A fully automated pretreatment method, system, and storage medium for soil samples

Abstract

This invention provides a fully automated pretreatment method, system, and storage medium for soil samples. The method involves acquiring a three-dimensional environmental model of the soil sample pretreatment process, determining a task sequence consisting of a starting point, task points, and target points based on the process flow, and planning a segmented, collision-free path using adjacent point pairs as start and end points. A Gaussian mixture probability model is constructed based on the endpoints, generating sampling points in the probability space to guide random tree growth. The node closest to the sampling point is selected as an expansion node, and multiple alternative directions are generated around the expansion direction. The direction with the lowest energy consumption is selected using a joint energy consumption model. New nodes are generated with a preset step size, and these nodes are added only when the weighted sum of the angular acceleration and angular jerk of the path segment is below a disturbance threshold. When a new node enters the endpoint's attraction domain, an initial path is generated and locally optimized to obtain a segmented path. All segmented paths are then concatenated into a global path, and a robotic arm is controlled to perform the soil pretreatment operation.

Description

Technical Field

[0001] This application belongs to the field of robotic arms, and in particular relates to a fully automated pretreatment method, system and storage medium for soil samples. Background Technology

[0002] The soil sample pretreatment process is currently based on manual operation. Operators may come into contact with potentially hazardous chemical reagents or samples during the process, posing health and safety risks, and the process is also inefficient. Using robotic arms to perform soil sample pretreatment can improve throughput and efficiency, isolate operators from potential hazards, and enhance work safety. However, the pretreatment workspace is filled with various instruments, test tube racks, and reagent bottles, requiring the robotic arm to navigate around all obstacles while performing a series of operations. Currently used path planning algorithms, such as the Rapid Expanding Random Tree (RRT) algorithm based on random sampling, lack guidance in their random exploration approach, resulting in slow convergence and low planning efficiency. The generated initial paths contain redundant nodes and unnecessary inflection points, leading to poor path quality and insufficient smoothness. When executing non-smooth paths, the robotic arm generates significant angular acceleration and jerk, causing vibration and shaking of the end effector, which can easily lead to liquid spillage and affect experimental accuracy. Therefore, how to devise a collision-free path that can quickly plan a smooth, stable, and energy-optimized path has become an urgent technical problem to be solved in order to improve the performance of automated soil sample pretreatment systems. Summary of the Invention

[0003] This invention proposes a fully automated pretreatment method for soil samples to address the problem that existing technologies have failed to achieve a smooth, stable, and energy-optimized collision-free path that can be quickly planned. The method includes: A three-dimensional environmental model of the soil sample pretreatment workspace is obtained, and a task sequence consisting of a starting configuration point, multiple task path points, and a target configuration point is determined according to the preset pretreatment process flow. Using the current configuration point in the task sequence as the starting point and the next configuration point as the ending point, segmented collision-free paths connecting each pair of starting and ending points are planned and generated sequentially. The generation of each segmented collision-free path includes: based on the ending position of the current segmented path, constructing a Gaussian mixture probability model centered at the specified position with probability density decreasing outwards from the center; generating random sampling points within the probability space defined by the model to guide the growth of a random tree; finding the node closest to the random sampling point in the random tree as an expansion node; generating multiple alternative expansion directions around the direction from the expansion node to the random sampling point, and utilizing pre-defined... The established joint energy consumption model calculates the motion energy consumption of each candidate direction and selects the direction with the lowest energy consumption as the expansion direction. A new node is generated along the expansion direction with a preset step size. The new node is added to the random tree only when the motion disturbance value of the path segment between the new node and the expansion node is lower than a preset disturbance threshold. The aforementioned random tree growth and expansion process is repeated. When a new node generated in the random tree enters the attraction domain of the current segment path endpoint, an attempt is made to connect the new node to the endpoint and perform collision detection. If successful, an initial path is obtained, and a local optimization algorithm is applied to the initial path to generate a collision-free path for the segment. Concatenate all generated segmented collision-free paths into a single complete global path; The robotic arm is controlled to perform pretreatment operations on the soil samples along the complete global path.

