Intelligent cotton spinning mechanical hand autonomous mobile operation control system and method

Abstract

The application belongs to the technical field of textile automation, and discloses an intelligent cotton spinning manipulator autonomous mobile operation control system and method; firstly, the spatiotemporal dynamic characteristics of point cloud clusters are extracted to realize accurate identification of flying interference and confidence obstacles; the interference clusters are mapped to a temporary observation buffer layer with a confidence decay mechanism, and are superimposed with a static map to construct a hybrid cost grid. In the path planning process, the system dynamically replans in real time according to the confidence evolution curve: if the interference disappears, it is smoothly passed through, and if the interference continues, it is upgraded to an entity obstacle and avoidance and stopping are performed. In addition, the application also covers envelope dynamic adjustment based on the posture of the manipulator and a sensing compensation mechanism for operation air flow disturbance. The application enhances the sensing robustness and operation continuity of the mobile manipulator in a complex dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of textile automation technology, and more specifically, to an intelligent cotton spinning robot autonomous mobile operation control system and method. Background Technology

[0002] In modern high-speed cotton spinning workshops, the autonomous navigation of mobile robotic arms faces severe environmental robustness challenges. Due to the characteristics of the production process, the workshop air is constantly filled with high concentrations of cotton lint and short fibers (commonly known as "flying fluff"), which float and accumulate irregularly with the airflow. During detection, traditional sensing systems such as lidar are easily reflected by these low-density, easily scattered clusters of flying fluff, generating a large number of diffuse and randomly distributed discrete noise points. This makes it difficult for sensing algorithms to distinguish the "illusory" suspended fibers from the "real" production equipment or workers based on their physical characteristics. In the SLAM (Simultaneous Localization and Mapping) process, this interference directly manifests as "burrs" and "ghosting" that are difficult to remove from the static map. In actual operation scenarios, the frequent acceleration and deceleration of the robotic arm and the grasping action of the robotic arm can induce violent fluctuations in local airflow, which in turn agitate the ground and create a secondary interference loop. This can trigger frequent braking of the control system or lead to an endless local path replanning dilemma, disrupting the continuity of the production process. As a result, the automatic navigation system faces perception failure and decision paralysis in the complex physical environment of cotton spinning.

[0003] In view of this, the present invention proposes an intelligent cotton spinning robot autonomous mobile operation control system and method to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent cotton spinning robot autonomous movement operation control method, comprising:

[0005] Step S1: Obtain the original three-dimensional point cloud data of the environment where the cotton spinning robot is located, perform spatial clustering processing on the original three-dimensional point cloud data to extract several discrete point cloud clusters, and calculate the spatiotemporal dynamic feature vector of each discrete point cloud cluster. The spatiotemporal dynamic feature vector includes the voxel density fluctuation rate between consecutive frames, the vertical axis suspension offset, and the bottom support connectivity index.

[0006] Step S2: Based on the spatiotemporal dynamic feature vector, perform fly flower interference identification on each discrete point cloud cluster to obtain floating interference candidate clusters and confidence obstacle clusters;

[0007] Step S3: The confidence obstacle clusters are directly written into the static environment map used for navigation, and the floating interference candidate clusters are mapped to the temporary observation buffer layer associated with the static environment map, wherein the temporary observation buffer layer is configured with a confidence coefficient that decays over time.

[0008] Step S4: When the cotton spinning robot receives the moving operation instruction, a hybrid cost grid is constructed based on the static environment map and the temporary observation buffer layer. The hybrid cost grid is formed by superimposing the fixed cost value in the static environment map and the floating cost value in the temporary observation buffer layer after weighting by the confidence coefficient.

[0009] Step S5: Path search is performed based on the hybrid cost grid to generate an initial operation path, and the confidence coefficient change of the suspended interference candidate clusters in the temporary observation buffer layer is monitored in real time as the cotton spinning robot moves along the initial operation path.

[0010] Step S6: When the confidence coefficient of the suspended interference candidate cluster decays to below the safety threshold, the suspended interference candidate cluster is removed from the temporary observation buffer layer, and local path replanning is triggered for the area where the suspended interference candidate cluster is located, so as to control the cotton spinning robot to smoothly pass through the area where the original suspended interference candidate cluster is located.

[0011] Step S7: When the confidence coefficient of the suspended interference candidate cluster continues to be higher than the safety threshold and is accompanied by spatial position convergence, the suspended interference candidate cluster is promoted to a confidence obstacle cluster and written into the static environment map, while controlling the cotton spinning robot to perform avoidance operation.

[0012] An autonomous mobile operation control system for an intelligent cotton spinning robot, used to implement the aforementioned autonomous mobile operation control method for the intelligent cotton spinning robot, includes:

[0013] The 3D perception module is used to acquire the original 3D point cloud data of the environment in which the cotton spinning robot is located;

[0014] The point cloud processing module is used to perform spatial clustering processing on the original 3D point cloud data to extract several discrete point cloud clusters, and to calculate the spatiotemporal dynamic feature vector of each discrete point cloud cluster.

[0015] The interference identification module is used to perform fly flower interference identification on each discrete point cloud cluster based on spatiotemporal dynamic feature vectors, and obtain floating interference candidate clusters and confidence obstacle clusters;

[0016] The map management module is used to directly write the clusters of believed obstacles into the static environment map used for navigation, and to map the candidate clusters of floating interference to a temporary observation buffer layer associated with the static environment map.

[0017] The cost grid construction module is used to construct a hybrid cost grid based on a static environment map and a temporary observation buffer layer when the cotton spinning robot receives a mobile operation instruction.

[0018] The path planning module is used to perform path search based on the hybrid cost grid to generate an initial job path;

[0019] The real-time monitoring module is used to monitor the changes in the confidence coefficient of suspended interference candidate clusters in the temporary observation buffer layer in real time as the cotton spinning robot moves along the initial working path.

[0020] The dynamic replanning module is used to remove the floating interference candidate cluster from the temporary observation buffer layer when the confidence coefficient of the floating interference candidate cluster decays to below the safety threshold, and to trigger local path replanning for the region where the floating interference candidate cluster is located.

[0021] The motion control module is used to control the cotton spinning robot to smoothly pass through the area where the original suspended interference candidate cluster is located, and when the confidence coefficient of the suspended interference candidate cluster is continuously higher than the safety threshold and accompanied by spatial position convergence, the suspended interference candidate cluster is promoted to a confidence obstacle cluster and written into the static environment map, while controlling the cotton spinning robot to perform avoidance and stopping operations.

[0022] The data storage module is used to store static environment maps, temporary observation buffer layers, and related control parameters.

[0023] The technical effects and advantages of the intelligent cotton spinning robot autonomous mobile operation control system and method of this invention are as follows:

[0024] This invention enhances the perception robustness and decision-making consistency of the operating platform in extreme dynamic environments. By accurately identifying and eliminating instantaneous suspension interference, it effectively avoids unnecessary braking, ensuring deep continuity of the production process. Its advantage lies in constructing an evaluation mechanism with evolutionary insight, achieving a dynamic balance between traffic efficiency and obstacle avoidance safety, and significantly improving path smoothness and space utilization under complex working conditions. Simultaneously, combined with collaborative compensation for environmental fluctuations induced by its own operational actions, this invention possesses adaptive capabilities, optimizing the flexibility of task execution and overall operational efficiency in high-dust environments. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the intelligent cotton spinning robot's autonomous movement operation control method of the present invention;

[0026] Figure 2 This is a schematic diagram of the intelligent cotton spinning robot autonomous mobile operation control system of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1 In this embodiment of the invention, the specific implementation process of the intelligent cotton spinning robot's autonomous movement operation control method includes:

[0029] Step S1: Acquire the original 3D point cloud data of the environment surrounding the cotton spinning robot. Perform spatial clustering on the original 3D point cloud data to extract several discrete point cloud clusters, and calculate the spatiotemporal dynamic feature vector of each discrete point cloud cluster. The spatiotemporal dynamic feature vector includes the voxel density fluctuation rate between consecutive frames, the vertical axis suspension offset, and the bottom support connectivity index. The original 3D point cloud data records the spatial geometric information of the environment surrounding the cotton spinning robot, including the 3D coordinates of various objects such as fixed obstacles, equipment structures, and airborne debris. Spatial clustering divides the point cloud into spatially continuous clusters through neighborhood density analysis, with each discrete point cloud cluster representing a potential independent physical entity. The spatiotemporal dynamic feature vector is the key basis for distinguishing between airborne debris interference and real obstacles. The voxel density fluctuation rate reflects the stability characteristics of the object in the time dimension, the vertical axis suspension offset characterizes the spatial relationship between the object and the ground, and the bottom support connectivity index quantifies the structural integrity of the object's bottom. These data provide multi-dimensional physical feature information for subsequent intelligent recognition, ensuring accurate differentiation between dynamic interference and static obstacles in complex cotton spinning environments.

