A hook obstacle avoidance method and system for a tower crane

Abstract

This invention provides a method and system for obstacle avoidance of a tower crane hook, relating to the field of intelligent equipment technology. The method includes: collecting multi-source environmental perception data of the hook's operating area based on a multi-source heterogeneous sensing unit; analyzing and fusing the multi-source environmental perception data to obtain the state vector of the dynamic obstacle in a unified coordinate system; predicting the obstacle's motion trajectory and trajectory uncertainty measure within a future time period using a time-series prediction model based on the state vector and scene context; constructing a swing dynamics model based on the load parameters and hook state, transforming the swing angle constraint into a hook acceleration boundary; constructing a dynamic safety restricted area based on the obstacle's motion trajectory and trajectory uncertainty measure, and obtaining an integrated obstacle avoidance and swing suppression hook trajectory through online optimization using the acceleration boundary as a constraint; and inversely decomposing the integrated obstacle avoidance and swing suppression hook trajectory into multi-mechanism collaborative control commands, which are then sent to the tower crane's execution system.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment technology, specifically to a hook obstacle avoidance method and system for tower cranes. Background Technology

[0002] As a core hoisting equipment in engineering construction, tower cranes operate in complex environments with numerous dynamic obstacles. The coordinated control of hook obstacle avoidance and load sway suppression directly determines the safety and efficiency of the operation.

[0003] Existing tower crane obstacle avoidance technologies mostly rely on single sensing devices, which have problems such as blind spots in the perception of nearby obstacles, recognition deviations caused by obstruction by suspended objects, and insufficient accuracy of multi-source data fusion, making it difficult to accurately obtain complete status information and movement intentions of dynamic obstacles.

[0004] In the trajectory prediction stage, traditional models often analyze obstacle movement in isolation, which can easily lead to inaccurate obstacle avoidance decisions due to prediction lag or bias.

[0005] Therefore, a hook obstacle avoidance method and system for tower cranes is provided, which can meet the requirements of high-precision and high-reliability hook obstacle avoidance in complex construction scenarios and greatly improve the reliability of hook obstacle avoidance. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide a hook obstacle avoidance method and system for tower cranes, which can meet the requirements for high-precision and high-reliability hook obstacle avoidance in complex construction scenarios, and greatly improve the reliability of hook obstacle avoidance.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a hook obstacle avoidance method for a tower crane, the method comprising: Based on multi-source heterogeneous sensing units, multi-source environmental sensing data of the hook operation area are collected; Multi-source environmental perception data is analyzed and fused to obtain the state vector of dynamic obstacles in a unified coordinate system; Based on state vectors and scene context, a time-series prediction model is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period. A swing dynamics model is constructed based on the parameters of the suspended object and the state of the hook, and the swing angle constraint is transformed into the hook acceleration boundary. A dynamic safety restricted area is constructed based on the obstacle motion trajectory and trajectory uncertainty measurement, and the integrated obstacle avoidance and swing suppression hook trajectory is obtained through online optimization with acceleration boundary as constraint. Based on the kinematic model of the tower crane, the trajectory of the obstacle avoidance and sway suppression integrated hook is decomposed into multi-mechanism collaborative control commands, which are then sent to the tower crane execution system to drive the hook to track the motion trajectory while suppressing the sway of the suspended object.

[0008] Preferably, the multi-source heterogeneous sensing unit includes a tower crane top vision device, a radar and panoramic vision device below the hook, and personnel positioning tags; By analyzing and fusing multi-source environmental perception data, the state vector of dynamic obstacles in a unified coordinate system is obtained, including: Visual detection results obtained through a target detection model based on depth and color images acquired by the vision device on the top of the tower crane; Based on point cloud data collected by millimeter-wave radar below the hook, velocity-space joint clustering is used to obtain radar point cloud clustering results; High-precision positioning information is obtained by combining the time-series signal output by personnel positioning tags with inertial auxiliary information and performing Kalman filtering. Based on the pre-calibrated coordinate transformation matrix, the visual detection results, radar point cloud clustering results, and high-precision positioning information are uniformly mapped to the world coordinate system to obtain spatially aligned multimodal observation data. Based on the distorted images acquired by the panoramic vision device under the hook, the images are corrected by the homography matrix and projected onto the world coordinate system to generate fisheye verification observations; Based on the spatial overlap between fisheye verification observation and visual detection results, low-confidence visual detection results are eliminated; combined with construction task stage information and semantic map of the work area, the prior occurrence probability distribution of each type of dynamic obstacle is determined; based on the prior occurrence probability distribution, modal-level confidence weights are calculated for each modal observation data. Based on the observation results of the same obstacle candidate in different modes, calculate the cross-modal geometric consistency index; Based on the geometric consistency index and the modal-level confidence weights, the observation data of each modality are adjusted and corrected to obtain a multi-source observation set with enhanced consistency. The multi-source observation set is input into the joint state estimator, and semantic association is performed by combining personnel location tags and job role information to obtain the state vector of the dynamic obstacle in the world coordinate system. The state vector includes the current position, current speed, motion type, identity identifier and motion intention tag.

[0009] Preferably, based on state vectors and scene context, a time-series prediction model is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period, including: Based on the identity identifier in the state vector, the job roles corresponding to each dynamic obstacle are parsed, and the set of possible behavioral intentions is determined according to the preset job behavior rule library. Based on the current position, velocity, and acceleration of the hook, calculate the real-time social interaction force field generated by the hook on each dynamic obstacle; A dynamic interactive graph neural network is constructed, wherein the nodes include all dynamic obstacles and hooks; the node features include state vectors, job roles and prior behavioral intentions; and the edge weights are modulated by BIM access path constraints, historical movement patterns and the social interaction force field. The dynamic interaction graph is input into a multi-head Transformer decoder, with each decoder head corresponding to a behavioral intent, and the future trajectory under the behavioral intent is generated. Based on the future trajectories of each decoder and the prior probabilities of intent in the job behavior rule base, an occurrence probability is assigned to each future trajectory to obtain a set of multimodal obstacle motion trajectories; Based on the set of multimodal obstacle motion trajectories, the spatial coverage ellipsoid of each trajectory is calculated at each prediction time to obtain the uncertainty metric corresponding to the obstacle motion trajectory. The uncertainty metric includes the probability of occurrence of each behavioral intention and the spatial coverage ellipsoid parameters of each trajectory at each time.