[0004] Furthermore, the present invention also relates to a fully automated pretreatment system for soil samples, comprising the following modules: The determination module is used to acquire a three-dimensional environmental model of the soil sample pretreatment workspace and, based on the preset pretreatment process, determine a task sequence consisting of a starting configuration point, multiple task path points, and a target configuration point. The generation module is used to plan and generate segmented collision-free paths connecting each pair of start and end points in the task sequence, starting from the current configuration point and ending at the next configuration point. The generation of each segmented collision-free path includes: constructing a Gaussian mixture probability model centered on the current end point of the segmented path, with probability density decreasing outwards from the center; generating random sampling points within the probability space defined by the model to guide the growth of a random tree; finding the node closest to the random sampling point in the random tree as an expansion node; and generating multiple alternative expansion directions around the direction from the expansion node to the random sampling point. The motion energy consumption of each candidate direction is calculated using a pre-built joint energy consumption model, and the direction with the lowest energy consumption is selected as the expansion direction. A new node is generated along the expansion direction with a preset step size. The new node is added to the random tree only when the motion disturbance value of the path segment between the new node and the expansion node is lower than a preset disturbance threshold. The aforementioned random tree growth and expansion process is repeated. When a new node generated in the random tree enters the attraction domain of the current segment path endpoint, an attempt is made to connect the new node to the endpoint and perform collision detection. If successful, an initial path is obtained, and a local optimization algorithm is applied to the initial path to generate a collision-free path for the segment. The stitching module is used to stitch all generated segmented collision-free paths into a complete global path. The execution module controls the robotic arm to perform pretreatment operations on the soil samples along the complete global path.

[0005] This invention guides the growth of a random tree by constructing a Gaussian mixture probability model centered on the endpoint, improving the efficiency and goal orientation of path planning and shortening the planning time. Simultaneously, during the random tree expansion process, a pre-built joint energy consumption model is used to select the expansion direction with the lowest energy consumption, and motion perturbation values ​​calculated based on angular acceleration and angular jerk are used as constraints for the expansion of new nodes. This results in a path that is not only energy-efficient but also smooth, stable, and with minimal impact. Applied to soil sample pretreatment, it ensures the stability and reliability of the robotic arm during grasping, transferring, and weighing operations, avoiding sample spillage, container collisions, or cross-contamination caused by robotic arm vibration or impact. This achieves highly efficient, stable, and high-quality automated pretreatment of soil samples. Attached Figure Description

[0006] Figure 1 A flowchart of the first embodiment; Figure 2 A schematic diagram illustrating the selection of new nodes for motion perturbation values; Figure 3 This is a schematic diagram of the smoothing of the B-spline interpolation path. Detailed Implementation

[0007] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0008] In the first embodiment, the present invention proposes a fully automated pretreatment method for soil samples, such as... Figure 1 ,include: S1. Obtain a three-dimensional environmental model of the soil sample pretreatment workspace, and determine the task sequence consisting of the starting configuration point, multiple task path points and target configuration points according to the preset pretreatment process flow. Using LiDAR or a depth camera, the robotic arm, worktable, sample rack, and centrifuge are scanned to generate point cloud data of the workspace, which is then processed into an octree map as a 3D environment model. The preset soil pretreatment process consists of five stages: weighing, liquid addition, shaking, centrifugation, and liquid transfer. Based on this process, the initial standby posture of the robotic arm is set as the starting configuration point. The poses of a series of key end effectors—sequentially grasping a sample tube, placing it on a balance, grasping a pipette to aspirate the extract, adding liquid to the sample tube, placing the sample tube into the shaker, then into the centrifuge, and aspirating the supernatant into another clean sample tube—are solved using inverse kinematics to obtain the corresponding robotic arm joint angles, which serve as multiple task path points. The standby posture after the task is completed is set as the target configuration point, forming a complete task sequence.

[0009] In an optional embodiment, obtaining a three-dimensional environmental model of the soil sample pretreatment workspace includes: Use an RGB-D camera mounted at the end of a robotic arm to collect point cloud data from multiple different perspectives; Register and fuse multi-view point cloud data to generate a dense point cloud for the workspace; An octree data structure is used to map the dense point cloud and construct the three-dimensional environment model.