[0030] Step S2 involves identifying cotton fluff interference based on spatiotemporal dynamic feature vectors for each discrete point cloud cluster, resulting in candidate clusters of suspended interference and clusters of believed obstacles. Cotton fluff interference identification is a core technology in cotton textile environmental perception, using multi-feature fusion to solve the problem of traditional navigation systems being unable to distinguish between floating debris and physical obstacles. The identification process is based on multi-dimensional threshold determination using spatiotemporal dynamic feature vectors. Point cloud clusters with high volatility, suspension characteristics, and weak bottom support are marked as candidate clusters of suspended interference; these clusters typically correspond to cotton fiber fluff floating in the air. Conversely, point cloud clusters with spatiotemporal stability, ground contact characteristics, and continuous bottom support are marked as believed obstacle clusters, representing real physical obstacles that need to be avoided. This intelligent classification strategy based on physical characteristics significantly reduces the misleading impact of cotton fluff on the navigation system, improving the accuracy of path planning and traffic efficiency.

[0031] Step S3 involves directly writing the confident obstacle clusters into the static environment map used for navigation, and mapping the suspended interference candidate clusters to a temporary observation buffer layer associated with the static environment map. This temporary observation buffer layer is configured with a confidence coefficient that decays over time. The static environment map stores confirmed persistent obstacle information, serving as the basic constraint layer for path planning and ensuring the robot avoids real obstacles such as fixed equipment and walls. The temporary observation buffer layer is an innovative mechanism for handling uncertain interference. It manages unverified suspended objects through an independent data layer, preventing the misidentification of transient debris as permanent obstacles and thus avoiding map contamination. The time decay mechanism of the confidence coefficients reflects Bayesian inference; as observation time progresses, if a suspended interference candidate cluster is not repeatedly observed, its confidence gradually decreases, and it is automatically removed from the navigation constraints. This two-layer map architecture ensures reliable avoidance of real obstacles while avoiding overly conservative responses to transient debris, achieving an adaptive dynamic balance for the navigation system.

[0032] In step S4, when the cotton spinning robot receives a movement command, a hybrid cost grid is constructed based on a static environment map and a temporary observation buffer layer. This hybrid cost grid is formed by superimposing the fixed cost value from the static environment map and the floating cost value from the temporary observation buffer layer, weighted by a confidence coefficient. The hybrid cost grid serves as the foundation for path planning, integrating deterministic and uncertain information. It uniformly expresses the spatial occupancy status of different confidence levels through a cost superposition mechanism. The fixed cost value corresponds to verified static obstacles, assigned a high cost for forced avoidance; the floating cost value reflects the dynamic threat level of suspended interference candidate clusters, with the cost adjusted by the confidence coefficient. High confidence levels approach fixed obstacle treatment, while low confidence levels approach free space. This cost superposition strategy transforms the binary occupancy grid into a continuous probability grid, enabling the path planning algorithm to perform risk assessment and path optimization based on obstacle confidence, achieving an intelligent trade-off between safety and efficiency.

[0033] Step S5 involves a path search based on a hybrid cost grid to generate an initial work path. As the cotton spinning robot moves along this initial path, the confidence coefficient changes of suspended interference candidate clusters in the temporary observation buffer layer are monitored in real time. The initial work path planning employs a heuristic search algorithm to find the globally optimal or suboptimal path on the hybrid cost grid, comprehensively considering path length, turning costs, and the risk costs of passing through different confidence regions. The real-time monitoring mechanism dynamically tracks the state evolution of suspended interference candidate clusters through continuous perception and map updates, including the time decay of confidence, confidence enhancement caused by new observations, and spatial drift characteristics. This dynamic monitoring provides real-time data support for subsequent adaptive path adjustments, ensuring the robot can respond to dynamic environmental changes, adjust the path promptly to improve efficiency after the fly debris dissipates, and avoid it in a timely manner to ensure safety after the fly debris has accumulated and stabilized.

[0034] Step S6: When the confidence coefficient of a suspended interference candidate cluster decays below the safety threshold, the candidate cluster is removed from the temporary observation buffer layer, and local path replanning for the area where the candidate cluster is located is triggered to control the cotton spinning robot to smoothly pass through the original area where the candidate cluster was located. A confidence coefficient decaying to the safety threshold indicates that the obstacle threat in the spatial area has been eliminated and safe passage is possible. The removal operation clears the corresponding grid cell from the temporary observation buffer layer, reducing the mixed cost of the area and creating conditions for path optimization. Local path replanning only optimizes the affected path segments, significantly reducing computational overhead and improving response speed compared to global replanning. Smooth passage control ensures the curvature continuity of the path through methods such as spline curve interpolation, avoiding sharp turns or sudden speed changes, and guaranteeing the stability and quality of the robot's movement. This adaptive adjustment mechanism fully utilizes dynamic environmental information, maximizing passage efficiency while ensuring safety.

[0035] Step S7: When the confidence coefficient of a candidate cluster of suspended interference remains above the safety threshold and is accompanied by spatial position convergence, the candidate cluster is upgraded to a confident obstacle cluster and written into the static environment map. Simultaneously, the cotton spinning robot is controlled to perform an obstacle avoidance operation. A consistently high confidence coefficient indicates that the object has been repeatedly observed and verified, and is not a transient flyer but a persistent entity. Spatial position convergence means that the object's spatial coordinates are stable within a fixed range, lacking the random drift characteristics of flyers, further verifying its entity obstacle attribute. Upgrading it to a confident obstacle cluster and writing it into the static map achieves a state transition from uncertainty management to deterministic constraints, ensuring that the obstacle is continuously avoided in subsequent planning. The obstacle avoidance operation embodies a safety-first strategy; when a newly confirmed entity obstacle appears on the path, progress is immediately halted and a new detour path is planned to avoid collision risks. This bidirectional state transition mechanism enables the navigation system to possess self-learning and adaptive capabilities, continuously optimizing the environmental perception model in dynamic environments.

[0036] In this embodiment of the invention, step 1 above may include:

[0037] Raw 3D point cloud data for the current frame is acquired using a LiDAR system mounted on a cotton spinning robot. This raw 3D point cloud data undergoes ground stripping and downsampling processing to obtain the target point cloud data. LiDAR acquisition obtains the 3D coordinates of spatial points by emitting laser beams and measuring reflection time, forming a dense point cloud dataset. Ground stripping is a preprocessing step that removes ground points. Points belonging to the ground are identified and deleted using height threshold filtering or plane fitting methods, reducing the amount of data for subsequent processing and eliminating interference from the ground in obstacle detection. Downsampling uses voxel grid filtering or random sampling methods to reduce point cloud density while maintaining the spatial distribution characteristics of the point cloud, significantly reducing computational load. The voxel size for downsampling is typically set to 3-5 cm, striking a balance between computational efficiency and feature preservation. The target point cloud data is a high-quality dataset that has been preprocessed and optimized, removing redundant information and interference components, providing clear input for subsequent clustering analysis.

[0038] A clustering algorithm based on Euclidean distance is used to grow regions from target point cloud data, extracting spatially continuous discrete point cloud clusters. Euclidean clustering, based on the principle of spatial proximity, groups points with a spatial distance less than a threshold into the same cluster, effectively identifying spatially separated independent objects. The region growing process starts from a seed point, iteratively adding neighboring points that meet the distance condition to the current cluster until no new points can be added. Then, a new seed is selected from the remaining points to continue growing new clusters. The choice of the distance threshold is crucial, typically set to 5-10 cm. Too small a threshold leads to over-segmentation of objects, while too large a threshold leads to incorrect merging of adjacent objects. The clustering result is several spatially continuous discrete point cloud clusters, each representing a potential independent physical object, including equipment parts, fly ash clumps, or other obstacles. This geometrically continuous clustering method is simple and efficient, suitable for real-time processing of large-scale point cloud data.

