Bridge crane obstacle avoidance system and method based on load adaptation and space-time prediction

Abstract

The application belongs to the technical field of crane automation control and path planning, in particular to a bridge crane obstacle avoidance system and method based on load self-adaptation and space-time prediction, comprising a perception module, a prediction module, a self-adaptive planning module and a trajectory generation module; global map information and load quality are mainly obtained through multi-sensor fusion technology, state estimation of dynamic obstacles is carried out by using extended Kalman filtering, and future multi-step space-time prediction trajectories are generated; at the same time, the maximum allowable acceleration under the current working condition is inversely solved according to the real-time load quality and the dynamic model; an improved dynamic window method is constructed, the obtained maximum allowable acceleration is taken as the dynamic boundary of the speed sampling space, and collision risk detection is carried out between each time step of the crane deduced trajectory and the corresponding time step of the obstacle prediction trajectory, and finally the optimal obstacle avoidance trajectory is output. The technical scheme of the application can solve the problem of dynamic obstacle avoidance lag, realize load self-adaptation and strong robustness.

Description

Technical Field

[0001] This invention relates to the field of crane automation control and path planning technology, and in particular to an obstacle avoidance system and method for bridge cranes based on load adaptation and spatiotemporal prediction. Background Technology

[0002] With the development of intelligent manufacturing, bridge cranes, as crucial logistics handling facilities in industries such as ports, docks, metallurgy, and warehousing, are undergoing a transformation in their control mode from traditional manual operation to automated and intelligent operation. However, the actual operating environment of bridge cranes is usually quite complex. In addition to various static goods, there may be dynamic moving targets such as AGVs and ground personnel in the working area. If the crane has a blind spot in perception or a delayed obstacle avoidance response during operation, it is very easy to cause collisions between the load and the goods, and may even cause personal injury accidents. For the existing dynamic obstacle avoidance technology of bridge cranes, there are still the following problems: (1) Delay in perception and planning. Existing technologies are mostly based on static map planning and are difficult to handle dynamic obstacles. Braking is often triggered only when the obstacle enters the danger zone, resulting in untimely obstacle avoidance; (2) Ignoring changes in load inertia. The weight of the crane varies each time it operates, and the braking distance is too different when it is unloaded and fully loaded. The existing DWA algorithm (dynamic window method) usually limits the maximum acceleration to a fixed value. If it is set according to the unload setting, the braking distance is too long when it is loaded. If it is set according to the load setting, the efficiency will be very low when it is unloaded. (3) Lack of prediction ability. The DWA algorithm usually treats dynamic obstacles as instantaneous static obstacles and lacks prediction of the movement trend of obstacles, resulting in the obstacle avoidance system lacking foresight. Summary of the Invention

[0003] To address the aforementioned technical problems, the present invention aims to provide a bridge crane obstacle avoidance system based on load adaptation and spatiotemporal prediction, comprising:

[0004] Perception module: It uses multi-sensor fusion to acquire blind-spot-free global map information and uses a pin-type load cell to acquire the load mass in real time;

[0005] Prediction module: Based on the blind-spot-free global map information obtained by the perception module, extract the centroid coordinate observation values ​​of dynamic obstacles, use the extended Kalman filter algorithm to fuse the centroid coordinate observation values ​​of dynamic obstacles to perform state estimation of dynamic obstacles, obtain the optimal state estimate of dynamic obstacles at the current moment, and generate the predicted trajectory of dynamic obstacles.

[0006] Adaptive planning module: Based on the load mass acquired in real time by the sensing module, the maximum allowable acceleration of the crane under the current working condition is solved by using the dynamic model of the load; an improved dynamic window method is constructed, and the obtained maximum allowable acceleration is used as the dynamic boundary of the velocity sampling space for velocity sampling to obtain the crane's deduced trajectory;

[0007] Trajectory generation module: Perform collision risk detection on each time step of the crane's simulated trajectory and the corresponding time step of the predicted trajectory of dynamic obstacles, generate multiple candidate obstacle avoidance trajectories, select the optimal obstacle avoidance trajectory from the candidate obstacle avoidance trajectories, and output the optimal obstacle avoidance trajectory.