[0010] An RGB-D camera, such as an Intel RealSense D435, is mounted on the end effector of a six-axis robotic arm. By controlling the movement of the robotic arm, the camera is moved to multiple preset observation positions around the workspace, for example, scanning the equipment and utensils on the worktable from four different perspectives: directly above, directly in front, to the left front, and to the right front. At each perspective, the camera detects a depth image and a color image, and combines these images into a local point cloud data.

[0011] Since each point cloud exists in its own camera coordinate system, they need to be unified into the same world coordinate system. The forward kinematics of the robotic arm can be used to obtain the camera pose at each viewpoint. This pose information serves as the initial transformation matrix for coarse alignment of the point clouds. The Iterative Closest Point (ICP) algorithm can be used to adjust the relative positions between point clouds, achieving registration. After registration, all point clouds are merged into a single dense point cloud covering the entire workspace. To facilitate rapid querying and collision detection, an octree data structure is used to organize the dense point cloud. The entire space is recursively divided into cubic units, or voxels. Based on whether a voxel contains point cloud data, it is marked as occupied, free, or unknown, thus constructing a multi-resolution 3D environment model, also known as an octree map.

[0012] S2, sequentially using the current configuration point in the task sequence as the starting point and the next configuration point as the ending point, plan and generate segmented collision-free paths connecting each pair of starting and ending points. The generation of each segmented collision-free path includes: based on the ending position of the current segmented path, constructing a Gaussian mixture probability model centered on the current position with probability density decreasing outwards from the center; generating random sampling points within the probability space defined by the model to guide the growth of a random tree; finding the node closest to the random sampling point in the random tree as an expansion node; generating multiple alternative expansion directions around the direction from the expansion node to the random sampling point, and utilizing... The pre-built joint energy consumption model calculates the motion energy consumption of each candidate direction and selects the direction with the lowest energy consumption as the expansion direction. A new node is generated along the expansion direction with a preset step size. The new node is added to the random tree only when the motion disturbance value of the path segment between the new node and the expansion node is lower than a preset disturbance threshold. The aforementioned random tree growth and expansion process is repeated. When a new node generated in the random tree enters the attraction domain of the current segment path endpoint, an attempt is made to connect the new node to the endpoint and perform collision detection. If successful, an initial path is obtained, and a local optimization algorithm is applied to the initial path to generate a collision-free path for the segment. For the path segment from the oscillator picking up the sample tube to the centrifuge door, the robotic arm joint angle corresponding to the endpoint pose of the centrifuge door is used as the principal center of the Gaussian mixture model, and three Gaussian distributions with large variance and small weights are set around it to obtain a multimodal probabilistic model. During sampling, a random joint angle is extracted from this model as a sampling point, so that most sampling points are concentrated near the endpoint, thereby accelerating the growth of the random tree in the target direction.

[0013] In the random tree, the Euclidean distance between each node and the previously generated random sampling point in the six-dimensional joint space is calculated, and the node with the smallest distance is selected as the expansion node. Using the vector pointing from the expansion node to the random sampling point as a reference, ten alternative expansion directions with small included angles are randomly generated in the up, down, left, and right directions. Based on a pre-established energy consumption model based on the square integral of joint torque, the theoretical energy consumption required for the robotic arm to move one small step along each of the ten alternative directions is calculated, and the direction with the lowest calculated energy consumption is selected as the direction for this expansion. In specific implementation, by loading the robot's URDF model, the RNEA (Recursive Newton-Euler Algorithm) is called to calculate the torque required for each joint in the current motion state. Combined with the joint velocity and motor resistivity, the sum of mechanical work and heat loss is calculated. The torque is calculated using pin.rnea(model,data,q,v,a), where model is the robot model, data is the robot data, q is the current joint angle, v is the current joint angular velocity, and a is the current joint angular acceleration.