[0010] Preferably, a swing dynamics model is constructed based on the parameters of the suspended object and the state of the hook, and the swing angle constraint is transformed into the hook acceleration boundary, including: Based on the mass of the hoisted object, the length of the wire rope, the slewing angular velocity of the tower crane, and the hoisting speed, a nonlinear dynamic equation containing radial and tangential swing degrees of freedom is constructed. Based on the safety management regulations of the construction site, the maximum permissible swing angle is determined to match the risk level of the hoisting operation. The risk level of the hoisting operation is determined by the presence of dynamic obstacles under the hoisted object, the hoisting height range, and the ambient wind speed. Based on the nonlinear dynamic equation, the current swing angle state, and the maximum allowable swing angle, the feasible region of the hook's acceleration in the horizontal plane is derived. The feasible region is an elliptical convex set with the current swing angle direction as the major axis. Based on the actual swing data fed back by the panoramic vision device below the hook, the air damping coefficient and the equivalent stiffness of the wire rope in the dynamic model are identified online, and the feasible acceleration domain is corrected in real time.

[0011] Preferably, a dynamic safety restricted area is constructed based on the obstacle's motion trajectory and trajectory uncertainty metric, and the integrated obstacle avoidance and sway suppression hook trajectory is obtained through online optimization using acceleration boundaries as constraints. Based on the uncertainty measure corresponding to the obstacle's motion trajectory, calculate the upper bound of the probability that the hook trajectory and the dynamic obstacle will collide in the future prediction time domain; Based on the maximum permissible collision risk threshold specified in the construction safety regulations, the minimum safe distance that the hook trajectory point needs to maintain at each moment is calculated in reverse, forming a probabilistic safety barrier function, and thus obtaining dynamic obstacle avoidance safety constraints. Based on the hook-load swing dynamics model and the maximum allowable swing angle constraint of the load, a safety metric function with the load swing angle as the state variable is defined, and a control barrier function that meets the preset requirements is constructed to obtain the swing stability constraint. Based on the dynamic obstacle avoidance safety constraints and the sway stability constraints, a joint safety barrier condition is constructed, which is then transformed into a quadratic inequality constraint to obtain a safe and feasible set. Based on the safe feasible set and the kinematic model of the tower crane, an objective function with trajectory tracking performance as the core is constructed. Combining the current position of the hook and the target lifting point, the objective function is solved through a preset optimization strategy, and the integrated obstacle avoidance and sway suppression hook trajectory is output.

[0012] Preferably, based on the tower crane kinematics model, the trajectory of the obstacle avoidance and sway suppression integrated hook is inversely decomposed into multi-mechanism collaborative control commands, including: Based on the kinematic model of the tower crane and the dynamic parameters of the luffing mechanism, slewing mechanism and hoisting mechanism, an electromechanical coupling inverse solution model including motor response delay, transmission friction and rotational inertia is constructed. Based on the electromechanical coupling inverse solution model, a differential flatness transformation is performed on the trajectory of the obstacle avoidance and sway suppression integrated hook to obtain basic collaborative control commands, which include luffing speed, slewing angular velocity and hoisting speed. Based on the hook-load swing dynamics model, the swing response of the load excited by the basic coordinated control command in the future time period is predicted to obtain the expected swing trajectory. Based on the deviation between the expected swing trajectory and the zero swing angle target, the sway suppression compensation amount is calculated; the sway suppression compensation amount includes high-frequency small-amplitude correction components for the rotational speed command and the amplitude speed command. The sway suppression compensation amount is superimposed on the basic cooperative control command to obtain the sway suppression enhanced cooperative control command; Based on the actual swing data fed back by the miniature panoramic vision device under the hook, the swing error between the actual swing trajectory and the expected swing trajectory is calculated. Based on the swing error, an adjustment feedback mechanism is used to adjust the amplitude gain and phase shift of the sway compensation amount online to form a closed-loop corrected sway compensation amount. The sway compensation amount after the closed-loop correction is superimposed with the basic cooperative control command to generate the final cooperative control command.

[0013] A second aspect of the present invention also provides a hook obstacle avoidance system for a tower crane, comprising: The acquisition module is used to acquire multi-source environmental sensing data of the hook operation area based on multi-source heterogeneous sensing units; The conversion module is used to analyze and fuse multi-source environmental perception data to obtain the state vector of dynamic obstacles in a unified coordinate system. The prediction module is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period based on the state vector and scene context, using a time-series prediction model. The module constructs a swing dynamics model based on the parameters of the suspended object and the state of the hook, and transforms the swing angle constraint into the hook acceleration boundary. The trajectory generation module is used to construct a dynamic safety restricted area based on the obstacle motion trajectory and trajectory uncertainty measurement, and obtain the integrated obstacle avoidance and swing suppression hook trajectory through online optimization with acceleration boundary as constraint. The conversion control module is used to convert the trajectory of the obstacle avoidance and sway suppression integrated hook into multi-mechanism collaborative control commands based on the tower crane kinematic model, and send them to the tower crane execution system to drive the hook to track the motion trajectory while suppressing the swing of the suspended object.