[0008] This invention also provides a method for obstacle avoidance of a bridge crane based on load adaptation and spatiotemporal prediction, comprising the following steps:

[0009] S1. Multi-sensor fusion is used to obtain blind-spot-free global map information, and a pin-type load cell is used to obtain the load mass in real time.

[0010] S2. Based on the blind-spot-free global map information obtained in step S1, extract the centroid coordinate observation values ​​of the dynamic obstacle, use the extended Kalman filter algorithm to fuse the centroid coordinate observation values ​​of the dynamic obstacle to perform state estimation of the dynamic obstacle, obtain the optimal state estimate of the dynamic obstacle at the current moment, and generate the predicted trajectory of the dynamic obstacle.

[0011] S3. Based on the load mass obtained in real time in step S1, use the load's dynamic model to solve for the maximum allowable acceleration of the crane under the current working condition; use the obtained maximum allowable acceleration as the dynamic boundary of the velocity sampling space to sample the velocity and obtain the crane's deduced trajectory.

[0012] S4. Perform collision risk detection on each time step of the crane's simulated trajectory and the corresponding time step of the dynamic obstacle prediction trajectory obtained in step S2, generate multiple candidate obstacle avoidance trajectories, select the optimal obstacle avoidance trajectory from the candidate obstacle avoidance trajectories, and output the optimal obstacle avoidance trajectory.

[0013] Furthermore, step S1 includes:

[0014] The blind-spot-free global map information of the working area is collected by LiDARs set at the four outer corners of the two end beams of the crane trolley and the center of the bottom of the crane trolley, and LiDAR point cloud data is generated.

[0015] The load mass is obtained in real time by a pin-type load cell installed at the fixed end of the wire rope on the crane trolley.

[0016] Furthermore, step S2 includes:

[0017] Preprocess the lidar point cloud data to extract the centroid coordinates of dynamic obstacles;

[0018] Establish the motion state vector of the dynamic obstacle, and use the "prediction-update" loop of the extended Kalman filter algorithm to fuse the centroid coordinate observations of the dynamic obstacle to obtain the optimal state estimate of the dynamic obstacle at the current moment.

[0019] Based on the optimal state estimate of the dynamic obstacle at the current moment, a nonlinear motion model is used for prediction iteration to generate the trajectory sequence of the dynamic obstacle for the next N time steps, thereby generating the predicted trajectory of the dynamic obstacle.

[0020] Furthermore, step S3 includes:

[0021] Based on the load mass collected by the pin-type load cell, as well as the mass of the crane trolley and the crane crane crane crane crane paving, and the rated power of the motor, the maximum permissible acceleration of the crane under the current working condition is solved by inversely applying Newton's second law:

[0022]

[0023] In the formula, This represents the maximum permissible acceleration of the crane under the current operating conditions. The load quality under the current operating conditions; This refers to the safety margin factor. Custom power for the motor; For the weight of the crane trolley; For the mass of the crane trolley;

[0024] Based on the improved dynamic window method, velocity sampling spaces for the crane trolley and crane ... ;

[0025] In the formula, For sampling space of crane trolley or crane speed; The current speed of the crane trolley or crane carriage; To predict the time step, This represents the maximum permissible acceleration of the crane under the current operating conditions.

[0026] The crane's trajectory is calculated using the sampling velocity obtained from velocity sampling, and the expression is as follows:

[0027] Where, v x v y For sampling speed, This represents the current position of the crane. Let k be the position at the k-th time step on the crane's simulated trajectory. .

[0028] Furthermore, step S4 includes:

[0029] Extract the corresponding time t from the predicted trajectory obtained in step S2. k Dynamic obstacle prediction state (P(t)) k ),∑(t k ));

[0030] Calculate the safe distance threshold:

[0031] ;

[0032] In the formula, This is the safe distance threshold; The radius of the dynamic obstacle itself; The coefficient of thermal expansion; The trace of the covariance matrix;

[0033] Calculate the distance between the end of the simulated crane trajectory and the end of the predicted trajectory of the dynamic obstacle:

[0034]

[0035] In the formula, The distance between the end of the crane's extrapolated trajectory and the end of the predicted trajectory of the dynamic obstacle at the k-th time step; The position of the end of the crane's trajectory at the k-th time step; Predict the end position of the trajectory of the dynamic obstacle at the k-th time step.