[0039] For any discrete point cloud cluster, a voxel count sequence is extracted from its historical point cloud data across multiple consecutive frames. The voxel density volatility is then calculated based on this sequence. The voxel count sequence records the scale change of the point cloud cluster over time, obtained by tracking the voxel occupancy within the same spatial region across consecutive frames. The tracking process is based on spatial location matching, identifying the current frame's point cloud cluster and the corresponding object in historical frames, and extracting the voxel counts from each frame to form a time series. The voxel density volatility quantifies the drastic degree of temporal change in object size, and its calculation formula is as follows:

[0040] ;

[0041] in, Voxel density fluctuation rate This represents the number of historical frames (usually 5-10 frames). For the first The number of voxels in a frame. This represents the time-averaged number of voxels. This indicator uses the coefficient of variation (standard deviation divided by the mean) to eliminate the influence of differences in object size, making point cloud clusters of different sizes comparable. Due to their unstable shape and blurred boundaries, the number of voxels in floating objects such as drifting debris fluctuates significantly with wind force and viewing angle, typically with a fluctuation rate greater than 0.3; while fixed obstacles have stable shapes, with a fluctuation rate close to 0. Voxel density fluctuation rate is an important component of spatiotemporal dynamic characteristics, providing a temporal dimension for drifting debris identification.

[0042] The lowest point height of the discrete point cloud cluster in the vertical direction is obtained, and the vertical axis suspension offset is calculated by combining this with the ground clearance parameter of the cotton spinning robot. The lowest point height is obtained by traversing the Z-coordinate (vertical coordinate) of all points in the point cloud cluster and taking the minimum value, representing the height of the bottom of the object from the ground. The ground clearance parameter of the cotton spinning robot is the distance between the lower edge of the robot's chassis and the ground, usually 8-15 cm, which serves as a reference for determining whether the object is on the ground. The formula for calculating the vertical axis suspension offset is:

[0043] ;

[0044] in, This is the vertical axis suspension offset. This represents the height of the lowest point in the point cloud cluster. This is the ground reference height (determined through the ground stripping process). This indicator visually reflects whether an object is in contact with the ground. A value close to 0 indicates that the object is in contact with the ground, which is a typical characteristic of solid obstacles. A value significantly greater than 0 (e.g., greater than 20 cm) indicates that the object is suspended in the air, consistent with the physical characteristics of flying flowers. The vertical axis suspension offset provides a basis for distinguishing spatial dimensions, forming a spatiotemporally complementary feature system with the voxel density fluctuation rate.

[0045] A morphological closing operation is performed on the projection of a discrete point cloud cluster onto a horizontal plane, and the bottom support connectivity index is calculated based on the area ratio before and after the closing operation. Horizontal projection projects the 3D point cloud cluster onto the XY plane, forming a 2D contour image that reflects the shape characteristics of the object's bottom. The morphological closing operation is an image processing operation that first dilates and then erodes, which can fill small holes and gaps within the contour, connect broken boundaries, and obtain a more complete continuous region. The size of the structuring element is set according to the point cloud resolution, typically 3-5 pixels. The formula for calculating the bottom support connectivity index is:

[0046] ;

[0047] in, This serves as a bottom support connectivity indicator. This represents the area of ​​the projected contour before the closing operation. This represents the area after the closing operation. For solid obstacles with intact bottom structures, the area change before and after the closing operation is small, with a ratio close to 1. For loose objects such as debris, the projected outline is usually broken and discontinuous, and the closing operation significantly increases the area, with a ratio significantly less than 1 (usually less than 0.6). This indicator quantifies the structural integrity of the object's bottom and is an important geometric feature for distinguishing between solid obstacles and loose debris.

[0048] By combining voxel density volatility, vertical axis levitation offset, and bottom support connectivity indices, a spatiotemporal dynamic feature vector for this discrete point cloud cluster is generated. The feature vector combination is expressed in a standardized multidimensional vector form as follows:

[0049] ;

[0050] in, For spatiotemporal dynamic feature vectors, The height normalization coefficient (usually 1 meter) is used to normalize the height value to a range comparable to other features. This three-dimensional feature vector integrates temporal dynamic characteristics (fluctuation), spatial location characteristics (hovering offset), and geometric structure characteristics (support connectivity), forming a comprehensive description of the physical properties of the point cloud cluster. The fusion of multi-dimensional features significantly improves the accuracy and robustness of fly debris recognition; misjudgments that may occur with a single feature are corrected through multi-feature cross-validation. The spatiotemporal dynamic feature vector provides a rich information foundation for subsequent intelligent classification decisions and is the core data structure for achieving reliable fly debris interference recognition.

[0051] In this embodiment of the invention, step 2 above may include:

[0052] Discrete point cloud clusters with voxel density fluctuation rates greater than a preset fluctuation threshold, vertical axis suspension offsets within a preset floating range, and bottom support connectivity indices lower than a preset support threshold are marked as floating interference candidate clusters. The remaining discrete point cloud clusters are marked as confidence obstacle clusters. This classification strategy based on multi-condition joint judgment reflects the decision logic of feature fusion, requiring point cloud clusters to simultaneously satisfy three dimensions of fly debris features to be identified as floating interference. The preset fluctuation threshold is typically set to 0.25-0.35; values ​​above this threshold indicate significant changes in object shape over time, exhibiting unstable characteristics of floating objects. The preset floating range is typically set to [0.15m, 2.0m], with the lower limit avoiding misclassification of small obstacles near the ground as fly debris and the upper limit excluding noise points measured by lidar. This range corresponds to the typical floating height range of fly debris in a cotton mill. The preset support threshold is typically set to 0.55-0.65; values ​​below this threshold indicate broken and discontinuous bottom projection contours, lacking continuous bottom support features of solid obstacles. The logical expression for joint judgment is: if... If the condition is met, it is marked as a candidate cluster for suspended interference; otherwise, it is marked as a cluster of confident obstacles. This conservative classification strategy prioritizes safety, identifying a point cloud cluster as negligible interference only when all three features clearly point to fly debris, and conservatively identifying any point cloud cluster whose features do not conform to fly debris characteristics as an obstacle to be avoided. The threshold parameter can be optimized and adjusted using machine learning methods according to the specific workshop environment and fly debris characteristics to improve recognition accuracy. This heuristic rule classification method based on physical features is computationally efficient, logically clear, and highly interpretable, making it suitable for real-time navigation system applications.

[0053] In this embodiment of the invention, mapping suspended interference candidate clusters to a temporary observation buffer layer associated with a static environment map includes:

[0054] A temporary observation buffer layer with the same scale and coordinate system as the static environment map is established. This temporary observation buffer layer stores the occupancy status of spatial entities that have not been identified as confident obstacle clusters. The temporary observation buffer layer adopts the same scale and coordinate system as the static environment map. Figure 1 A consistent grid structure, with each grid cell corresponding to the same physical spatial size (typically 5-10 cm), ensures accurate spatial alignment and overlay operations between the two data layers. The coordinate system adopts the global reference frame used for robotic arm navigation, typically a Cartesian coordinate system established with a fixed point in the workshop as the origin. The temporary observation buffer layer's data structure includes attributes such as grid occupancy status (idle / occupied), confidence coefficient, and timestamps, adding temporal dynamic information compared to static maps. This buffer layer is designed to isolate and manage uncertain obstacle information, preventing transient interference from being written into the static map, thus avoiding map pollution and long-term conservative biases in path planning. The independent buffer layer architecture supports flexible update and clear operations, providing an efficient data management mechanism for dynamic environment modeling.