[0014] Compared with the prior art, the beneficial effects of the present invention are: Using the panoramic vision device directly below the hook as the core perception node, and combining it with the vision device on the top of the tower crane, the radar below the hook, and personnel positioning tags to form a multi-angle perception matrix, the system achieves blind-spot-free capture and distortion correction of near-distance obstacles around the suspended object by relying on the camera directly below the hook. Combined with the long-distance detection and positioning capabilities of other angle devices, a fusion system of "precise near-distance imaging - long-distance range coverage - multi-source data complementarity" is constructed. After coordinate unification, semantic verification, cross-modal consistency calibration, and job role association, a dynamic obstacle state vector containing motion intention in a unified coordinate system is output. The core application of the camera directly beneath the hook effectively solves problems such as occlusion by the suspended object and missed detection of close-range obstacles inherent in traditional single-angle sensing. Through distortion correction and precise imaging, it provides high-fidelity foundational data for low-confidence visual data elimination and cross-modal geometric consistency verification. Multi-angle sensing collaboration enables the system to cover the entire operation area while focusing on high-risk areas around the suspended object. Combined with other equipment, it enhances adaptability to complex scenarios, ultimately achieving a dual improvement in obstacle state recognition accuracy and positioning reliability. This provides accurate and comprehensive environmental data support for subsequent trajectory prediction and obstacle avoidance decisions, overcoming the technical pain points of blind spots in close-range obstacle perception and accurate fusion of multi-source sensing data in complex lifting scenarios. It breaks through the limitations of single sensing devices in terms of occlusion, accuracy, and environmental adaptability. Through fisheye verification to eliminate low-confidence data, cross-modal geometric consistency verification, and semantic prior empowerment, it significantly improves obstacle state recognition accuracy and positioning reliability, providing highly reliable data support for subsequent trajectory prediction and obstacle avoidance decisions. This solves the technical challenges of blurred dynamic obstacle perception and multi-source data conflicts in complex construction scenarios.

[0015] Based on a dynamic interactive graph neural network and a multi-head Transformer decoder, and incorporating a job behavior rule base, BIM access constraints, and a hook social interaction force field, a time-series prediction mechanism is constructed, consisting of "behavioral intent analysis – dynamic correlation modeling – multi-intent trajectory generation – uncertainty ellipsoid quantification." This mechanism outputs a set of multimodal obstacle trajectories containing occurrence probability and spatial coverage parameters. It overcomes the shortcomings of traditional prediction models that ignore the correlation between obstacle behavior and the randomness of trajectories, achieving accurate prediction and risk quantification of the future movement trajectory of dynamic obstacles. Its uncertainty measurement can fully cover the actual movement range, providing a refined basis for the construction of dynamic safety no-go zones and avoiding obstacle avoidance failures due to prediction lag or bias.

[0016] Based on the swing dynamics model, the swing angle constraint is transformed into an acceleration boundary. A probabilistic safety barrier is constructed by integrating obstacle trajectory uncertainties. A safe and feasible set is built through quadratic inequality constraints, achieving integrated online trajectory optimization for obstacle avoidance and swing suppression. Combining an electromechanical coupling inverse kinematics model and closed-loop swing suppression compensation, the trajectory is accurately converted into multi-mechanism collaborative control commands and dynamically corrected. This breaks the disconnect between obstacle avoidance and swing suppression control, ensuring that the hook trajectory meets dynamic obstacle avoidance safety requirements while strictly controlling the swing angle of the load within a safe threshold. The closed-loop correction mechanism can compensate for nonlinear effects such as motor delay and friction in real time, significantly improving trajectory tracking accuracy, effectively reducing the risk of lifting collisions and safety hazards caused by load swing, and enhancing the efficiency and stability of tower crane operations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a schematic diagram of a hook obstacle avoidance method for tower cranes.

[0019] Figure 2 This is a schematic diagram of a hook obstacle avoidance system for tower cranes. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] Example 1 This embodiment discloses a hook obstacle avoidance method for tower cranes, the method comprising: Based on a multi-source heterogeneous sensing unit, multi-source environmental sensing data of the hook operation area is collected; the multi-source heterogeneous sensing unit includes a tower crane top vision device, a radar and panoramic vision device below the hook, and personnel positioning tags. Multi-source environmental perception data is analyzed and fused to obtain the state vector of dynamic obstacles in a unified coordinate system; It should be noted that the analysis and fusion of multi-source environmental perception data yields the following state vectors of dynamic obstacles in a unified coordinate system: Based on depth and color images acquired by the vision device on top of the tower crane, a visual detection model is used to obtain visual detection results containing obstacle categories and two-dimensional bounding boxes. In this embodiment, the object detection model adopts the DETR model or its variant (such as Deformable DETR) based on the Transformer architecture, and its input is four-channel (RGB-D) data composed of aligned color and depth images. The visual detection results include: category, two-dimensional bounding box, confidence score, and corresponding depth value (mean or median of the region within the bounding box in the depth image).

[0023] Based on point cloud data acquired by a millimeter-wave radar beneath the hook, velocity-spatial joint clustering is used to obtain the radar point cloud clustering results. In this embodiment, the millimeter-wave radar is in the 77GHz band and has velocity measurement capabilities. Radar. Its raw output is a list of point clouds for each frame, with each point containing... That is, three-dimensional coordinates, radial velocity, and radar cross section.

[0024] The specific clustering logic is as follows: The features of each radar point are extracted from three-dimensional space... Extended to five-dimensional feature space .in, and This is a normalization coefficient used to balance the scales of spatial distance, velocity differences, and scattering intensity differences. Velocity dimension. The introduction of this method is key; it separates targets with similar spatial locations but vastly different speeds (such as stationary equipment and walking people) in a feature space. Considering the large variation in target density in the area below the hook, the clustering parameter (neighborhood radius) is crucial. and minimum points (This is not fixed.) Dynamically adjust based on the current average density of the point cloud; A lower value is then set based on experience to suit small targets (such as a single person). DBSCAN is performed on the five-dimensional feature space. The algorithm considers mutually reachable points (within the eps neighborhood) and the number of core points satisfies... Points are grouped into the same cluster. For each cluster, its centroid position, average velocity vector (fitted from point velocities), 3D bounding box size, and point cloud density are calculated. The radar point cloud clustering results are output: cluster ID, centroid... average speed Bounding box size, number of points, and main scattering type (judged as human body, metal device, etc. based on RCS mean).

[0025] Based on the time-series signal output by the personnel positioning tag, Kalman filtering is performed in conjunction with inertial auxiliary information to obtain high-precision positioning information. In this embodiment, the original distance observation sequence between the UWB tag worn by the personnel and at least four UWB base stations deployed at the construction site is measured using the two-way time-of-flight method. Inertial auxiliary information: data from the microelectromechanical system inertial measurement unit built into the tag, specifically including: triaxial accelerometer data: used to calculate displacement changes and detect motion status over a short period of time. Triaxial gyroscope data: used to estimate the tag's attitude changes (although personnel positioning does not have high requirements for absolute attitude, it can be used to assist in determining the direction of motion). Motion status detection: by analyzing the variance and pattern of the accelerometer data, the tag is determined in real time to be in a "stationary", "walking", "running", or "falling" state.