[0036] Furthermore, the optimal obstacle avoidance trajectory output in step S4 includes:

[0037] Each candidate obstacle avoidance trajectory is scored using an evaluation function:

[0038] In the formula, These are the weighting coefficients; For goal-oriented items; For dynamic obstacle distance; For speed;

[0039] Select the candidate obstacle avoidance trajectory with the highest score as the optimal obstacle avoidance trajectory, and output the optimal obstacle avoidance trajectory.

[0040] The aforementioned obstacle avoidance system and method for bridge cranes based on load adaptation and spatiotemporal prediction can achieve the following beneficial effects:

[0041] (1) Load Adaptation: The constantly changing load quality at the physical level is mapped to the dynamic constraints at the algorithm level, ensuring safety and efficiency.

[0042] (2) Predictive obstacle avoidance: By using the extended Kalman filter algorithm to predict the obstacle trajectory sequence at multiple time steps, the problem of the existing algorithm lagging behind in dynamic obstacle avoidance is solved, and the transformation from reactive obstacle avoidance to predictive obstacle avoidance is realized.

[0043] (3) Strong robustness: The obstacle avoidance radius is dynamically adjusted by using the covariance matrix, and the safety envelope is automatically expanded as the prediction time increases, effectively dealing with prediction errors. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0045] Figure 1 This is a flowchart illustrating the obstacle avoidance method for bridge cranes based on load adaptation and spatiotemporal prediction according to the present invention.

[0046] Figure 2 This is a schematic diagram showing the installation location of the lidar in this invention;

[0047] Figure 3 This is a schematic diagram showing the installation position of the pin-type weighing sensor in this invention;

[0048] Figure 4 , Figure 5 The images show simulation results of obstacle avoidance trajectories using the traditional DWA algorithm and the improved DWA algorithm with dynamic obstacle prediction mechanism in this embodiment, respectively.

[0049] Attached reference numerals: 1-Trolley guide rail, 2-Crane trolley, 301-First LiDAR, 302-Second LiDAR, 303-Third LiDAR, 304-Fourth LiDAR, 305-Fifth LiDAR, 4-Pin-type load cell, a1-Load start point, a2-Load end point, A-Load, b1-Dynamic obstacle start point, b2-Dynamic obstacle end point, C-Static obstacle, S-Obstacle avoidance trajectory. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of protection.

[0051] Please see Figures 1-5 This invention discloses an obstacle avoidance system for bridge cranes based on load adaptation and spatiotemporal prediction, comprising:

[0052] The sensing module uses multi-sensor fusion to acquire blind-spot-free global map information and a pin-type load cell to collect load mass in real time.

[0053] Prediction Module: Based on the blind-spot-free global map information obtained by the perception module, the centroid coordinate observation values ​​of dynamic obstacles are extracted. The extended Kalman filter algorithm is used to fuse the centroid coordinate observation values ​​of dynamic obstacles to estimate the state of dynamic obstacles, obtain the optimal state estimate of dynamic obstacles at the current moment, and generate the predicted trajectory of dynamic obstacles.

[0054] Adaptive planning module: Based on the load mass acquired in real time by the sensing module, the maximum allowable acceleration of the crane under the current working condition is solved by using the dynamic model of the load; the obtained maximum allowable acceleration is used as the dynamic boundary of the velocity sampling space for constructing the improved dynamic window method, and velocity sampling is performed to obtain the crane's deduced trajectory.

[0055] Trajectory generation module: Perform collision risk detection on each time step of the crane's simulated trajectory and the corresponding time step of the predicted trajectory of dynamic obstacles, generate multiple candidate obstacle avoidance trajectories, select the optimal obstacle avoidance trajectory from the candidate obstacle avoidance trajectories, and output the optimal obstacle avoidance trajectory.