[0055] When the candidate clusters of floating interference are first mapped to the temporary observation buffer layer, an initial confidence coefficient is assigned to the grid cells covered by the candidate clusters. This initial confidence coefficient is determined based on the mean point cloud reflection intensity of the candidate clusters. The mapping process converts the spatial coordinates of the point cloud clusters into grid indices, projects all points within the cluster onto the corresponding grid cells, and marks them as occupied. The initial confidence coefficient quantifies the initial credibility of the obstacle's existence, reflecting the mean intensity using an innovative method of lidar reflection intensity information. Point cloud reflection intensity reflects the material properties and reflection characteristics of an object's surface. Solid obstacles typically have high and stable reflection intensity, while fluffy materials such as fly debris have low and uneven reflection intensity. The formula for calculating the initial confidence coefficient is:

[0056] ;

[0057] in, This is the initial confidence level coefficient. This is the normalization coefficient (usually taken as 0.7-0.9). The mean reflection intensity of the point cloud cluster. This is a reference intensity value (calibrated according to the lidar model). The function maps the ratio to the [0,1] interval and provides saturation characteristics. This confidence initialization method based on physical measurements more accurately reflects the reliability of the observations than simple fixed assignment, providing a reasonable initial state for subsequent time decay and cumulative correction.

[0058] Based on the moving speed of the cotton spinning robot and the characteristics of the ambient airflow, a time decay step size is configured for each grid cell in the temporary observation buffer layer. Within each control cycle, the initial confidence coefficient is decreased according to the time decay step size. The time decay mechanism embodies the Bayesian update principle of gradually reducing the confidence of objects that have not been repeatedly observed, conforming to the physical laws governing the dissipation of transient interference such as fly debris. The moving speed affects the refresh rate of the sensor's field of view; the faster the speed, the shorter the interval between repeated observations of the same area, and the decay step size should be adjusted accordingly to avoid excessively rapid clearing. Ambient airflow characteristics include workshop ventilation intensity and airflow direction; the stronger the airflow, the faster the fly debris dissipates, and the decay step size should be adjusted accordingly. The formula for calculating the time decay step size is:

[0059] ;

[0060] in, The time decay step size, The base decay rate (usually 0.02-0.05 per control cycle) For the movement speed of the robotic arm, For speed influence coefficient, For ambient airflow velocity, This represents the airflow influence coefficient. The control period is typically 100-200 milliseconds, matching the sensor scanning frequency. The confidence-decreasing operation employs a linear decay model. ,when The decay stops when the value drops to 0. This adaptive decay mechanism enables the temporary observation buffer layer to automatically adjust the timeliness of information according to the dynamic characteristics of the environment. It can quickly eliminate outdated information in a high-dynamic environment and maintain information stability in a low-dynamic environment, thus achieving intelligent adaptation to the rate of environmental change.

[0061] In subsequent frame observations, if the same spatial location is again marked as a candidate cluster for floating interference, an incremental correction operation is performed on the confidence coefficient of the grid cell. Repeated observations indicate that the obstacle at this location is not transient but persistent, and its confidence should be increased to reflect a higher probability of existence. The incremental correction uses an incremental update strategy; each time a new candidate cluster for floating interference is observed at the same location, the confidence is increased by a correction amount. The incremental correction formula is:

[0062]

[0063] in, This is the corrected confidence coefficient. This represents the current confidence level coefficient. This is the observation increment (usually taken as 0.15-0.25). This is the upper confidence level (usually set to 1.0). The formula is derived through... The function limits the confidence level to a maximum value to prevent numerical overflow. A dynamic balance is formed between the repeated observation accumulation mechanism and time decay: the confidence level of frequently observed objects remains high or even increases, while the confidence level of objects that have not been observed for a long time gradually decreases until they are cleared. This two-way adjustment mechanism enables the temporary observation buffer layer to adaptively distinguish between persistent obstacles and transient interference. Persistent objects will eventually be upgraded to confident obstacles because their confidence level accumulates to exceed a threshold, while randomly passing debris will be automatically cleared due to confidence decay, realizing intelligent lifecycle management of uncertain obstacles.

[0064] In this embodiment of the invention, a hybrid cost raster is constructed based on a static environment map and a temporary observation buffer layer, including:

[0065] The algorithm reads the fixed obstacle attributes of each basic grid in the static environment map and assigns corresponding fixed costs based on these attributes. The static environment map stores verified persistent obstacle information, and each basic grid has an occupancy attribute (free / occupied / unknown). The fixed cost is a quantitative indicator used by the path planning algorithm to evaluate path quality; a higher cost indicates a greater risk or cost in traversing that grid. The cost allocation strategy typically employs a hierarchical scheme: occupied grids (inside obstacles) are assigned a fatal cost (e.g., 255, indicating no passage); the expanded area surrounding occupied grids is assigned a high cost (e.g., 200-254, indicating strong discouragement); free space is assigned a low cost (e.g., 0, indicating free passage); and unknown areas are assigned a medium cost (e.g., 127, indicating cautious passage). This hierarchical cost system provides fine-grained spatial preference guidance for the path search algorithm, ensuring that the planned path stays away from obstacles while maintaining a safety margin.

[0066] The confidence coefficient of the associated grid cells in the temporary observation buffer layer is read, and then multiplied by a preset interference cost weight to obtain the floating cost value of each associated grid cell. The associated grid cells are the grids marked as occupied in the temporary observation buffer layer, corresponding to the spatial location of the floating interference candidate cluster. The confidence coefficient reflects the degree of credibility of the obstacle's existence; by weighting and converting it into path cost, a quantitative mapping from uncertainty to planning cost is achieved. The interference cost weight is a parameter that adjusts the intensity of the floating interference's impact. It is set according to the security requirements of the application scenario; a higher weight (e.g., 150-200) is set in conservative applications, and a lower weight (e.g., 80-120) is set in aggressive applications. The floating cost value calculation formula is:

[0067] ;

[0068] in, For floating value, This is the confidence coefficient (value 0-1). This represents the interference cost weight. The formula achieves a linear mapping from confidence level to cost; higher confidence results in higher cost, and the path planning algorithm is more inclined to avoid that region. The dynamic characteristic of the floating cost is reflected in the time variation of the confidence coefficient. As the confidence level decays, the floating cost automatically decreases, and the path planning gradually allows passage through that region, achieving an adaptive response of the path strategy to dynamic environmental changes.

[0069] For overlapping base and associated raster cells, the fixed cost and floating cost are arithmetically superimposed to obtain the mixed cost value at the corresponding location in the mixed cost raster. The overlapping area refers to the spatial location where obstacle information exists simultaneously in the static environment map and the temporary observation buffer layer, possibly originating from floating debris near fixed obstacles or newly appearing unconfirmed obstacles. The arithmetic superposition uses a simple cost summation strategy, comprehensively considering the influence of both obstacle sources. The formula for calculating the mixed cost value is:

[0070] ;

[0071] in, For mixed generation value, For fixed value, For floating value, This is the upper limit of the cost (typically 255). The combined cost, after being stacked, may exceed the cost of a single obstacle source, reflecting the cumulative effect of multiple threats. The function is constrained within a reasonable range to avoid overflow. For non-overlapping graticules, the hybrid cost is directly taken as either a fixed cost or a floating cost. This cost fusion mechanism achieves a unified expression of static deterministic information and dynamic uncertain information, providing a complete environmental cost map for path planning algorithms.

[0072] The hybrid cost grid is expanded, with the expansion range determined by the contour envelope of the robotic arm in its retracted state. Expansion extends the obstacle region outward, creating a safety buffer zone around the obstacle to compensate for the robotic arm's physical dimensions and positioning errors. The contour envelope is the minimum circumscribed geometry of the robotic arm in its fully retracted state, typically rectangular or circular, with its size determined by the width and length of the robotic arm's chassis, typically 50-80 cm. The expansion range is set to the envelope radius plus a safety margin (usually 10-20 cm) to ensure the expanded space can fully accommodate the robotic arm's passage. The expansion operation employs a morphological expansion algorithm, using circular or square structuring elements to expand the high-cost regions of the hybrid cost grid, raising the cost of the free space grid outside the obstacle boundary to a high cost, forming a no-passage buffer zone. The expanded hybrid cost grid accurately reflects the actual passage constraints considering the robotic arm's size, providing a safe and reliable cost map for path planning and avoiding collision risks caused by planned paths being too close to obstacles.