[0014] Following the expansion direction selected in the previous step, increase the joint angle values ​​of each node of the expansion node by a fixed step size, such as 0.05 radians, thereby generating a new node. By performing differential calculations on the expansion node, the path midpoint, and the new node, the angular acceleration and angular jerk of the path segment are estimated. According to the formula, the perturbation value equals 0.6 × the modulus of angular acceleration + 0.4 × the modulus of angular jerk. If the calculated perturbation value is less than a preset threshold, such as 1.5, the motion of this segment is considered smooth, and the new node and its connection edge to the expansion node are added to the random tree, such as... Figure 2 Otherwise, the new node is abandoned. In another embodiment, the motion disturbance value is the weighted sum of the squares of the instantaneous acceleration values ​​of each joint of the robotic arm during the movement from the extended node to the new node, or the rate of change of the deviation of the direction of the synthesized acceleration of the end effector relative to the direction of gravity.

[0015] The process involves continuous sampling, finding nearest neighbors, expanding in low-energy directions, identifying motion disturbances, and adding nodes in a loop. When the joint space distance between a newly generated node and the endpoint is less than 0.2 times the preset attraction field radius, random expansion stops, and an attempt is made to connect the new node to the endpoint. The connection method involves linear interpolation between the two points to generate a straight path segment consisting of 20 intermediate points. Each of these 20 intermediate points is checked to see if its pose collides with the 3D environment model. If all intermediate points are free of collisions, the connection is successful, and an initial path from the starting point to the endpoint has been found. This initial path is then optimized using a shortcut path, for example, by randomly selecting two non-adjacent points on the path and attempting to connect them. If the new path segment is free of collisions, the original part is replaced with this shortcut path. This process is repeated, for example, 500 times, to obtain a shorter and smoother path.

[0016] In an optional embodiment, the step of constructing a Gaussian mixture probability model centered on the endpoint position of the current segmented path, with the probability density decreasing outwards from the center, includes: Multiple Gaussian components are set, with the mean of one Gaussian component set as the endpoint of the current segment path, and the mean of the remaining Gaussian components distributed around the endpoint. The covariance of each Gaussian component is set such that the covariance of the central Gaussian component is smaller than the covariance of the surrounding Gaussian components, forming a sampling probability distribution that converges towards the endpoint.

[0017] The structure of a Gaussian mixture model consists of a weighted sum of multiple Gaussian distributions, with the probability density function being... , where K is the number of Gaussian components, for example, set to 5. It is the weight of the i-th component. It is a mean vector. It is the covariance matrix. The model is used to generate random points in the sampling space, guiding the path planning tree to grow towards the target.

[0018] Assume the three-dimensional coordinates of the endpoint of the current segmented path are The mean of the first Gaussian component. Set as The mean of the remaining four Gaussian components arrive Then symmetrically distributed in The surrounding preset locations, such as positions 0.1m apart on the xy-plane. To ensure the sampling points appear densely near the endpoint, the covariance matrix of the central Gaussian component... The values ​​are set relatively small, for example, the diagonal elements of a diagonal matrix are 0.01, while the covariance matrix of the surrounding components is... arrive The weights are set relatively high, with diagonal elements at 0.05. Meanwhile, the weight of the center component is also set relatively high. The weight is set to 0.6, while the weights of the other four components are each 0.1. Points sampled from this Gaussian mixture model have a high probability of falling within the region of the central Gaussian distribution, thus guiding the search process toward the target endpoint.

[0019] In an optional embodiment, the step of calculating the motion energy consumption of each candidate direction using a pre-built joint energy consumption model and selecting the direction with the lowest energy consumption as the expansion direction includes: Based on the direction from the extended node to the random sampling point, multiple alternative extended directions are generated within a preset angle range centered on the reference. The energy consumption value of each candidate expansion direction is calculated using the energy consumption model, and the direction with the minimum energy consumption value is selected as the expansion direction.

[0020] When the path planning algorithm selects an extended node and a random sampling point Then, a reference direction vector is determined, which is formed by... point to Using this reference direction as the central axis, define a conical space, for example, a cone with a vertex angle of 30°. Within this conical space, generate a set of candidate direction vectors uniformly or randomly, for example, generating 10 candidate directions. Each candidate direction represents a random tree node. One possible direction for expansion.