[0056] This invention also discloses an obstacle avoidance method for bridge cranes based on load adaptation and spatiotemporal prediction, comprising the following steps:

[0057] S1. Multi-sensor fusion is used to obtain blind-spot-free global map information, and a pin-type weighing sensor is used to obtain the load mass in real time.

[0058] For details, please refer to Figure 2 , Figure 3 The system employs a "four-corner coverage + vertical top-down" strategy, installing a lidar 3 at each of the four outer corners of the two end beams of the crane trolley and at the center of the bottom of the crane trolley 2. These lidars are designated as the first lidar 301, the second lidar 302, the third lidar 303, the fourth lidar 304, and the fifth lidar 305. In this embodiment, each lidar is a multi-line 3D lidar. The system collects 360° blind-spot-free global environmental map information of the working area using multiple lidars and generates lidar point cloud data. The load mass is acquired in real time using a pin-type load cell 4 installed at the fixed end of the wire rope on the crane trolley.

[0059] S2. Based on the blind-spot-free global map information obtained in step 1, extract the centroid coordinate observation values ​​of the dynamic obstacles, use the extended Kalman filter algorithm to fuse the centroid coordinate observation values ​​of the dynamic obstacles to perform state estimation of the dynamic obstacles, obtain the optimal state estimate of the dynamic obstacles at the current moment, and generate the predicted trajectory of the dynamic obstacles.

[0060] Specifically, after acquiring global map information using LiDAR, the obstacle avoidance system receives the LiDAR point cloud data and preprocesses it, such as removing the ground and ceiling backgrounds and extracting the centroid coordinates of dynamic obstacles. .

[0061] In this embodiment, the motion state vector of the dynamic obstacle is established. (representing the x-coordinate, y-coordinate, linear velocity, heading angle, and angular velocity, respectively), utilizing the "prediction-update" loop mechanism of the extended Kalman filter algorithm and fusing the centroid coordinate observations of the dynamic obstacle, sensor noise is eliminated to obtain the optimal state estimate of the dynamic obstacle at the current moment. This specifically includes the following two stages:

[0062] (1) Time update: based on the posterior state of the previous time step The prior predicted state at the current moment is derived using a constant rotation rate and velocity kinematic model (CTRV). .

[0063]

[0064] in, To predict the time step.

[0065] Calculate the state transition Jacobian matrix And superimpose the process noise covariance matrix Calculate the prior prediction covariance matrix :

[0066]

[0067] (2) Measurement update: Acquisition The dynamic obstacle center position vector observed by the real-time lidar Establish observation equations Here, h(X) is the observation function. Since the established dynamic obstacle motion state vector X is a five-dimensional vector, while the position (x, y) observed by the lidar is a two-dimensional vector, the function h(X) transforms the five-dimensional vector into a two-dimensional vector. n is the observation noise, representing the error in the measurement process, which is usually modeled as a Gaussian random vector with zero mean and covariance R. The observation Jacobian matrix is ​​then calculated. and observation residuals :

[0068]

[0069]

[0070] Next, the measurement noise covariance matrix is ​​introduced. Calculate Kalman gain The prior state at the current moment is then weighted and corrected to obtain the posterior optimal state estimate. With the posterior covariance matrix :

[0071]

[0072]

[0073]

[0074] Through this step, Kalman gain The weights of model predictions and sensor observations are dynamically adjusted when the measurement noise R is small. By increasing the size of the system, the prediction bias is corrected using observed data, thereby filtering out random noise from the sensors and obtaining the state that is closest to the true value at the current moment, i.e., the posterior optimal state estimate. .

[0075] In this embodiment, based on the optimal state estimate of the dynamic obstacle at the current moment, a nonlinear motion model is used for prediction iteration to generate the trajectory sequence of the dynamic obstacle for the next N time steps, thereby generating the predicted trajectory of the dynamic obstacle.

[0076] Specifically, based on the current optimal state estimation (Similar to the posterior optimal state estimation mentioned earlier) (Equivalent), using a constant rotational speed and velocity kinematics model (CTRV) for prediction iteration, generating a trajectory sequence for the next N time steps.