[0073] In this embodiment of the invention, step S5 may include:

[0074] A heuristic search algorithm is used to search for the lowest-cost path from the starting position to the target job position in a mixed-cost grid, which is then used as the initial job path. The heuristic search algorithm typically employs A* or D* algorithms, and evaluates the function... Guiding the search direction, among which From the starting point to the node The actual cost, For the node Heuristic cost estimation to the target. Actual cost. Accumulated path through the grid's mixed cost, heuristic function Euclidean or Manhattan distances are typically used. The search process maintains open and closed lists, iteratively expanding the nodes with the lowest cost until the target location is reached or no feasible path is confirmed. The lowest-cost path is the optimal or near-optimal solution that comprehensively considers path length, obstacle avoidance, and avoidance of uncertain areas, representing the best movement strategy under the current environmental perception. The initial operation path is expressed as a waypoint sequence, containing the coordinates and heading angles of key locations on the path, providing navigation instructions for the robot's motion control.

[0075] When a cotton spinning robot moves along its initial work path, data from a temporary observation buffer layer within a pre-defined viewing window is acquired at a set frequency. This set frequency, typically 5-10 Hz, matches the sensor scanning frequency and control cycle to ensure real-time tracking of environmental changes. The pre-defined viewing window is defined as a fan-shaped or rectangular area in front of the robot, usually 3-8 meters in distance, covering a 180-degree field of view. This range corresponds to the robot's braking distance at its current speed and the anticipated distance required for path adjustment. Data acquisition within the viewing window involves querying the corresponding grid cells in the temporary observation buffer layer to extract information such as occupancy status, confidence coefficient, and timestamps, forming a dynamic snapshot of the local environment. This forward predictive monitoring strategy allows the robot to anticipate environmental changes ahead of its path, providing ample decision-making and execution time for path adjustment and speed control.

[0076] The confidence coefficients for all suspended interference candidate clusters within a preset viewport are extracted, and a trend curve showing the evolution of these confidence coefficients over time is plotted. Confidence coefficient extraction involves identifying grid cells belonging to suspended interference candidate clusters within the viewport, reading the current confidence value of each cell, and aggregating them into cluster-level confidence values ​​(usually the average or maximum confidence value of the grid cells within the cluster). The trend curve records the numerical changes of the confidence coefficients at continuous time points, with time on the horizontal axis and confidence on the vertical axis, represented by a line graph or smooth curve. The trend curve visually demonstrates the dynamic evolution pattern of suspended interference candidate clusters: a monotonically decreasing curve indicates that the pollen is dissipating and the confidence is decreasing; a fluctuating increasing curve indicates that the object is being repeatedly observed and the confidence is increasing; a stable curve indicates that the object's state is stable. This time series analysis provides a data foundation for predicting the future state of confidence.

[0077] Based on the slope change rate of the trend curve, the confidence level of the suspended interference candidate cluster at the next time step is predicted. If the predicted state is lower than the safety threshold, a path smoothing transition instruction is generated in advance. The slope represents the rate of change of confidence, and the slope change rate (second derivative) reflects the acceleration or deceleration trend of the rate of change. The prediction uses linear extrapolation or polynomial fitting methods, estimating the confidence value for the next control cycle based on the confidence values ​​at several historical time points. The prediction formula uses first-order linear extrapolation:

[0078] ;

[0079] in, For the confidence level of the predicted future, At the current confidence level, The rate of change of confidence at the current moment (approximated by difference). The prediction time step is typically 1-2 control cycles. The safety threshold is usually set to 0.2-0.3; below this value, the obstacle threat is considered eliminated and safe passage is permitted. When the prediction confidence level falls below the safety threshold, a smooth path transition instruction is generated in advance, triggering local path replanning preparation. This allows the robot to seamlessly switch to the optimized path when the confidence level actually drops to the threshold. This prediction-based feedforward control strategy offers better smoothness and anticipation compared to passive threshold responses, reducing lag and abrupt changes in path adjustments and improving motion continuity and operational quality.

[0080] In this embodiment of the invention, step S6 may include:

[0081] The system compares the current confidence coefficient of each suspended interference candidate cluster with the safety threshold in real time. When the current confidence coefficient drops below the safety threshold, the corresponding grid cell is cleared from the temporary observation buffer layer. Real-time comparison is achieved by traversing all suspended interference candidate clusters in the temporary observation buffer layer in each control cycle, reading the confidence coefficient, and comparing it with the safety threshold. The safety threshold is set by comprehensively considering the risk of false detection and passage efficiency. An excessively high threshold leads to overly conservative behavior and reduced efficiency, while an excessively low threshold may misjudge residual fly debris as dissipated, increasing the risk of collision. It is typically determined to be 0.2-0.3 through experimental optimization based on actual application scenarios. The clearing operation resets the occupancy status of the corresponding grid cell to idle, resets the confidence coefficient to zero, and removes it from the active data structure of the temporary observation buffer layer, freeing up storage space and computing resources. This threshold-based automatic clearing mechanism achieves lifecycle management of transient interference, ensuring that the temporary observation buffer layer retains only currently valid uncertain obstacle information, avoiding memory overflow and computational burden caused by the accumulation of outdated data.

[0082] In the hybrid cost raster, the hybrid cost value of cleared raster cells is updated to include only fixed costs or set to passability costs. Cost updates are a synchronous operation between environmental perception changes and path planning inputs, ensuring that the hybrid cost raster accurately reflects the current environmental state. For cleared raster cells, if the location is also an empty area in the static environment map, the hybrid cost value is set to passability costs (usually 0), indicating complete free passage; if the location has fixed obstacle information in the static environment map, the hybrid cost value is restored to the fixed cost value, retaining the constraints of static obstacles. This differentiated update strategy ensures the correct fusion of information from the two map layers and avoids the accidental deletion of static obstacle constraints after temporary observation data is cleared. After the cost update, the attractiveness of the area for path planning is significantly improved, creating conditions for path optimization.

[0083] The process extracts affected local path segments between the current position and the target work position of the cotton spinning robot. A local search is then performed on the updated hybrid cost grid to generate replacement path segments. Affected path segments are identified by analyzing the spatial relationship between the cleared area and the currently planned path. If the cleared area is within a certain range of the planned path (typically 2-5 meters), the path segments near that area are marked as affected segments. The local search employs a heuristic algorithm similar to global path planning, but the search space is limited to a local area surrounding the affected path segments, significantly reducing computational complexity. The search starts at the beginning waypoint of the affected segment and ends at the end waypoint, searching for a new lowest-cost connection path on the updated hybrid cost grid. Replacement path segments leverage the low-cost advantage of the cleared area, typically being shorter or smoother than the original path, representing an optimized response to environmental changes. Compared to global replanning, local search requires less computation and has a faster response, making it suitable for real-time dynamic adjustments.

[0084] Spline curve interpolation is used to smooth the curvature at the junction of the replacement path segment and the original path, controlling the cotton spinning robot to smoothly enter and pass through the area where the original suspended disturbance candidate cluster is located without changing its basic travel speed. The junction is the connection point between the replacement path segment and the retained original path segment, which may have corners or curvature discontinuities; direct switching would cause abrupt motion changes. Spline curve interpolation uses cubic B-splines or Bézier curves, with waypoints near the junction point as control points, to fit and generate a smooth transition curve. Curvature smoothing ensures the continuity of the first derivative (tangent direction) and second derivative (curvature) of the path, eliminating corners and curvature abrupt changes, and satisfying the robot's kinematic constraints. The smoothed path allows the robot to maintain its basic travel speed (without deceleration or stopping), achieving smooth path switching. The control strategy uses path tracking algorithms (such as pure tracking or model predictive control) to make the robot move along the smooth path, smoothly enter and pass through the original suspended disturbance area, and complete the adaptive response to environmental changes. This smooth transition mechanism avoids frequent speed adjustments and sharp turns, improves motion efficiency and stability, and reduces disturbances to the robotic arm's operation.