[0021] For each of the 10 alternative expansion directions, from node Extending this direction by a fixed step size yields a potential new node. For from arrive For each short path segment, the motion energy consumption is calculated using a pre-established robotic arm joint energy consumption model. This energy consumption model might be a function based on joint velocity and required torque, for example... The energy consumption values ​​of the 10 candidate path segments are calculated, for example, a set of data: 1.5J, 1.2J, and 1.8J. These energy consumption values ​​are compared, and the one with the smallest value, such as 1.2J, is selected as the expansion direction to generate new nodes and edges in the tree. In an alternative embodiment, the N directions with the lowest energy consumption are selected, and then a direction is randomly chosen from these N directions to increase the randomness of the expansion direction and avoid getting trapped in local optima.

[0022] In an optional embodiment, the motion disturbance value is calculated based on a weighted sum of the angular acceleration and angular jerk of the path segment, specifically: For the path segment connecting the extended node and the new node, the motion disturbance value S is calculated by the following formula: ; Where T is the motion time of the path segment, and q(t) is the joint angle vector of the robotic arm. and These are the joint angular acceleration and angular jerk vectors, respectively. and These are the preset non-negative weighting coefficients. This represents the Euclidean norm of a vector.

[0023] For a connection extension node and new nodes For the path segments, a method such as fifth-order polynomial interpolation is used to generate time-varying joint angle vector functions. This function represents the smooth process of all joints of the robotic arm moving from the initial configuration to the final configuration. The total motion time T is determined based on the length of the path segment and the maximum joint speed and acceleration limits of the robotic arm, for example, set to 1.5s.

[0024] Once the joint trajectory function q(t) is determined, the joint angular acceleration vector can be obtained by differentiating the function. Angular jerk, i.e., jerk vector The parsing expression. Set the weighting coefficients, for example, by... Set to 1.0. Setting it to 0.5 indicates different emphases on acceleration and jerk smoothness. The product function is applied over the time interval [0, T]. Numerical integration is performed, for example using the trapezoidal rule or Simpson's rule, to approximate the integration result by evaluating and summing the results at multiple discrete time points. The resulting scalar value S represents the smoothness of motion of the path segment; the smaller the value, the less vibration and impact during the motion.

[0025] In an optional embodiment, applying a local optimization algorithm to the initial path to generate the collision-free path for the segments includes: The initial path is smoothed using the B-spline interpolation algorithm to generate a B-spline curve; Dense sampling is performed along the B-spline curve, and collision detection is performed at each sampling point; If all sampling points are collision-free, then the B-spline curve is used as the collision-free path of the segment.

[0026] The initial path generated by the basic path planning algorithm is usually a series of discrete path points connected by straight line segments. , , , To enable smooth operation of the robotic arm, the path points are used as control points, and a B-spline interpolation algorithm is applied to generate a continuous and smooth B-spline curve, such as... Figure 3 For example, using cubic B-spline curves can ensure the curvature continuity of the path, thereby eliminating sharp corners in the initial path and generating a kinematically superior candidate path.

[0027] While the generated B-spline curve is smooth, it may cross obstacles. To ensure path safety, collision detection is required. Dense discrete sampling is performed along the B-spline curve. For example, for a curve of 0.5m length, sampling can be performed every 0.01m, resulting in 50 sampling points. For each sampling point, the complete 3D configuration of the robotic arm is calculated and collided with the pre-built environment model. If the robotic arm configuration corresponding to all 50 sampling points does not collide with the environment, the B-spline curve is considered collision-free, and the curve is accepted as the path for that segment. If a collision is detected at any sampling point, the optimization fails, and the B-spline curve needs adjustment or the path needs to be replanned.

[0028] S3, concatenate all generated segmented collision-free paths into a complete global path; The segmented collision-free paths generated for all task path points in the preceding steps of weighing, adding liquid, and oscillation are concatenated end-to-end according to the chronological order of the tasks. For example, the path from the starting point to the weighing point, the path from the weighing point to the adding liquid point, and all path segments are sequentially connected to obtain a single, continuous sequence of nodes containing all motion commands from the start to the end of the task, which is the complete global path.