[0077]

[0078] In the formula, P t+i for Real-time dynamic obstacle location prediction;

[0079] for The position covariance matrix at time t.

[0080] Specifically, the prediction for each time step is obtained in the following way:

[0081] (1) Location prediction P t+i The acquisition of.

[0082] From the discretized form of the CTRV differential equation, the state vector components at the next time step are calculated using the following formula:

[0083]

[0084]

[0085] Through the above nonlinear recursion, the center position coordinates of the dynamic obstacle at each future moment are calculated successively.

[0086] (2) Covariance matrix The acquisition of.

[0087] Using the state transition matrix F (i.e., the Jacobian matrix) The covariance matrix from the previous time step is transferred using the following formula:

[0088]

[0089] Where Q is the process noise covariance matrix.

[0090] S3. Based on the load mass obtained in real time in step S1, use the load's dynamic model to solve for the maximum allowable acceleration under the current working condition; use the obtained maximum allowable acceleration as the dynamic boundary of the velocity sampling space to sample the velocity and obtain the crane's deduced trajectory.

[0091] Specifically, based on the load mass collected by the pin-type load cell, as well as the mass of the crane trolley and crane crane crane crane pawl, and the rated power of the motor, the maximum permissible acceleration under the current working condition is solved by inversely applying Newton's second law:

[0092]

[0093] In the formula, This represents the maximum permissible acceleration of the crane under the current operating conditions. The load quality under the current operating conditions; This refers to the safety margin factor. Custom power for the motor; For the weight of the crane trolley; For the mass of the crane trolley;

[0094] Based on the improved dynamic window method, velocity sampling spaces for the crane trolley and crane ...

[0095] In the formula, For sampling space of crane trolley or crane speed; The current speed of the crane trolley or crane carriage; To predict the time step.

[0096] Specifically, considering the independent decoupled drive characteristics of the crane's main trolley (X-axis) and trolley (Y-axis), speed sampling spaces for the main trolley and trolley are established separately. The speed sampling spaces are not only limited by the rated maximum speed of the crane's main trolley drive motor and trolley drive motor, but also by the maximum allowable acceleration calculated in the aforementioned steps.

[0097] This embodiment obtains the sampling speed through velocity sampling. Calculate the crane's projected trajectory. Specifically, let the crane's current position be... The crane's trajectory is calculated at the k-th time step (corresponding to time). The position of ) is The expression is:

[0098] S4. Perform collision risk detection on each time step of the crane's simulated trajectory and the corresponding time step of the dynamic obstacle prediction trajectory obtained in step S2, generate multiple candidate obstacle avoidance trajectories, select the optimal obstacle avoidance trajectory from the candidate obstacle avoidance trajectories, and output the optimal obstacle avoidance trajectory.

[0099] Specifically, the corresponding time is extracted from the predicted trajectory obtained in step S2. Dynamic obstacle prediction state (P(t)) k ),∑(t k )); Calculate the safe distance threshold:

[0100] ;

[0101] In the formula, This is the safe distance threshold; The radius of the dynamic obstacle itself; The coefficient of thermal expansion; The trace of the covariance matrix;

[0102] Calculate the distance between the end of the crane's simulated trajectory and the end of the dynamic obstacle's trajectory:

[0103]

[0104] In the formula, The distance between the end of the crane's extrapolated trajectory and the end of the predicted trajectory of the dynamic obstacle at the k-th time step; The position of the end of the crane's trajectory at the k-th time step; Predict the end position of the trajectory of the dynamic obstacle at the k-th time step.

[0105] when If the crane's predicted trajectory does not collide with the obstacle's predicted trajectory, then the crane's predicted trajectory will be discarded.

[0106] Furthermore, the optimal obstacle avoidance trajectory output in step S4 includes:

[0107] Each candidate obstacle avoidance trajectory is scored using an evaluation function, which is:

[0108] In the formula, These are the weighting coefficients; For goal-oriented items; For dynamic obstacle distance; This is the speed term.