[0085] In this embodiment of the invention, step S7 may include:

[0086] If, over multiple consecutive control cycles, the confidence coefficient of a candidate suspended disturbance cluster consistently remains above a safety threshold, and the boundary variance of this candidate cluster in the horizontal plane converges to a preset stable interval, then the candidate cluster is determined to be a physical obstacle. A consistently high confidence level is the first condition for determining a physical obstacle. Multiple consecutive control cycles are typically set to 10-20 cycles (corresponding to 1-4 seconds), and the duration must ensure that it is not a random short-term observation but a stable and continuous phenomenon. Boundary variance convergence is a quantitative indicator of spatial stability. It involves calculating the statistical variance of the boundary coordinates (e.g., the vertex coordinates of the minimum bounding rectangle) of the candidate suspended disturbance cluster in consecutive frames; a small variance indicates a stable boundary position. The preset stable interval is typically a standard deviation of less than 5-10 cm, corresponding to the typical error range of point cloud resolution and positioning accuracy. The formula for calculating the boundary variance is:

[0087] ;

[0088] in, For boundary variance, To count the number of frames, For the first The coordinates of the frame's boundary center. The time-averaged coordinates of the boundary center are used. Meeting both the confidence and spatial stability conditions indicates that the object is not a drifting flower, but a fixed physical obstacle, such as newly placed equipment, leftover materials, or other static objects. Classifying it as a physical obstacle represents a state transition from an uncertain assumption to a deterministic conclusion, triggering subsequent map updates and control responses.

[0089] The spatial coordinates and contour information of suspended interference candidate clusters identified as physical obstacles are transformed into static occupancy data and written to the corresponding area in the static environment map. Simultaneously, the record of the suspended interference candidate cluster is deleted from the temporary observation buffer layer. Spatial coordinate transformation maps the 3D geometric information of the point cloud cluster to raster occupancy markers on the static map through projection and rasterization operations. Contour information includes the boundary shape and size parameters of the point cloud cluster, used to accurately define the occupancy range of the obstacle. The static occupancy data writing operation sets the occupancy attribute of the corresponding raster cell to the occupancy state and optionally records extended attributes such as the obstacle's height and type. Writing to the static map makes the obstacle a persistent environmental constraint, continuously considered and avoided in all subsequent path planning. Deletion from the temporary observation buffer layer avoids information duplication and conflict, ensuring that the same object is managed only in one data layer. The deletion operation clears the occupancy state and confidence information of the corresponding raster in the temporary buffer layer, releasing resources. This state transition mechanism achieves an automatic upgrade from dynamic uncertainty management to static deterministic management, demonstrating the adaptive learning capability of environmental cognition.

[0090] The hybrid cost raster is refreshed based on the updated static environment map to determine whether the initial work path interferes with physical obstacles. Hybrid cost raster refresh is a synchronous operation between map update and path planning. It rereads the fixed cost value of the static map and overlays it with the floating cost value of the temporary observation buffer layer to generate a new hybrid cost raster. Interference is determined by checking whether the raster cells traversed by the initial work path contain newly added high-cost or fatal-cost areas. If a path point falls into an obstacle-occupied area or an expanded safety zone, interference is considered. Interference detection employs a geometric collision detection algorithm, considering the actual size envelope of the robot to ensure accuracy. Interference indicates that the current path is no longer feasible; continued execution will result in a collision, requiring immediate countermeasures.

[0091] If interference occurs, the dynamic distance between the cotton spinning robot and the physical obstacle is calculated. If the dynamic distance is less than the braking threshold, an emergency stop command is triggered, and a new detour path is planned. The dynamic distance is the shortest Euclidean distance from the robot's current position to the obstacle boundary, calculated by querying the spatial relationship between the robot's position grid and the obstacle's occupied grid. The braking threshold is determined based on the robot's dynamic characteristics, typically the emergency braking distance at the current speed plus a safety margin, with a typical value of 0.5-1.5 meters. The distance calculation is updated in real time and changes dynamically with the robot's movement. The emergency stop command is the highest priority safety control command, immediately sending a zero-speed target to the motion control system and activating the braking system to stop the robot within the shortest distance. Emergency stop ensures the last line of defense before a collision, preventing equipment damage and safety accidents. Immediately after the emergency stop, global path replanning is initiated, searching for a new path from the current stopping position to the target work position on the updated hybrid cost grid, bypassing newly identified physical obstacles. The detour path planning uses the same heuristic search algorithm as the initial path planning, ensuring that a feasible and low-cost alternative is found. After replanning, the robotic arm resumes movement along the new path, completing the full response process to sudden environmental changes. This layered protection mechanism (interference detection - distance monitoring - emergency stop protection - path replanning) ensures safe operation in dynamic environments, reflecting the design principle of prioritizing safety.

[0092] In this embodiment of the invention, during the movement of the cotton spinning robot along the initial working path, the following further steps are included:

[0093] Real-time acquisition of joint angle data of the robotic arm mounted on the cotton spinning robot is used to calculate the real-time sweep radius of the robotic arm on the horizontal plane. Joint angle data is collected through encoders or angle sensors at each joint of the robotic arm, reflecting the current rotation angle or extension position of each joint. The real-time sweep radius is the farthest distance that the end effector of the robotic arm can reach on the horizontal plane, determined by the horizontal projection distance from the center of the robot's chassis to the outermost point of the robotic arm. The calculation employs the forward kinematics method, calculating the end effector position coordinates based on the link parameters of the robotic arm (such as link length and DH parameters) and joint angles, extracting the horizontal plane coordinate components, and calculating the distance to the chassis center. For multi-degree-of-freedom robotic arms, the sweep radius is taken as the maximum horizontal distance under all possible postures. The calculation formula is based on the kinematic transformation matrix:

[0094] ;

[0095] in, To determine the real-time sweep radius, This represents the horizontal coordinate of the robotic arm's end effector in the chassis coordinate system. The real-time sweep radius dynamically reflects the impact of the robotic arm's extension state on the overall envelope size of the robotic arm and is a key parameter for dynamic collision detection.

[0096] The expansion range of the hybrid cost grid is dynamically adjusted based on the real-time sweep radius. This expansion range adjustment is a dynamic mechanism that adapts the obstacle safety buffer zone to changes in the actual size of the robotic arm. When the robotic arm extends, the sweep radius increases, requiring a larger safety buffer zone; when the robotic arm retracts, the sweep radius decreases, allowing for a narrower safety buffer zone and narrower passageways. The expansion range is set to the real-time sweep radius plus a fixed safety margin (typically 10-15 cm). Dynamic adjustment is achieved by re-performing the expansion operation of the hybrid cost grid, using the updated expansion range as the structural element size to generate a new expansion cost distribution. This dynamic expansion mechanism is more precise and efficient than fixed expansion (always expanding at the maximum sweep radius), avoiding excessive conservatism during robotic arm retraction and improving space utilization and passage capacity.

[0097] When a floating interference candidate cluster with a floating cost value greater than zero but lower than the blocking threshold is detected in the mixed cost grid ahead, a robotic arm retraction command is generated to reduce the real-time sweep radius. This ensures that the overall envelope of the cotton spinning robot can safely pass through the outer boundary of the floating interference candidate cluster after retraction, without triggering an avoidance operation. A floating cost value greater than zero indicates the presence of a floating interference candidate cluster, while a value lower than the blocking threshold (usually set to 200-220) indicates that the threat level of the interference is insufficient to completely prohibit passage, representing a moderate threat level between negligible and mandatory avoidance. Detection is achieved by querying the mixed cost grid within a preset distance range ahead of the path, identifying grid areas that meet the cost conditions. The robotic arm retraction command is an active morphological adjustment strategy that controls the joint angle of the robotic arm to retract it inward, reducing the horizontal sweep radius. The retraction objective is to ensure that the radius of the retracted envelope plus a safety margin is less than the distance from the robotic arm to the boundary of the floating interference candidate cluster, ensuring safe passage. The formula for determining the retraction condition is:

[0098] ;

[0099] in, For the target shrinkage radius, For safety margin, This represents the minimum clearance distance from the path to the interference boundary. If the current sweep radius does not meet this condition but can be met through retraction, a retraction command is generated; if it still does not meet the condition even with full retraction, detour path planning is triggered. The robotic arm retraction is executed through joint motion control, smoothly adjusting the robotic arm to the retracted posture while maintaining the motion continuity of the mobile base. This proactive morphological adjustment strategy significantly improves the ability to traverse narrow spaces and areas with moderate interference, avoiding unnecessary detours and stops while ensuring safety, thus improving operational efficiency and path directness. This strategy embodies the concept of overall coordinated control of the robotic arm, optimizing the mobile platform and robotic arm as a unified system, and fully utilizing the flexibility of the robotic arm to assist navigation decisions.

