, and a load characteristic database was established. Gripper structure parameter calibration: Determine structural parameters such as the friction coefficient of the gripper clamping surface, the effective clamping area, the equivalent radius r of the output shaft, and the cylinder transmission efficiency; Mathematical Model Construction: Based on rigid body dynamics and contact mechanics theory, the core relationships are established: T L =F c ⋅μ⋅r+F s ⋅r (10) Among them, T L For load torque, F c F is the gripping force of the grippers, μ is the coefficient of friction between the grippers and the skeleton. s The instantaneous separation force at the bone-flesh interface; Model validation and calibration: A high-precision torque sensor is installed at the end of the actuator to collect the actual load torque under different working conditions. The model parameters are calibrated iteratively using the least squares method, and the final model prediction error is ≤5%.

[0069] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for intelligent planning of flexible deboning paths for a robotic arm to separate meat from bone in livestock heads, characterized in that, Includes the following steps: By analyzing the bone-meat interface features of livestock heads, a bone-meat interface separation feature dataset is constructed. The bone-meat interface features include at least the location of the bone-meat separation point, the separation angle, and the coordinates of the grasping point. Based on the mechanical constraints of the robot, its workspace coordinate system is determined, and the origin of the workspace coordinate system is taken as the initial position of the robot. The bone-flesh interface separation feature dataset is transformed and calculated to map the bone-flesh interface separation feature dataset to the workspace coordinate system of the robot arm, thereby obtaining the flexible bone removal path parameters of the robot arm, which include the pose of the robot arm at the grasping point. Based on the parameters of the robotic arm's flexible deboning path, a set of instructions for flexible deboning of livestock heads after slaughter is constructed. This includes planning the robotic arm's idle motion path and the bone-meat separation and deboning path. The idle motion path is from the initial position to the grasping point, and the bone-meat separation and deboning path is from the grasping point to the bone disposal point. The A* algorithm is used to calculate the movement path parameters for the idle motion path, and the preset path planning rules are used to calculate the movement path parameters for the bone-meat separation and deboning path. Based on the movement path parameters of the robotic arm, the spatial motion and angular adjustment of the robotic arm in the idle motion path and the bone separation and deboning path are calculated by inverse kinematics, and the spatial motion and angular adjustment are converted into control parameters of the robotic arm.

2. The intelligent planning method for flexible bone removal path of the robotic arm as described in claim 1, characterized in that, The bone-flesh interface separation feature dataset includes: the coordinates of the grasping points of the mandible and maxilla, the angles at which the gripper of the robotic arm cuts into the mandible and maxilla, the position coordinates of the bone-flesh separation point of the mandible, and the position coordinates of the bone-flesh separation point of the maxilla.

3. The intelligent planning method for flexible bone removal path of the robotic arm as described in claim 2, characterized in that, The movement path parameters of the robot in its workspace coordinate system include: the pose coordinates of the turning points in the movement trajectory of the robot from the initial position to the gripping point, the pose coordinates of the turning points in the movement trajectory of the robot from the gripping point to the bone disposal point, and the pose coordinates of the bone disposal point.

4. The intelligent planning method for flexible bone removal path of the robotic arm as described in claim 3, characterized in that, The preset path planning rules include the mandibular bone removal path planning rules; The mandibular bone removal path planning rules include: rotating from the mandibular grasping point P1 to the mandibular bone-flesh separation point O as the center, moving away from the maxilla by a first angle δ to the turning point P1', and then translating along the y-axis of the manipulator away from the maxilla by a first distance L1 to the mandibular bone removal point P1'', wherein the first angle δ is not less than 60°, and the first distance L1 is not less than the width of the mandible. Let the pose coordinates of the mandibular grasping point P1 be (x1, y1, z1, θ1), where (x1, y1, z1) are the position coordinates of the mandibular grasping point P1 in the workspace coordinate system of the robot arm, θ1 is the angle at which the gripper of the robot arm cuts into the mandible, and the position coordinates of the mandibular bone-muscle separation point O are (x1, y1, z1, θ1). o , y o , z o Given that the pose coordinates of the turning point P1' are (x1', y1', z1', θ1'), then... θ1'=θ1; Let the pose coordinates of the mandibular bone rejection point P1'' be (x1'', y1'', z1'', θ1''), then θ1''=θ0, where θ0 is the initial angle of the gripper of the robot.