[0109] Select the candidate obstacle avoidance trajectory with the highest score as the optimal obstacle avoidance trajectory, and output the optimal obstacle avoidance trajectory.

[0110] Specifically, all collision-free trajectories are traversed, and the "target approach," "velocity," and "spatiotemporal distance to dynamic obstacles" are comprehensively evaluated based on the improved DWA algorithm evaluation function. The candidate obstacle avoidance trajectory with the highest score is the optimal obstacle avoidance trajectory, which is then output to the motor for execution. Figure 4 The image shows a simulation diagram of the obstacle avoidance trajectory using the traditional DWA algorithm, which does not incorporate a dynamic obstacle prediction mechanism. Figure 5 The diagram shows the simulation effect of the obstacle avoidance trajectory using the method described in this embodiment, namely the improved DWA algorithm incorporating a dynamic obstacle prediction mechanism. A comparison of the two diagrams shows that the obstacle avoidance trajectory obtained by the method described in this embodiment is smoother and the obstacle avoidance effect is better.

[0111] The embodiments of the present invention have the following beneficial effects:

[0112] (1) Load Adaptation: The constantly changing load quality at the physical level is mapped to the dynamic constraints at the algorithm level, ensuring safety and efficiency.

[0113] (2) Predictive obstacle avoidance: By using EKF multi-time step obstacle prediction trajectory sequence, the problem of existing algorithms lagging behind in dynamic obstacle avoidance is solved, realizing the transformation from reactive obstacle avoidance to predictive obstacle avoidance.

[0114] (3) Strong robustness: The obstacle avoidance radius is dynamically adjusted by using the covariance matrix, and the safety envelope is automatically expanded as the prediction time increases, effectively dealing with prediction errors.

[0115] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort, such as modifications to the technical solutions described in the following embodiments or equivalent substitutions of some technical features, are within the scope of protection of the present invention.

Claims

1. A method for obstacle avoidance of bridge cranes based on load adaptation and spatiotemporal prediction, characterized in that, Includes the following steps: S1. Multi-sensor fusion is used to obtain blind-spot-free global map information, and a pin-type load cell is used to obtain the load mass in real time. S2. Based on the blind-spot-free global map information obtained in step S1, extract the centroid coordinate observation values ​​of the dynamic obstacle, use the extended Kalman filter algorithm to fuse the centroid coordinate observation values ​​of the dynamic obstacle to perform state estimation of the dynamic obstacle, obtain the optimal state estimate of the dynamic obstacle at the current moment, and generate the predicted trajectory of the dynamic obstacle. S3. Based on the load mass obtained in real time in step S1, the maximum allowable acceleration of the crane under the current working condition is solved by using the dynamic model of the load; the obtained maximum allowable acceleration is used as the dynamic boundary of the velocity sampling space for velocity sampling, and then the crane trajectory is obtained. S4. Perform collision risk detection on each time step of the crane's simulated trajectory and the corresponding time step of the dynamic obstacle prediction trajectory obtained in step S2, generate multiple candidate obstacle avoidance trajectories, select the optimal obstacle avoidance trajectory from the candidate obstacle avoidance trajectories, and output the optimal obstacle avoidance trajectory. Step S3 includes: Based on the load mass collected by the pin-type load cell, as well as the mass of the crane trolley and the crane crane crane crane crane paving, and the rated power of the motor, the maximum permissible acceleration of the crane under the current working condition is solved by inversely applying Newton's second law: ； In the formula, This represents the maximum permissible acceleration of the crane under the current operating conditions. The load quality under the current operating conditions; This refers to the safety margin factor. Custom power for the motor; For the weight of the crane trolley; For the mass of the crane trolley; Based on the improved dynamic window method, velocity sampling spaces for the crane trolley and crane ... ; In the formula, For sampling space of crane trolley or crane speed; The current speed of the crane trolley or crane carriage; To predict the time step, This represents the maximum permissible acceleration of the crane under the current operating conditions. The crane's trajectory is calculated using the sampling velocity obtained from velocity sampling, and the expression is as follows: ; Among them, v x v y For sampling speed, This represents the current position of the crane. Let this be the position of the crane at the k-th time step in the trajectory simulation. .