[0100] In this embodiment of the invention, when the cotton spinning robot reaches the target work position and performs the grasping or placing operation, it further includes:

[0101] This study monitors local airflow disturbance data generated by a cotton spinning robot during its operations and assesses the impact range of this disturbance on a temporary observation buffer layer. Local airflow disturbances are generated by the robot's rapid movements, arm swings, and grasping actions, which stir up the surrounding air and stir up dust adhering to the ground or equipment. Airflow disturbance data is obtained through environmental sensors (such as wind speed and barometric pressure sensors) or kinematic models, quantifying the intensity and direction of the airflow. The impact range assessment is based on airflow intensity and the amplitude of the operational actions, typically defined as a spherical or cylindrical area of ​​0.5–2 meters around the operational position, within which suspended particles may be re-floated by the disturbance. The assessment considers differences in operation type; grasping actions generally generate smaller disturbances, while placement actions, especially rapid placement, can generate larger airflow impacts. The spatial relationship between the impact range and the temporary observation buffer layer is determined through geometric calculations, identifying grid cells within the buffer layer that fall within the impact range.

[0102] When the affected area covers the blind zone or effective field of view of the lidar, the confidence coefficient attenuation operation for the corresponding region in the temporary observation buffer layer is paused, and the hybrid cost update for that region is frozen. The blind zone is an area that the lidar cannot effectively observe due to obstruction, reflection characteristics, or installation location limitations; it typically includes the vicinity of the robotic arm or below it. The effective field of view is the spatial range within which the lidar can normally acquire reliable point cloud data. When operational disturbances affect these areas, the sensor may observe swirled debris and mistakenly identify it as newly emerging suspended interference, leading to false suspended interference candidate clusters in the temporary observation buffer layer or incorrectly increasing the confidence of dissipated debris. Pausing the attenuation operation freezes the confidence coefficient of the corresponding region's grid cells, does not perform time-based decrementing, and maintains the current value. Freezing the cost update stops the refresh calculation of the hybrid cost grid for that region, maintaining the cost distribution before the operation. This pause mechanism prevents false environmental changes caused by self-disturbance from being incorporated into navigation decisions, preventing the operation's actions from contaminating the environmental perception model. The pause area is limited to the intersection of the affected area and the perception-sensitive area, and does not affect the normal updates of other areas.

[0103] After the cotton spinning robot completes its operation, a preset time delay is applied before resuming the confidence coefficient decay process on the temporary observation buffer layer. This prevents fly debris stirred up by the operation from being misjudged as newly generated clusters of confidence obstacles. The preset time period is the time required for airflow disturbances to subside and for the stirred-up fly debris to settle again; it is typically set between 2 and 5 seconds, depending on the workshop environment and operational intensity. The delay mechanism is implemented using a timer, which starts when the operation is completed and resumes the normal confidence update process after the delay time expires. During the delay, the sensor continues to collect point cloud data, but observations within the affected area are handled specially: newly observed suspended point cloud clusters are not immediately written into the temporary observation buffer layer or are assigned a lower initial confidence level, and existing suspended interference candidate clusters are not accumulated with confidence levels even if they are observed repeatedly. After the delay ends, the observation processing returns to normal, confidence decay restarts, and new observations are written into the buffer layer normally. This delay-recovery mechanism provides sufficient time for the environment to stabilize, preventing transient disturbances from becoming persistent obstacle perceptions. In particular, this mechanism prevents the rapid accumulation of confidence in fly debris generated by operational actions due to repeated observations in a short period, which could lead to it being mistakenly elevated to a cluster of confident obstacles and written into the static map, causing map pollution and unnecessary detours in subsequent routes. This mechanism demonstrates intelligent identification of the causes of dynamic environmental changes, distinguishing between natural environmental changes (which require response) and self-disturbances (which require suppression), improving the accuracy and robustness of environmental cognition, and ensuring the stability and reliability of the navigation system in dynamic environments with frequent operations.

[0104] like Figure 2 As shown, embodiments of the present invention also provide an intelligent cotton spinning robot autonomous mobile operation control system, including:

[0105] The 3D perception module is used to acquire the original 3D point cloud data of the environment in which the cotton spinning robot is located;

[0106] The point cloud processing module is used to perform spatial clustering processing on the original 3D point cloud data to extract several discrete point cloud clusters, and to calculate the spatiotemporal dynamic feature vector of each discrete point cloud cluster.

[0107] The interference identification module is used to perform fly flower interference identification on each discrete point cloud cluster based on spatiotemporal dynamic feature vectors, and obtain floating interference candidate clusters and confidence obstacle clusters;

[0108] The map management module is used to directly write the clusters of believed obstacles into the static environment map used for navigation, and to map the candidate clusters of floating interference to a temporary observation buffer layer associated with the static environment map.

[0109] The cost grid construction module is used to construct a hybrid cost grid based on a static environment map and a temporary observation buffer layer when the cotton spinning robot receives a mobile operation instruction.

[0110] The path planning module is used to perform path search based on the hybrid cost grid to generate an initial job path;

[0111] The real-time monitoring module is used to monitor the changes in the confidence coefficient of suspended interference candidate clusters in the temporary observation buffer layer in real time as the cotton spinning robot moves along the initial working path.

[0112] The dynamic replanning module is used to remove the floating interference candidate cluster from the temporary observation buffer layer when the confidence coefficient of the floating interference candidate cluster decays to below the safety threshold, and to trigger local path replanning for the region where the floating interference candidate cluster is located.

[0113] The motion control module is used to control the cotton spinning robot to smoothly pass through the area where the original suspended interference candidate cluster is located, and when the confidence coefficient of the suspended interference candidate cluster is continuously higher than the safety threshold and accompanied by spatial position convergence, the suspended interference candidate cluster is promoted to a confidence obstacle cluster and written into the static environment map, while controlling the cotton spinning robot to perform avoidance and stopping operations.

[0114] The data storage module is used to store static environment maps, temporary observation buffer layers, and related control parameters;

[0115] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0116] This invention achieves autonomous mobile operation control of a cotton spinning robot in complex fly-feather environments through spatiotemporal dynamic feature extraction, intelligent identification of fly-feather interference, dynamic management of a two-layer map, a confidence time decay mechanism, hybrid cost grid construction, adaptive path replanning, and collaborative control of the robotic arm. The multi-level perception and decision-making method of this invention can accurately distinguish between fly-feather interference and real obstacles, effectively addressing the dynamic environmental characteristics of cotton spinning workshops, and significantly improving the safety and efficiency of mobile operations.

[0117] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0118] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0119] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for controlling autonomous movement of a smart cotton spinning mechanical hand, characterized in that, include: Step S1: Obtain the original three-dimensional point cloud data of the environment where the cotton spinning robot is located, perform spatial clustering processing on the original three-dimensional point cloud data to extract several discrete point cloud clusters, and calculate the spatiotemporal dynamic feature vector of each discrete point cloud cluster. Step S2: Based on the spatiotemporal dynamic feature vector, perform fly flower interference identification on each of the discrete point cloud clusters to obtain floating interference candidate clusters and confidence obstacle clusters; Step S3: The confident obstacle cluster is directly written into the static environment map used for navigation, and the floating interference candidate cluster is mapped to a temporary observation buffer layer associated with the static environment map; Step S4: When the cotton spinning robot receives a moving operation instruction, a hybrid cost grid is constructed based on the static environment map and the temporary observation buffer layer; Step S5: Path search is performed based on the hybrid cost grid to generate an initial operation path, and the confidence coefficient change of the suspended interference candidate clusters in the temporary observation buffer layer is monitored in real time during the movement of the cotton spinning robot along the initial operation path. Step S6: When the confidence coefficient of the floating interference candidate cluster decays to below the safety threshold, the floating interference candidate cluster is removed from the temporary observation buffer layer, and local path replanning is triggered for the region where the floating interference candidate cluster is located. Step S7: When the confidence coefficient of the suspended interference candidate cluster is continuously higher than the safety threshold and accompanied by spatial position convergence, the suspended interference candidate cluster is promoted to a confidence obstacle cluster and written into the static environment map, while controlling the cotton spinning robot to perform avoidance operation.