[0026] This embodiment uses error-state Kalman filtering to fuse UWB ranging and IMU data.

[0027] Using raw data from the gyroscope and accelerometer (excluding the currently estimated zero bias), the nominal states (position, velocity, attitude) are updated through integration using the inertial navigation mechanics equations. Simultaneously, the covariance matrix of the error states is updated based on the IMU's noise characteristics (accelerometer white noise, zero bias random walk, etc.). This is done when a set of UWB ranging values ​​from multiple base stations is received. At that time, the predicted ranging value is calculated based on the current nominal state and the base station location. The observation residual is The key innovation lies in the observation model: when an abnormally large observation residual is detected for a certain link (exceeding a threshold based on historical statistics), the weight of the observation is temporarily reduced or it is marked as an NLOS (non-line-of-sight) observation for deweighting, combining the geometric relationship of the link and the motion consistency inferred from the IMU, instead of using it directly. The filter output is a corrected nominal state, i.e., high-precision positioning information is a timestamp and position. ,speed Location mode , Horizontal positioning accuracy factor.

[0028] Based on the pre-calibrated coordinate transformation matrix, the visual detection results, radar point cloud clustering results, and high-precision positioning information are uniformly mapped to the world coordinate system to obtain spatially aligned multimodal observation data. Based on the distorted images acquired by the panoramic vision device under the hook, the images are corrected by the homography matrix and projected onto the world coordinate system to generate fisheye verification observations; Based on the spatial overlap between fisheye verification observation and visual detection results, low-confidence visual detection results are eliminated; combined with construction task stage information and semantic map of the work area, the prior occurrence probability distribution of each type of dynamic obstacle is determined; based on the prior occurrence probability distribution, modal-level confidence weights are calculated for each modal observation data. Based on the observation results of the same obstacle candidate in different modalities, a cross-modal geometric consistency index is calculated. The geometric consistency index includes the Euclidean distance between the center of the visual frame and the centroid of the radar point cloud, and the angle between the velocity vectors of the localization position and the visual position. Based on the geometric consistency index and the modal-level confidence weights, the observation data of each modality are adjusted and corrected to obtain a multi-source observation set with enhanced consistency. The multi-source observation set is input into the joint state estimator, and semantic association is performed by combining personnel location tags and job role information to obtain the state vector of the dynamic obstacle in the world coordinate system. The state vector includes the current position, current velocity, motion type, identity identifier, and motion intention tag. In this embodiment, the joint state estimator is a system combining a multi-hypothesis tracker and a semantic knowledge base. For each observation in the current frame (from vision, radar, and UWB), it is associated with the existing target trajectory. The association cost function comprehensively considers: Motion cost: Mahalanobis distance between the predicted position and the observed position based on Kalman filter.

[0029] Appearance cost: For visual observation, the cosine similarity between appearance features extracted using the ReID network and historical appearance features of the trajectory.

[0030] Semantic cost: Consistency between the historical type of the trajectory (e.g., 'personnel') and the current observation type. Job role information comes into play here: for example, if the UWB tag ID is bound to "signalman", its reasonable area of ​​activity is limited to the vicinity of the hoisting command area. If the tag is observed to appear below the tower crane cab, a higher semantic conflict cost will occur.

[0031] Successfully associated observations are used to update the Kalman filter state of the corresponding trajectory. For multimodal observation fusion, covariance intersection or weighted fusion based on modal-level confidence weights is used to address the varying uncertainties of observations from different sensors. New trajectories are generated, trajectories are confirmed, and lost trajectories are deleted. Motion intent labels are generated based on: the trajectory's short-term historical path, target type, and positional relationship relative to the hook and the preset work area. For example, if a "personnel" trajectory moves rapidly from the rest area directly towards the projected area of ​​the suspended object, its motion intent might be labeled as "entering a risk area."

[0032] Based on state vectors and scene context, a time-series prediction model is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period. It should be noted that, based on the state vector and scene context, the temporal prediction model predicts the obstacle's trajectory and trajectory uncertainty measure within a future time period, including: Based on the identity identifier in the state vector, the job role corresponding to each dynamic obstacle is parsed, and according to the preset job behavior rule library, the set of possible behavioral intentions is determined. The behavioral intentions include one of passing, avoiding, staying or cooperating. Based on the hook's current position, velocity, and acceleration, a real-time social interaction force field is calculated on the hook's interaction with each dynamic obstacle. This social interaction force field strengthens as the hook approaches the obstacle. In this embodiment, the social interaction force field... It is a scalar field representing the intensity of the psychological influence of the hook on the behavior of an obstacle. It consists of three parts, located at the obstacle's position. The calculation formula at the location is: In the formula, For spatial proximity forces, For the threat of movement, For enhanced sound and light warning capabilities, , , The corresponding coefficients can be calibrated using field data.

[0033] A dynamic interactive graph neural network is constructed, wherein the nodes include all dynamic obstacles and hooks; the node features include state vectors, job roles and prior behavioral intentions; and the edge weights are modulated by BIM access path constraints, historical movement patterns and the social interaction force field. In this embodiment, node feature encoding is used, and the feature vector of each node i is... It is a concatenated vector: ; Let be the position and velocity of the state vector. The vector for job roles is obtained through a learnable embedding layer. The intent probability distribution vector is obtained from the rule base, and social_force is the social interaction force field value at the location of this node. This value is 0 for hook nodes.

[0034] The initial weight of the edge connecting nodes i and j is calculated using the following formula: In the formula, The path accessibility factor is calculated by querying the BIM model to determine if there is a path from position i to j that is not blocked by permanent walls or equipment. If so, the value is 1; if completely blocked, the value is 0; if a detour is required, the value is a decay function of the path length. The historical coordination factor is used to calculate the correlation (Pearson correlation coefficient) between the velocity vectors of i and j over a past period. A higher value indicates a greater likelihood of historically coordinated movement (like that of a work group), and a larger edge weight. This is the social force coupling factor. If i or j is a hook, this factor is the social_force of the corresponding node. If they are two obstacles, the similarity (cosine similarity) of their respective social_forces is calculated. A high similarity means that they may react similarly to the same threat, and the edge weight is increased.