5. The intelligent planning method for flexible bone removal path of the robotic arm as described in claim 3, characterized in that, The preset path planning rules include maxillary bone removal path planning rules; The maxillary bone removal path planning rules include: translating from the maxillary bone grasping point P2 along the x-axis of the manipulator away from the animal's head by a second distance L2 to the turning point P2', and then translating along the y-axis of the manipulator away from the animal's head by a third distance L3 to the maxillary bone discarding point P2'', wherein the second distance L2 is not less than 0.5 times the length of the maxilla, and the third distance L3 is not less than the width of the maxilla; Let the pose coordinates of the maxillary grasping point P2 be (x2, y2, z2, θ2), where (x2, y2, z2) are the position coordinates of the maxillary grasping point P2 in the workspace coordinate system of the robot arm, θ2 is the angle at which the gripper of the robot arm cuts into the maxilla, and the pose coordinates of the turning point P2' be (x2', y2', z2', θ2'). θ2'=θ2; Let the pose coordinates of the maxillary bone rejection point P2'' be (x2'', y2'', z2'', θ2''), then θ1''=θ0, where θ0 is the initial angle of the gripper of the robot.

6. The intelligent planning method for flexible bone removal path of a robotic arm as described in claim 4 or 5, characterized in that, The inverse kinematics solution includes: for a circular trajectory, the coordinates of the interpolation points on the circular trajectory are obtained by using the interpolation equation based on the center of the circle and the angular velocity; for a straight trajectory, the coordinates of the interpolation points on the straight trajectory are calculated by linear interpolation, and the spatial motion and rotation adjustment are calculated based on the current pose coordinates and the target pose coordinates, and the spatial motion and rotation adjustment are converted into the control parameters of the manipulator.

7. The intelligent planning method for flexible bone removal path of a robotic arm as described in claim 6, characterized in that, The robotic arm includes a three-dimensional motion module with servo motors. Converting the spatial motion quantities into control parameters for the robotic arm includes: based on the single-pulse displacement δ of the three-dimensional motion module and a preset movement speed. Calculate the number of pulses N and frequency f required for the servo motor. in,( , , The spatial motion quantity is obtained by calculating the difference between the target position coordinates and the current position coordinates of the three-dimensional moving module.

8. The intelligent planning method for flexible bone removal path of a robotic arm as described in claim 6, characterized in that, The robotic arm includes a rotary cylinder that drives the gripper to rotate. Converting the rotation angle adjustment into control parameters for the robotic arm includes converting the rotation angle adjustment into a target rotation angle for the rotary cylinder. Then, combining the rotational inertia J of the rotary cylinder with the target angular acceleration... and load torque T L The minimum output torque T required for the rotary cylinder can be calculated using the following formula. m : T m =J +T L Therefore, based on the minimum output torque T m The structural parameters of the rotary cylinder are matched with the corresponding flow control valve opening parameter K. v With the working air pressure parameter P, where: K v =C In the formula, C is the valve port flow coefficient, A is the effective area of ​​the cylinder piston, r is the equivalent radius of the output shaft, and the working air pressure parameter P is determined according to the air supply pressure and safety factor of the rotary cylinder. Based on the target rotation angle Given the preset rotary cylinder motion time parameters, the velocity curve of the rotary motion is planned, and the target angular acceleration is calculated from it. ; Based on the target rotation angle Based on the corresponding gripper pose and a pre-established gripper-load interaction mechanical model, the load torque T is calculated. L ; The gripper-load interaction mechanical model was established through the following methods: testing the peeling force, adhesion force, elastic modulus, and plastic deformation parameters of the muscle tissue at the bone-meat interface of the livestock head under different working conditions, and establishing a load characteristic database; measuring the friction coefficient of the gripper clamping surface, the effective clamping area, the equivalent radius of the output shaft, and the transmission efficiency; establishing the relationship between the load torque, gripper clamping force, and instantaneous peeling force at the bone-meat interface based on rigid body dynamics and contact mechanics; installing a torque sensor at the end of the manipulator to collect the actual load torque, and iteratively calibrating the model parameters using the least squares method to control the model prediction error within a preset range.

9. The intelligent planning method for flexible bone removal path of a robotic arm as described in claim 1, characterized in that, The bone-meat interface separation feature dataset was obtained using a deep learning-based method, specifically including: collecting samples categorized by breed, weight, and degree of cooking, covering different livestock breeds and processing conditions; using an instance segmentation model to complete pixel-level segmentation of the bone-meat interface and detection of deboning sites, and improving generalization ability through transfer learning; using a clustering algorithm to cluster samples of the same breed and specification to establish a dataset of deboning path parameters and control parameters, forming a basic deboning template; and using an incremental learning mechanism to back-feed the path parameters with the highest net meat yield in actual operation to the database, iteratively optimizing the model accuracy.

10. The intelligent planning method for flexible bone removal path of a robotic arm as described in claim 1, characterized in that, The free path planning of the airborne motion path is implemented using the A* algorithm, which specifically includes: pre-constructing a grid map of the robot's workspace and marking fixed obstacles; using the robot's current real-time pose as the starting point and the next grasping point as the ending point, using the A* algorithm to search for the global shortest collision-free path and generate a sequence of key points for the path.

Images