2. The method according to claim 1, characterized in that, Step S1 includes: The original 3D point cloud data of the current frame is acquired, and the original 3D point cloud data is subjected to ground stripping and downsampling processing to obtain the target point cloud data. A clustering algorithm based on Euclidean distance is used to perform region growing on the target point cloud data to extract spatially continuous discrete point cloud clusters; For any discrete point cloud cluster, extract its voxel count sequence in multiple consecutive historical point clouds, and calculate the voxel density fluctuation rate based on the voxel count sequence. Obtain the lowest point height value of the discrete point cloud cluster in the vertical direction, and calculate the vertical axis suspension offset by combining the ground clearance parameter of the cotton spinning robot. A morphological closing operation is performed on the projection of the discrete point cloud cluster onto the horizontal plane, and the bottom support connectivity index is calculated based on the area ratio before and after the closing operation. The spatiotemporal dynamic feature vector of the discrete point cloud cluster is generated by combining the voxel density volatility, the vertical axis levitation offset, and the bottom support connectivity index.

3. The method according to claim 1, characterized in that, The step of mapping the suspended interference candidate clusters to a temporary observation buffer layer associated with the static environment map includes: Establish a temporary observation buffer layer with the same scale and coordinate system as the static environment map; When the floating interference candidate cluster is first mapped to the temporary observation buffer layer, an initial confidence coefficient is assigned to the grid cell covered by the floating interference candidate cluster; Based on the moving speed of the cotton spinning robot and the characteristic parameters of the ambient airflow, a time decay step size is configured for each grid cell in the temporary observation buffer layer. In each control cycle, the initial confidence coefficient is decremented according to the time decay step size. In subsequent frame observations, if the same spatial location is again marked as a floating interference candidate cluster, an accumulation correction operation is performed on the confidence coefficient of the grid cell.

4. The method according to claim 1, characterized in that, The construction of a hybrid cost raster based on the static environment map and the temporary observation buffer layer includes: Read the fixed obstacle attributes of each basic grid in the static environment map, and assign corresponding fixed cost values ​​based on the fixed obstacle attributes; Read the current confidence coefficient of the associated grid cell in the temporary observation buffer layer, multiply the confidence coefficient by a preset interference cost weight, and obtain the floating cost value of each associated grid cell; For overlapping basic grid cells and associated grid cells, the fixed cost value and the floating cost value are arithmetically superimposed to obtain the mixed cost value at the corresponding position in the mixed cost grid. The mixed cost grid is expanded, and the extent of the expansion is determined based on the contour envelope of the robotic arm in the contracted state of the cotton spinning robot.

5. The method according to claim 1, characterized in that, Step S5 includes: A heuristic search algorithm is used to search for the lowest cost path from the starting position to the target job position in the hybrid cost grid, which is then used as the initial job path; When the cotton spinning robot performs a moving operation according to the initial operation path, it acquires data of the temporary observation buffer layer within the preset viewing window range at a set frequency; Extract the confidence coefficients corresponding to all floating interference candidate clusters within the preset window range, and plot the trend curve of the confidence coefficients over time; Based on the slope change rate of the trend curve, the confidence state of the suspended interference candidate cluster at the next moment is predicted. If the predicted state is lower than the safety threshold, a path smooth transition instruction is generated in advance.

6. The method according to claim 1, characterized in that, Step S6 includes: The current confidence coefficient of each of the floating interference candidate clusters is compared with the safety threshold in real time. When the current confidence coefficient drops below the safety threshold, the corresponding grid cell is cleared from the temporary observation buffer layer. In the hybrid cost grid, the hybrid cost value of the cleared grid cell is updated to contain only the fixed cost value or set as the passage cost value; Extract the affected local path segments between the current position of the cotton spinning robot and the target working position, perform a local search on the updated hybrid cost grid, and generate replacement path segments; The curvature smoothing process is performed at the junction of the replacement path segment and the original path using spline curve interpolation method, so that the cotton spinning robot can smoothly cut into and pass through the area where the original suspended interference candidate cluster is located without changing the basic travel speed.

7. The method according to claim 1, characterized in that, Step S7 includes: If, within multiple consecutive control cycles, the confidence coefficient of a certain suspended interference candidate cluster remains above the safety threshold, and the boundary variance of the suspended interference candidate cluster on the horizontal plane converges to a preset stable interval, then the suspended interference candidate cluster is determined to be a solid obstacle. The spatial coordinates and contour information of the suspended interference candidate clusters identified as physical obstacles are converted into static occupancy data and written into the corresponding area in the static environment map. At the same time, the record of the suspended interference candidate cluster is deleted from the temporary observation buffer layer. The hybrid cost grid is refreshed based on the updated static environment map, and it is determined whether the initial operation path interferes with the physical obstacles. If interference occurs, the dynamic distance between the cotton spinning robot and the physical obstacle is calculated, and an emergency stop command is triggered when the dynamic distance is less than the braking threshold. At the same time, the detour operation path is replanned.

8. The method according to claim 4, characterized in that, During the movement of the cotton spinning robot along the initial work path, the method further includes: The joint angle data of the robotic arm mounted on the cotton spinning robot are acquired in real time, and the real-time sweeping radius of the robotic arm on the horizontal plane is calculated based on the joint angle data. The expansion range of the hybrid cost grid is dynamically adjusted based on the real-time sweep radius. When a floating interference candidate cluster with a floating cost value greater than zero but lower than the blocking threshold is detected in front of the mixed cost grid, a robotic arm retraction command is generated to reduce the real-time sweep radius.

9. The method according to claim 1, characterized in that, When the cotton spinning robot reaches the target work position and performs a grasping or placing operation, the method further includes: Monitor the local airflow disturbance data generated by the cotton spinning robot during the execution of its work actions, and assess the impact range of the local airflow disturbance data on the temporary observation buffer layer; When the influence range covers the scanning blind zone or effective field of view of the lidar, the confidence coefficient attenuation operation for the corresponding area in the temporary observation buffer layer is suspended, and the hybrid cost update for that area is frozen. After the cotton spinning robot completes its operation, the confidence coefficient decay operation on the temporary observation buffer layer is resumed after a preset time delay.

10. An intelligent cotton spinning robot autonomous mobile operation control system, used to implement the intelligent cotton spinning robot autonomous mobile operation control method according to claim 1, characterized in that, include: The 3D perception module is used to acquire the original 3D point cloud data of the environment in which the cotton spinning robot is located; The point cloud processing module is used to perform spatial clustering processing on the original three-dimensional point cloud data to extract several discrete point cloud clusters, and to calculate the spatiotemporal dynamic feature vector of each discrete point cloud cluster. The interference identification module is used to perform fly flower interference identification on each of the discrete point cloud clusters based on the spatiotemporal dynamic feature vectors to obtain floating interference candidate clusters and confidence obstacle clusters; The map management module is used to directly write the believed obstacle clusters into a static environment map for navigation, and to map the floating interference candidate clusters to a temporary observation buffer layer associated with the static environment map; The cost grid construction module is used to construct a hybrid cost grid based on the static environment map and the temporary observation buffer layer when the cotton spinning robot receives a moving operation instruction. The path planning module is used to perform path search based on the hybrid cost grid to generate an initial job path; The real-time monitoring module is used to monitor the change in confidence coefficient of suspended interference candidate clusters in the temporary observation buffer layer in real time as the cotton spinning robot moves along the initial operation path. The dynamic replanning module is used to remove the floating interference candidate cluster from the temporary observation buffer layer when the confidence coefficient of the floating interference candidate cluster decays to below the safety threshold, and to trigger local path replanning for the region where the floating interference candidate cluster is located. The motion control module is used to control the cotton spinning robot to smoothly pass through the area where the original suspended interference candidate cluster is located, and when the confidence coefficient of the suspended interference candidate cluster is continuously higher than the safety threshold and accompanied by spatial position convergence, the suspended interference candidate cluster is promoted to a confidence obstacle cluster and written into the static environment map, while controlling the cotton spinning robot to perform avoidance operation. The data storage module is used to store the static environment map, the temporary observation buffer layer, and related control parameters.

Images