[0029] S4, control the robotic arm to perform soil sample pretreatment operations along the complete global path.

[0030] A series of joint angle nodes in the complete global path are sent to the underlying motion controller of the robotic arm. The controller uses a fifth-order polynomial interpolation algorithm to smooth the path between adjacent nodes, generating time-continuous joint position, velocity, and acceleration commands. Based on these commands, the controller drives the servo motors of the six joints, and simultaneously provides real-time feedback on the actual position through the encoders of each joint, forming a closed-loop control. This guides the end effector of the robotic arm to move along the planned path, completing a series of soil sample pretreatment operations such as grasping, placing, and transferring liquid.

[0031] In an optional embodiment, the controlled robotic arm performs soil sample pretreatment operations along the complete global path, including: The robotic arm is controlled to perform weighing, adding a preset volume of extract, vortexing, and centrifugation.

[0032] A robotic arm picks up an empty centrifuge tube and places it on an electronic balance with an accuracy of 0.001g, reading and recording the tare weight. The robotic arm removes the tube from the balance, moves it to a soil sample container, adds a small amount of soil sample to the tube using an end effector, and then returns it to the balance. The computer compares the current reading with a target weight, such as 2.000g, and controls the robotic arm to repeat the sample addition operation until the sample weight is within ±0.01g of the target value.

[0033] After weighing, the robotic arm moves the centrifuge tube containing soil to the liquid workstation. According to a pre-programmed procedure, a pipette pump connected to the robotic arm system injects 10.0 mL of extraction solution into the tube. The robotic arm caps the centrifuge tube and inserts it into a vortex mixer, vortexing at 2500 rpm for 60 seconds to ensure thorough mixing of the sample and liquid. After vortexing, the robotic arm removes the centrifuge tube and places it in a centrifuge. The centrifuge runs at 8000 rpm for 5 minutes to allow solid residue to settle at the bottom of the tube. The robotic arm removes the centrifuge tube from the centrifuge; the supernatant in the tube is the processed sample extract.

[0034] In a second embodiment, the present invention also proposes a fully automated pretreatment system for soil samples, comprising the following modules: The determination module is used to acquire a three-dimensional environmental model of the soil sample pretreatment workspace and, based on the preset pretreatment process, determine a task sequence consisting of a starting configuration point, multiple task path points, and a target configuration point. The generation module is used to plan and generate segmented collision-free paths connecting each pair of start and end points in the task sequence, starting from the current configuration point and ending at the next configuration point. The generation of each segmented collision-free path includes: constructing a Gaussian mixture probability model centered on the current end point of the segmented path, with probability density decreasing outwards from the center; generating random sampling points within the probability space defined by the model to guide the growth of a random tree; finding the node closest to the random sampling point in the random tree as an expansion node; and generating multiple alternative expansion directions around the direction from the expansion node to the random sampling point. The motion energy consumption of each candidate direction is calculated using a pre-built joint energy consumption model, and the direction with the lowest energy consumption is selected as the expansion direction. A new node is generated along the expansion direction with a preset step size. The new node is added to the random tree only when the motion disturbance value of the path segment between the new node and the expansion node is lower than a preset disturbance threshold. The aforementioned random tree growth and expansion process is repeated. When a new node generated in the random tree enters the attraction domain of the current segment path endpoint, an attempt is made to connect the new node to the endpoint and perform collision detection. If successful, an initial path is obtained, and a local optimization algorithm is applied to the initial path to generate a collision-free path for the segment. The stitching module is used to stitch all generated segmented collision-free paths into a complete global path. The execution module controls the robotic arm to perform pretreatment operations on the soil samples along the complete global path.

[0035] In an optional embodiment, the step of constructing a Gaussian mixture probability model centered on the endpoint position of the current segmented path, with the probability density decreasing outwards from the center, includes: Multiple Gaussian components are set, with the mean of one Gaussian component set as the endpoint of the current segment path, and the mean of the remaining Gaussian components distributed around the endpoint. The covariance of each Gaussian component is set such that the covariance of the central Gaussian component is smaller than the covariance of the surrounding Gaussian components, forming a sampling probability distribution that converges towards the endpoint.