2. The obstacle avoidance method for bridge cranes based on load adaptation and spatiotemporal prediction according to claim 1, characterized in that, Step S1 includes: The blind-spot-free global map information of the working area is collected by LiDARs set at the four outer corners of the two end beams of the crane trolley and the center of the bottom of the crane trolley, and LiDAR point cloud data is generated. The load mass is obtained in real time by a pin-type load cell installed at the fixed end of the wire rope on the crane trolley.

3. The obstacle avoidance method for bridge cranes based on load adaptation and spatiotemporal prediction according to claim 2, characterized in that, Step S2 includes: Preprocess the lidar point cloud data to extract the centroid coordinates of dynamic obstacles; Establish the motion state vector of the dynamic obstacle, and use the "prediction-update" loop of the extended Kalman filter algorithm to fuse the centroid coordinate observations of the dynamic obstacle to obtain the optimal state estimate of the dynamic obstacle at the current moment. Based on the optimal state estimate of the dynamic obstacle at the current moment, a nonlinear motion model is used for prediction iteration to generate the trajectory sequence of the dynamic obstacle for the next N time steps, thereby generating the predicted trajectory of the dynamic obstacle.

4. The obstacle avoidance method for bridge cranes based on load adaptation and spatiotemporal prediction according to claim 1, characterized in that, Step S4 includes: Extract the corresponding time t from the predicted trajectory obtained in step S2. k Dynamic obstacle prediction state (P(t)) k ),∑(t k )); Calculate the safe distance threshold: ； In the formula, This is the safe distance threshold; The radius of the dynamic obstacle itself; The coefficient of thermal expansion; The trace of the covariance matrix; Calculate the distance between the end of the simulated crane trajectory and the end of the predicted trajectory of the dynamic obstacle: ； In the formula, The distance between the end of the crane's extrapolated trajectory and the end of the predicted trajectory of the dynamic obstacle at the k-th time step; The position of the end of the crane's trajectory at the k-th time step; Predict the end position of the trajectory of the dynamic obstacle at the k-th time step.

5. The obstacle avoidance method for bridge cranes based on load adaptation and spatiotemporal prediction according to claim 1, characterized in that, The optimal obstacle avoidance trajectory output in step S4 includes: Each candidate obstacle avoidance trajectory is scored using an evaluation function: ; In the formula, These are the weighting coefficients; For goal-oriented items; For dynamic obstacle distance; For the speed term, For sampling rate; Select the candidate obstacle avoidance trajectory with the highest score as the optimal obstacle avoidance trajectory, and output the optimal obstacle avoidance trajectory.

6. A bridge crane obstacle avoidance system based on load adaptation and spatiotemporal prediction, characterized in that, The obstacle avoidance system is used to implement the obstacle avoidance method according to any one of claims 1-5, including: Perception module: It uses multi-sensor fusion to acquire blind-spot-free global map information and uses a pin-type load cell to acquire the load mass in real time; Prediction module: Based on the blind-spot-free global map information obtained by the perception module, extract the centroid coordinate observation values ​​of dynamic obstacles, use the extended Kalman filter algorithm to fuse the centroid coordinate observation values ​​of dynamic obstacles to perform state estimation of dynamic obstacles, obtain the optimal state estimate of dynamic obstacles at the current moment, and generate the predicted trajectory of dynamic obstacles. Adaptive planning module: Based on the load mass acquired in real time by the sensing module, the maximum allowable acceleration of the crane under the current working condition is solved by using the dynamic model of the load; an improved dynamic window method is constructed, and the obtained maximum allowable acceleration is used as the dynamic boundary of the velocity sampling space for velocity sampling, thereby obtaining the crane's deduced trajectory; Trajectory generation module: Perform collision risk detection on each time step of the crane's simulated trajectory and the corresponding time step of the predicted trajectory of dynamic obstacles, generate multiple candidate obstacle avoidance trajectories, select the optimal obstacle avoidance trajectory from the candidate obstacle avoidance trajectories, and output the optimal obstacle avoidance trajectory.

Images