[0035] Message passing is performed using a 2-3 layer graph attention network. In each layer, nodes aggregate the features of their neighbors, with weights determined by the dynamic edge weights described above. This is determined together with the attention coefficient. The final output of each node incorporates contextual features that integrate global interaction information.

[0036] The dynamic interaction graph is input into a multi-head Transformer decoder, with each decoder head corresponding to a behavioral intent, and the future trajectory under the behavioral intent is generated. Based on the future trajectories of each decoder and the prior probabilities of intent in the job behavior rule base, an occurrence probability is assigned to each future trajectory to obtain a set of multimodal obstacle motion trajectories; Based on the set of multimodal obstacle motion trajectories, the spatial coverage ellipsoid of each trajectory is calculated at each prediction time to obtain the uncertainty measure corresponding to the obstacle motion trajectory. The uncertainty measure includes the probability of occurrence of each behavioral intention and the spatial coverage ellipsoid parameters of each trajectory at each time. The scale of the spatial coverage ellipsoid is jointly determined by the trajectory variance and the perceived uncertainty of the input state vector.

[0037] A swing dynamics model is constructed based on the parameters of the suspended object and the state of the hook, and the swing angle constraint is transformed into the hook acceleration boundary. Specifically, a swing dynamics model is constructed based on the parameters of the suspended object and the state of the hook, and the swing angle constraint is transformed into the hook acceleration boundary, including: Based on the mass of the hoisted object, the length of the wire rope, the slewing angular velocity of the tower crane, and the hoisting speed, a nonlinear Lagrange dynamic equation containing radial and tangential swing degrees of freedom is constructed. Based on the safety management regulations of the construction site, a maximum permissible swing angle matching the risk level of the hoisting operation is determined. The risk level of the hoisting operation is jointly determined by the presence and status of dynamic obstacles below the hoisted object, the hoisting height range, and the ambient wind speed. In this embodiment, risk factors are collected and normalized, including personnel intrusion risk. The system monitors whether there are dynamic obstacles (especially personnel) in the projection area directly below the hoisted object. An intrusion risk coefficient is calculated based on the number of intruding targets, their distance from the center of the hoisted object, and their movement speed. Height risk: The system calculates a height risk coefficient based on the current height of the hoisted object above the ground, combined with preset "low, medium, and high" risk height ranges. The higher the height, the greater the risk coefficient. Environmental risk: The system accesses meteorological station data to obtain real-time wind speed. The higher the wind speed, the more uncontrollable the impact on the swing of the hoisted object, and an environmental risk coefficient is calculated accordingly. The above three risk coefficients are weighted and summed according to preset weights to obtain a comprehensive risk index.

[0038] The system has a pre-installed risk level-swing angle limit lookup table. This table divides the comprehensive risk index into several levels, such as "low risk," "medium risk," and "high risk," with each level corresponding to a maximum permissible swing angle value with varying degrees of stringency. Based on the real-time calculated comprehensive risk index, the applicable dynamic maximum permissible swing angle can be obtained from the table.

[0039] Based on the aforementioned nonlinear dynamic equations, the current swing angle state, and the maximum permissible swing angle, the feasible region of acceleration of the hook in the horizontal plane is derived using the Lyapunov stability condition. This feasible region is an elliptical convex set with the current swing angle direction as its major axis. In this embodiment, the elliptical region is characterized by its major axis always aligning with the current swing direction of the suspended object. The overall size of the ellipse is determined by the "margin" between the current swing angle and the maximum permissible swing angle. The smaller the margin, the smaller the ellipse contracts, meaning a more stringent allowable acceleration range and requiring gentler control. The narrowest direction of the ellipse is perpendicular to the current swing direction.

[0040] Based on the actual swing data fed back by the panoramic vision device below the hook, the air damping coefficient and the equivalent stiffness of the wire rope in the dynamic model are identified online, and the feasible region of acceleration is corrected in real time. In this embodiment, the panoramic camera below the hook continuously captures images of the suspended object. Through image algorithms, specific marker points on the suspended object are tracked in real time, or video motion analysis is used to directly measure the two-dimensional swing angle sequence of the suspended object in the image. After coordinate system transformation, these data are converted into actual radial and tangential swing angle observations in the world coordinate system.

[0041] A dynamic safety restricted area is constructed based on the obstacle motion trajectory and trajectory uncertainty measurement, and the integrated obstacle avoidance and swing suppression hook trajectory is obtained through online optimization with acceleration boundary as constraint. It should be noted that a dynamic safety restricted area is constructed based on the obstacle movement trajectory and trajectory uncertainty measurement, and the integrated obstacle avoidance and sway suppression hook trajectory is obtained through online optimization using acceleration boundaries as constraints. Based on the uncertainty metric corresponding to the obstacle's trajectory, an upper bound on the probability of a collision between the hook trajectory and the dynamic obstacle in the future prediction time domain is calculated. This upper bound is conservatively estimated using the Gaussian distribution tail inequality or Chebyshev inequality. In this embodiment, for each possible trajectory (and its uncertainty ellipsoid) of each dynamic obstacle, we do not precisely calculate complex integral probabilities. Instead, we employ a computationally efficient and absolutely conservative estimation method to estimate the collision probability. Specifically, the positional error of the obstacle in the short-term prediction follows a diffusion process (described by the uncertainty ellipsoid). We calculate the minimum distance between a point on the hook trajectory and the worst-case position of the obstacle (i.e., the error diffuses in the direction of maximum probability). Then, using a pre-defined lookup table related to the diffusion coefficient, this minimum distance is mapped to an upper bound on the collision probability.