[0036] In an optional embodiment, the step of calculating the motion energy consumption of each candidate direction using a pre-built joint energy consumption model and selecting the direction with the lowest energy consumption as the expansion direction includes: Based on the direction from the extended node to the random sampling point, multiple alternative extended directions are generated within a preset angle range centered on the reference. The energy consumption value of each candidate expansion direction is calculated using the energy consumption model, and the direction with the minimum energy consumption value is selected as the expansion direction.

[0037] In an optional embodiment, the motion disturbance value is calculated based on a weighted sum of the angular acceleration and angular jerk of the path segment, specifically: For the path segment connecting the extended node and the new node, the motion disturbance value S is calculated by the following formula: ; Where T is the motion time of the path segment, and q(t) is the joint angle vector of the robotic arm. and These are the joint angular acceleration and angular jerk vectors, respectively. and These are the preset non-negative weighting coefficients. This represents the Euclidean norm of a vector.

[0038] In an optional embodiment, applying a local optimization algorithm to the initial path to generate the collision-free path for the segments includes: The initial path is smoothed using the B-spline interpolation algorithm to generate a B-spline curve; Dense sampling is performed along the B-spline curve, and collision detection is performed at each sampling point; If all sampling points are collision-free, then the B-spline curve is used as the collision-free path of the segment.

[0039] In an optional embodiment, obtaining a three-dimensional environmental model of the soil sample pretreatment workspace includes: Use an RGB-D camera mounted at the end of a robotic arm to collect point cloud data from multiple different perspectives; Register and fuse multi-view point cloud data to generate a dense point cloud for the workspace; An octree data structure is used to map the dense point cloud and construct the three-dimensional environment model.

[0040] In an optional embodiment, the controlled robotic arm performs soil sample pretreatment operations along the complete global path, including: The robotic arm is controlled to perform weighing, adding a preset volume of extract, vortexing, and centrifugation.

[0041] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A fully automated pretreatment method for soil samples, characterized in that, Includes the following steps: A three-dimensional environmental model of the soil sample pretreatment workspace is obtained, and a task sequence consisting of a starting configuration point, multiple task path points, and a target configuration point is determined according to the preset pretreatment process flow. Using the current configuration point in the task sequence as the starting point and the next configuration point as the ending point, segmented collision-free paths connecting each pair of starting and ending points are planned and generated sequentially. The generation of each segmented collision-free path includes: based on the ending position of the current segmented path, constructing a Gaussian mixture probability model centered at the specified position with probability density decreasing outwards from the center; generating random sampling points within the probability space defined by the model to guide the growth of a random tree; finding the node closest to the random sampling point in the random tree as an expansion node; generating multiple alternative expansion directions around the direction from the expansion node to the random sampling point, and utilizing pre-defined... The established joint energy consumption model calculates the motion energy consumption of each candidate direction and selects the direction with the lowest energy consumption as the expansion direction. A new node is generated along the expansion direction with a preset step size. The new node is added to the random tree only when the motion disturbance value of the path segment between the new node and the expansion node is lower than a preset disturbance threshold. The aforementioned random tree growth and expansion process is repeated. When a new node generated in the random tree enters the attraction domain of the current segment path endpoint, an attempt is made to connect the new node to the endpoint and perform collision detection. If successful, an initial path is obtained, and a local optimization algorithm is applied to the initial path to generate a collision-free path for the segment. Concatenate all generated segmented collision-free paths into a single complete global path; The robotic arm is controlled to perform pretreatment operations on the soil samples along the complete global path.

2. The method according to claim 1, characterized in that, The step of constructing a Gaussian mixture probability model centered at the endpoint of the current segmented path, with probability density decreasing outwards from the endpoint, includes: Multiple Gaussian components are set, with the mean of one Gaussian component set as the endpoint of the current segment path, and the mean of the remaining Gaussian components distributed around the endpoint. The covariance of each Gaussian component is set such that the covariance of the central Gaussian component is smaller than the covariance of the surrounding Gaussian components, forming a sampling probability distribution that converges towards the endpoint.