[0042] For each future time point t on the hook trajectory, the system iterates through all obstacles to find the most stringent upper bound requirement for the collision probability at that moment. The hook trajectory point must maintain a distance of at least D_safe(t) from the uncertainty ellipsoid boundary of the corresponding obstacle at that moment. D_safe(t) is not a fixed value, but varies with time: the further the prediction, the greater the uncertainty, and the larger the required safe distance D_safe(t).

[0043] Based on the maximum permissible collision risk threshold specified in the construction safety regulations, the minimum safe distance that the hook trajectory point needs to maintain at each moment is calculated to form a probabilistic safety barrier function, resulting in dynamic obstacle avoidance safety constraints. In this embodiment, hard constraints and soft constraints are defined. For hard constraints, a classic zero-order CBF is constructed, the form of which ensures that as long as the initial state is safe, there exists a control input that prevents the system from going out of bounds. This directly corresponds to the elliptical feasible region of acceleration derived from the dynamic model. For soft constraints, a first-order CBF is constructed, which not only requires the state to be within the safe set but also requires the state to evolve in the direction of reducing the swing angle. This guides the system to actively suppress swaying. The elliptical feasible region of acceleration (from the hard constraints) is used as the basic constraint set of the optimization problem. Based on this, the first-order CBF condition (from the soft constraints) is used as a penalty term in the optimization objective.

[0044] Based on the hook-load swing dynamics model and the maximum allowable swing angle constraint of the load, a safety metric function with the load swing angle as the state variable is defined, and a control barrier function that meets the preset requirements is constructed to obtain the swing stability constraint. Based on the dynamic obstacle avoidance safety constraints and the sway stability constraints, a joint safety barrier condition is constructed, and the joint safety barrier condition is transformed into a quadratic inequality constraint to obtain a safe and feasible set. Based on the aforementioned safe and feasible set and the tower crane kinematic model, an objective function with trajectory tracking performance as its core is constructed. Combining the current position of the hook and the target lifting point, the objective function is solved through a preset optimization strategy, and the integrated obstacle avoidance and sway suppression hook trajectory is output.

[0045] In this embodiment, the decision variables are defined as a series of horizontal position points of the hook within a finite future prediction time domain. The objective function consists of a weighted sum of three parts: first, the square of the terminal position tracking error, ensuring that the hook eventually reaches the target point; second, the sum of the squares of the trajectory jerk, used to penalize abrupt velocity changes and ensure smooth motion; and third, an active sway control term. All constraints, including the dynamic obstacle avoidance safety distance constraint, the hook acceleration ellipse feasible region constraint, and the velocity and acceleration limit constraints of each mechanism of the tower crane, are organized into a standard form.

[0046] The preset optimization strategy employs a sequential convex optimization method: at the beginning of each iteration, based on the optimal trajectory guess obtained from the previous iteration, a first-order Taylor expansion is performed on all non-convex obstacle avoidance constraints, approximating them as linear inequality constraints with respect to the current decision variable. Through this step, the original complex non-convex problem is transformed into a standard and easily solvable quadratic programming problem in this iteration.

[0047] To meet the millisecond-level response requirements of tower crane real-time control, the highly computationally efficient active set algorithm is used to solve the aforementioned quadratic programming problem. The key lies in introducing a "warm start" mechanism: the optimal solution calculated in the previous control cycle is shifted and extended as the initial guess value for the current solution, and its active constraint set is initialized simultaneously.

[0048] Using the current solution of the quadratic programming problem as a new linearization guess point, the corresponding steps are repeated for sequential iteration. A dual convergence criterion is set: first, the change in the objective function value is less than a set threshold; second, the change in the trajectory itself is sufficiently small.

[0049] When the solver reports no feasible solution on the first iteration, it indicates that the current environmental constraints are too stringent (such as being completely blocked by dynamic obstacles). The system immediately initiates a tiered relaxation strategy: first, it temporarily removes the active sway-damping term from the objective function to fully ensure obstacle avoidance; if this is still infeasible, it gradually relaxes the conservative estimate of the safety distance by a preset step size until the problem is solvable. As a final safeguard, if there is absolutely no solution, the system will switch to "safe hovering mode," plan a trajectory that smoothly decelerates the hook to a standstill, and simultaneously issue the highest-level audible and visual alarm to the operator, requesting manual intervention.

[0050] Based on the kinematic model of the tower crane, the trajectory of the obstacle avoidance and sway suppression integrated hook is decomposed into multi-mechanism collaborative control commands, which are then sent to the tower crane execution system to drive the hook to track the motion trajectory while suppressing the sway of the suspended object.

[0051] It should be noted that, based on the tower crane kinematic model, the inverse kinematics of the obstacle avoidance and sway suppression integrated hook trajectory is converted into multi-mechanism collaborative control commands, including: Based on the kinematic model of the tower crane and the dynamic parameters of the luffing mechanism, slewing mechanism, and hoisting mechanism, an electromechanical coupling inverse kinematic model is constructed, incorporating motor response delay, transmission friction, and rotational inertia. In this embodiment, the electromechanical coupling inverse kinematic model is specifically implemented by establishing a dynamic transfer function model for each mechanism, from "speed command to motor electromagnetic torque to transmission system deformation and friction to mechanism output speed / position." The motor response delay is simulated using a first-order inertial element; the transmission friction employs a model incorporating Coulomb friction and viscous friction. The model; the moment of inertia is a time-varying parameter that varies with the position of the hook and the weight of the load, and needs to be calculated online.

[0052] Based on the electromechanical coupling inverse solution model, a differential flatness transformation is performed on the hook motion trajectory to obtain the basic coordinated control command, which includes luffing speed, slewing angular velocity and hoisting speed. Based on the hook-load swing dynamics model, the swing response of the load excited by the basic coordinated control command in the future time period is predicted to obtain the expected swing trajectory. Based on the deviation between the expected swing trajectory and the zero swing angle target, the sway suppression compensation amount is calculated; the sway suppression compensation amount includes high-frequency small-amplitude correction components for the rotational angular velocity command and the amplitude-changing speed command; in this embodiment, the calculation of high-frequency small-amplitude correction classification is performed through a feedforward shaper and a disturbance observer, specifically, the feedforward shaper processes the basic control command... (Such as variable amplitude speed) passes through a digital filter. The design goal of this filter is to make the filter output... When applied to the simplified swing angle model, the resulting residual oscillation is minimized. Simultaneously, a disturbance observer is run. This observer takes the "base command" and the "actual hook motion" (obtainable from encoder feedback) as inputs and estimates an equivalent "unmodeled oscillation disturbance force" in real time. This disturbance force essentially incorporates factors such as model error, wind disturbance, and wire rope torsion.

[0053] Final swing compensation It is a weighted sum of the correction part of the feedforward shaper output and the feedback correction part generated based on the disturbance observer estimate. The specific calculation formula is as follows: In the formula, , For the corresponding adjustable weights, For model-based design gain, For disturbance forces.

[0054] The sway suppression compensation amount is superimposed on the basic cooperative control command to obtain the sway suppression enhanced cooperative control command; Based on the actual swing data fed back by the miniature panoramic vision device under the hook, the swing error between the actual swing trajectory and the expected swing trajectory is calculated. Based on the swing error, an adjustment feedback mechanism is used to adjust the amplitude gain and phase shift of the sway compensation amount online to form a closed-loop corrected sway compensation amount. The sway compensation amount after the closed-loop correction is superimposed with the basic cooperative control command to generate the final cooperative control command.

[0055] This embodiment also discloses a hook obstacle avoidance system for tower cranes, including: The acquisition module is used to acquire multi-source environmental sensing data of the hook operation area based on multi-source heterogeneous sensing units; The conversion module is used to analyze and fuse multi-source environmental perception data to obtain the state vector of dynamic obstacles in a unified coordinate system. The prediction module is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period based on the state vector and scene context, using a time-series prediction model. The module constructs a swing dynamics model based on the parameters of the suspended object and the state of the hook, and transforms the swing angle constraint into the hook acceleration boundary. The trajectory generation module is used to construct a dynamic safety restricted area based on the obstacle motion trajectory and trajectory uncertainty measurement, and obtain the integrated obstacle avoidance and swing suppression hook trajectory through online optimization with acceleration boundary as constraint. The conversion control module is used to convert the trajectory of the obstacle avoidance and sway suppression integrated hook into multi-mechanism collaborative control commands based on the tower crane kinematic model, and send them to the tower crane execution system to drive the hook to track the motion trajectory while suppressing the swing of the suspended object.

[0056] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0057] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0058] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of this application.

[0059] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0060] In the several embodiments provided in this application, it should be understood that the disclosed application can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0061] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0062] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0063] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for obstacle avoidance of a hook for a tower crane, characterized in that, The method includes: Based on multi-source heterogeneous sensing units, multi-source environmental sensing data of the hook operation area are collected; Multi-source environmental perception data is analyzed and fused to obtain the state vector of dynamic obstacles in a unified coordinate system; Based on state vectors and scene context, a time-series prediction model is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period. A swing dynamics model is constructed based on the parameters of the suspended object and the state of the hook, and the swing angle constraint is transformed into the hook acceleration boundary. A dynamic safety restricted area is constructed based on the obstacle motion trajectory and trajectory uncertainty measurement, and the integrated obstacle avoidance and swing suppression hook trajectory is obtained through online optimization with acceleration boundary as constraint. Based on the kinematic model of the tower crane, the trajectory of the obstacle avoidance and sway suppression integrated hook is decomposed into multi-mechanism collaborative control commands, which are then sent to the tower crane execution system to drive the hook to track the motion trajectory while suppressing the sway of the suspended object.

2. The method for hook obstacle avoidance for a tower crane according to claim 1, characterized in that, The multi-source heterogeneous sensing unit includes a tower crane top vision device, a radar and panoramic vision device below the hook, and personnel positioning tags. By analyzing and fusing multi-source environmental perception data, the state vector of dynamic obstacles in a unified coordinate system is obtained, including: Visual detection results obtained through a target detection model based on depth and color images acquired by the vision device on the top of the tower crane; Based on point cloud data collected by millimeter-wave radar below the hook, velocity-space joint clustering is used to obtain radar point cloud clustering results; High-precision positioning information is obtained by combining the time-series signal output by personnel positioning tags with inertial auxiliary information and performing Kalman filtering. Based on the pre-calibrated coordinate transformation matrix, the visual detection results, radar point cloud clustering results, and high-precision positioning information are uniformly mapped to the world coordinate system to obtain spatially aligned multimodal observation data. Based on the distorted images acquired by the panoramic vision device under the hook, the images are corrected by the homography matrix and projected onto the world coordinate system to generate fisheye verification observations; Based on the spatial overlap between fisheye verification observation and visual detection results, low-confidence visual detection results are eliminated; combined with construction task stage information and semantic map of the work area, the prior occurrence probability distribution of each type of dynamic obstacle is determined; based on the prior occurrence probability distribution, modal-level confidence weights are calculated for each modal observation data. Based on the observation results of the same obstacle candidate in different modes, calculate the cross-modal geometric consistency index; Based on the geometric consistency index and the modal-level confidence weights, the observation data of each modality are adjusted and corrected to obtain a multi-source observation set with enhanced consistency. The multi-source observation set is input into the joint state estimator, and semantic association is performed by combining personnel location tags and job role information to obtain the state vector of the dynamic obstacle in the world coordinate system. The state vector includes the current position, current speed, motion type, identity identifier and motion intention tag.

3. The hook obstacle avoidance method for a tower crane according to claim 2, characterized in that, Based on state vectors and scene context, a time-series prediction model is used to predict the obstacle's trajectory over a future time period, along with a trajectory uncertainty metric, including: Based on the identity identifier in the state vector, the job roles corresponding to each dynamic obstacle are parsed, and the set of possible behavioral intentions is determined according to the preset job behavior rule library. Based on the current position, velocity, and acceleration of the hook, calculate the real-time social interaction force field generated by the hook on each dynamic obstacle; A dynamic interactive graph neural network is constructed, wherein the nodes include all dynamic obstacles and hooks; the node features include state vectors, job roles and prior behavioral intentions; and the edge weights are modulated by BIM access path constraints, historical movement patterns and the social interaction force field. The dynamic interaction graph is input into a multi-head Transformer decoder, with each decoder head corresponding to a behavioral intent, and the future trajectory under the behavioral intent is generated. Based on the future trajectories of each decoder and the prior probabilities of intent in the job behavior rule base, an occurrence probability is assigned to each future trajectory to obtain a set of multimodal obstacle motion trajectories; Based on the set of multimodal obstacle motion trajectories, the spatial coverage ellipsoid of each trajectory is calculated at each prediction time to obtain the uncertainty metric corresponding to the obstacle motion trajectory. The uncertainty metric includes the probability of occurrence of each behavioral intention and the spatial coverage ellipsoid parameters of each trajectory at each time.