3. The method according to claim 1, characterized in that, The process of calculating the motion energy consumption of each candidate direction using a pre-built joint energy consumption model and selecting the direction with the lowest energy consumption as the expansion direction includes: Based on the direction from the extended node to the random sampling point, multiple alternative extended directions are generated within a preset angle range centered on the reference. The energy consumption value of each candidate expansion direction is calculated using the energy consumption model, and the direction with the minimum energy consumption value is selected as the expansion direction.

4. The method according to claim 1, characterized in that, The motion disturbance value is calculated based on the weighted sum of the angular acceleration and angular jerk of the path segment, specifically: For the path segment connecting the extended node and the new node, the motion disturbance value S is calculated by the following formula: ； Where T is the motion time of the path segment, and q(t) is the joint angle vector of the robotic arm. and These are the joint angular acceleration and angular jerk vectors, respectively. and These are the preset non-negative weighting coefficients. This represents the Euclidean norm of a vector.

5. The method according to claim 1, characterized in that, The step of applying a local optimization algorithm to the initial path to generate the segmented collision-free path includes: The initial path is smoothed using the B-spline interpolation algorithm to generate a B-spline curve; Dense sampling is performed along the B-spline curve, and collision detection is performed at each sampling point; If all sampling points are collision-free, then the B-spline curve is used as the collision-free path of the segment.

6. The method according to claim 1, characterized in that, The three-dimensional environment model of the soil sample pretreatment workspace includes: Use an RGB-D camera mounted at the end of a robotic arm to collect point cloud data from multiple different perspectives; Register and fuse multi-view point cloud data to generate a dense point cloud for the workspace; An octree data structure is used to map the dense point cloud and construct the three-dimensional environment model.

7. The method according to any one of claims 1-6, characterized in that, The controlled robotic arm performs soil sample pretreatment operations along the complete global path, including: The robotic arm is controlled to perform weighing, adding a preset volume of extract, vortexing, and centrifugation.

8. A fully automated pretreatment system for soil samples, characterized in that, Includes the following modules: The determination module is used to acquire a three-dimensional environmental model of the soil sample pretreatment workspace and, based on the preset pretreatment process, determine a task sequence consisting of a starting configuration point, multiple task path points, and a target configuration point. The generation module is used to plan and generate segmented collision-free paths connecting each pair of start and end points in the task sequence, starting from the current configuration point and ending at the next configuration point. The generation of each segmented collision-free path includes: constructing a Gaussian mixture probability model centered on the current end point of the segmented path, with probability density decreasing outwards from the center; generating random sampling points within the probability space defined by the model to guide the growth of a random tree; finding the node closest to the random sampling point in the random tree as an expansion node; and generating multiple alternative expansion directions around the direction from the expansion node to the random sampling point. The motion energy consumption of each candidate direction is calculated using a pre-built joint energy consumption model, and the direction with the lowest energy consumption is selected as the expansion direction. A new node is generated along the expansion direction with a preset step size. The new node is added to the random tree only when the motion disturbance value of the path segment between the new node and the expansion node is lower than a preset disturbance threshold. The aforementioned random tree growth and expansion process is repeated. When a new node generated in the random tree enters the attraction domain of the current segment path endpoint, an attempt is made to connect the new node to the endpoint and perform collision detection. If successful, an initial path is obtained, and a local optimization algorithm is applied to the initial path to generate a collision-free path for the segment. The stitching module is used to stitch all generated segmented collision-free paths into a complete global path. The execution module controls the robotic arm to perform pretreatment operations on the soil samples along the complete global path.

9. The system according to claim 8, characterized in that, The step of constructing a Gaussian mixture probability model centered at the endpoint of the current segmented path, with probability density decreasing outwards from the endpoint, includes: Multiple Gaussian components are set, with the mean of one Gaussian component set as the endpoint of the current segment path, and the mean of the remaining Gaussian components distributed around the endpoint. The covariance of each Gaussian component is set such that the covariance of the central Gaussian component is smaller than the covariance of the surrounding Gaussian components, forming a sampling probability distribution that converges towards the endpoint.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when executed by a processor, implements the method as described in any one of claims 1-7.

Images