4. The hook obstacle avoidance method for a tower crane according to claim 3, characterized in that, A swing dynamics model is constructed based on the parameters of the suspended object and the state of the hook, and the swing angle constraint is transformed into the hook acceleration boundary, including: Based on the mass of the hoisted object, the length of the wire rope, the slewing angular velocity of the tower crane, and the hoisting speed, a nonlinear dynamic equation containing radial and tangential swing degrees of freedom is constructed. Based on the safety management regulations of the construction site, the maximum permissible swing angle is determined to match the risk level of the hoisting operation. The risk level of the hoisting operation is determined by the presence of dynamic obstacles under the hoisted object, the hoisting height range, and the ambient wind speed. Based on the nonlinear dynamic equation, the current swing angle state, and the maximum allowable swing angle, the feasible region of the hook's acceleration in the horizontal plane is derived. The feasible region is an elliptical convex set with the current swing angle direction as the major axis. Based on the actual swing data fed back by the panoramic vision device below the hook, the air damping coefficient and the equivalent stiffness of the wire rope in the dynamic model are identified online, and the feasible acceleration domain is corrected in real time.

5. A hook obstacle avoidance method for a tower crane according to claim 4, characterized in that, A dynamic safety restricted area is constructed based on obstacle motion trajectory and trajectory uncertainty measurement. Constrained by acceleration boundaries, the integrated obstacle avoidance and sway suppression hook trajectory is obtained through online optimization, including: Based on the uncertainty measure corresponding to the obstacle's motion trajectory, calculate the upper bound of the probability that the hook trajectory and the dynamic obstacle will collide in the future prediction time domain; Based on the maximum permissible collision risk threshold specified in the construction safety regulations, the minimum safe distance that the hook trajectory point needs to maintain at each moment is calculated in reverse, forming a probabilistic safety barrier function, and thus obtaining dynamic obstacle avoidance safety constraints. Based on the hook-load swing dynamics model and the maximum allowable swing angle constraint of the load, a safety metric function with the load swing angle as the state variable is defined, and a control barrier function that meets the preset requirements is constructed to obtain the swing stability constraint. Based on the dynamic obstacle avoidance safety constraints and the sway stability constraints, a joint safety barrier condition is constructed, which is then transformed into a quadratic inequality constraint to obtain a safe and feasible set. Based on the safe feasible set and the kinematic model of the tower crane, an objective function with trajectory tracking performance as the core is constructed. Combining the current position of the hook and the target lifting point, the objective function is solved through a preset optimization strategy, and the integrated obstacle avoidance and sway suppression hook trajectory is output.

6. A method for hook obstacle avoidance for a tower crane according to claim 5, characterized in that, Based on the tower crane kinematics model, the trajectory of the obstacle avoidance and sway suppression integrated hook is decomposed into multi-mechanism collaborative control commands, including: Based on the kinematic model of the tower crane and the dynamic parameters of the luffing mechanism, slewing mechanism and hoisting mechanism, an electromechanical coupling inverse solution model including motor response delay, transmission friction and rotational inertia is constructed. Based on the electromechanical coupling inverse solution model, a differential flatness transformation is performed on the trajectory of the obstacle avoidance and sway suppression integrated hook to obtain basic collaborative control commands, which include luffing speed, slewing angular velocity and hoisting speed. Based on the hook-load swing dynamics model, the swing response of the load excited by the basic coordinated control command in the future time period is predicted to obtain the expected swing trajectory. Based on the deviation between the expected swing trajectory and the zero swing angle target, the sway suppression compensation amount is calculated; the sway suppression compensation amount includes high-frequency small-amplitude correction components for the rotational speed command and the amplitude speed command. The sway suppression compensation amount is superimposed on the basic cooperative control command to obtain the sway suppression enhanced cooperative control command; Based on the actual swing data fed back by the miniature panoramic vision device under the hook, the swing error between the actual swing trajectory and the expected swing trajectory is calculated. Based on the swing error, an adjustment feedback mechanism is used to adjust the amplitude gain and phase shift of the sway compensation amount online to form a closed-loop corrected sway compensation amount. The sway compensation amount after the closed-loop correction is superimposed with the basic cooperative control command to generate the final cooperative control command.

7. A hook obstacle avoidance system for a tower crane, implementing the hook obstacle avoidance method for a tower crane as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to acquire multi-source environmental sensing data of the hook operation area based on multi-source heterogeneous sensing units; The multi-source heterogeneous sensing unit includes a tower crane top vision device, a radar and panoramic vision device below the hook, and personnel positioning tags. The conversion module is used to analyze and fuse multi-source environmental perception data to obtain the state vector of dynamic obstacles in a unified coordinate system. The prediction module is used to predict the obstacle's trajectory and trajectory uncertainty measure within a future time period based on the state vector and scene context, using a time-series prediction model. The module constructs a swing dynamics model based on the parameters of the suspended object and the state of the hook, and transforms the swing angle constraint into the hook acceleration boundary. The trajectory generation module is used to construct a dynamic safety restricted area based on the obstacle motion trajectory and trajectory uncertainty measurement, and obtain the integrated obstacle avoidance and swing suppression hook trajectory through online optimization with acceleration boundary as constraint. The conversion control module is used to convert the trajectory of the obstacle avoidance and sway suppression integrated hook into multi-mechanism collaborative control commands based on the tower crane kinematic model, and send them to the tower crane execution system to drive the hook to track the motion trajectory while suppressing the swing of the suspended object.

Images