Self-propelled winding machine control method based on visual identification and autonomous path planning

Abstract

The invention relates to the technical field of machine vision, in particular to a self-propelled winding machine control method based on visual identification and autonomous path planning. The space around goods is divided into a plurality of sector units according to height and azimuth angles by taking a tray as a reference, a stability risk value is calculated for each sector unit, a target coverage degree is set for each sector unit in combination with goods categories and working conditions, a stability and coverage degree combined risk map is written in uniformly, and the risk value is calculated. Compared with the method that the number of turns is roughly configured only according to the overall height or the single direction in the prior art, the method has the advantages that the risk and reinforcement requirements of each local area can be finely described on the two-dimensional scale of the height matching orientation, and particularly, smaller grid division step length is adopted on the bottom reinforcement layer; the risk resolution of key areas such as the fork inserting direction and the unbalance loading direction of the forklift is improved, and a quantitative basis is provided for subsequent differential winding parameter planning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine vision, in particular to a self-walking wrapping machine control method based on visual recognition and autonomous path planning. BACKGROUND

[0002] Pallet cargo wrapping is a common anti-shedding, anti-slip and moisture-proof reinforcement means in the warehouse and logistics link, and various fixed or self-walking wrapping machine control schemes have been proposed in the prior art. The fixed wrapping machine usually realizes the wrapping of the pallet cargo by rotating the rotating disc or lifting the film frame, and executes simple program control according to the preset number of turns, lifting speed and tension parameters of the operator; some self-walking wrapping machines try to introduce visual sensors or depth sensors to identify the shape of the cargo, which is used to automatically determine the height and approximate volume of the cargo, so as to automatically set the wrapping height range and basic parameters. In addition, with the development of automatic guided vehicles and mobile robots, self-walking wrapping equipment combining mobile chassis and wrapping mechanism has also appeared, which can autonomously travel between storage locations and perform wrapping operations near the pallet position.

[0003] However, the above prior art in the self-walking wrapping scene is mostly based on the overall height, weight or simplified shape profile of the cargo to set uniform wrapping process parameters, which cannot finely characterize the stability differences of different circumferential and height positions in the pallet coordinate system, leading to difficulties in timely identifying and implementing targeted reinforcement for the forklift insertion side, the off-load side, the upper stacking irregularity and other local high-risk directions. In addition, the key process parameters such as the number of wrapping layers, the wrapping overlap rate and the film tension level are usually kept basically constant during the entire wrapping process, which is difficult to balance the film consumption and overall stability optimization while meeting the transportation safety requirements. Finally, there is generally a lack of real-time monitoring and feedback adjustment mechanism for the attitude changes of the cargo such as tilting and shaking during the wrapping process. When the wrapping process is disturbed by external disturbances or the center of gravity of the cargo shifts, it cannot achieve closed-loop compensation by adjusting the motion trajectory and wrapping parameters, which easily leads to risks such as insufficient local restraint, decreased overall stability and even cargo scattering.

[0004] Therefore, a self-walking wrapping machine control method based on visual recognition and autonomous path planning is proposed. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the present application provides a self-walking wrapping machine control method based on visual recognition and autonomous path planning.

[0006] To solve the above technical problems, the present application provides the following technical solutions: including the following steps: S1, obtaining three-dimensional perception data of the target tray and the goods by the visual sensor or the depth sensor of the self-walking winding machine, determining the boundary of the tray and the outer contour of the goods, dividing the space around the goods into multiple sector units according to height and azimuth sector, calculating the stability risk value of each sector unit, and setting the target coverage of each sector unit, thereby constructing a risk map; S2, based on the camera image sequence obtained by the visual sensor, performing a simultaneous localization and mapping algorithm, when planning a navigation path and a close direction, based on the risk map as a reference, preferentially selecting the side close to the high-risk direction as the starting side of the detour, and driving the self-walking winding machine to move along the navigation path to the vicinity of the target tray, determining the starting pose of the detour and the first winding direction; S3, based on the risk map and the outer contour of the goods, the control unit divides different height layers based on the height section of the winding area in the height direction, and configures the basic coverage constant respectively, and determines the target winding layer number, the target overlap rate and the target tension level of the sector unit according to the corresponding stability risk value and the target coverage, and constructs the control parameter trajectory on this basis; S4, while performing the winding operation according to the control parameter trajectory, periodically detecting the state of the goods, updating the actual coverage of each sector unit in the risk map, when detecting that the displacement or inclination of the goods to a certain direction exceeds the preset threshold, increasing the stability risk value of the sector unit in the corresponding direction in the risk map and increasing its target coverage, limiting the maximum walking speed and acceleration of the self-walking winding machine, and performing directional reinforcement winding on the direction and adjacent sector units in the subsequent turns; S5, after reaching the preset winding turn number and height coverage condition, controlling the self-walking winding machine to walk around the goods periphery in detection mode, identifying the quality defects of the film in each sector unit through the visual sensor, writing the defect result into the quality state in the risk map, and obtaining the target sector unit set that needs to be supplemented based on the joint determination of the stability risk and the quality risk of each sector unit, generating a local supplementary winding path according to the current machine position and the position of the target sector unit, and performing supplementary winding; S6, after the winding and supplementary winding are completed, the final risk map, the control parameter trajectory used in the winding process, the goods category and the corresponding subsequent stacking and transportation working conditions are associated and stored, and based on the preset experience updating rule, the initial parameters and winding strategies of the risk map of the subsequent similar goods and similar working conditions are updated.

[0007] As a preferred technical solution of the present application, the sector unit division comprises: first, taking the geometric center of the pallet as the origin to establish a space rectangular coordinate system, letting the X-axis and Y-axis be parallel to the two mutually perpendicular edges of the pallet, and the Z-axis be vertically upward, thereby obtaining the pallet coordinate system; determining the pallet surface height and the maximum height of the goods according to the three-dimensional perception data, and determining the circumscribed circle radius of the pallet outer edge in the XY plane; in the height direction, the interval from the pallet surface to the maximum height of the goods is divided into several height sections; in the horizontal direction, taking the X-axis of the pallet coordinate system as the zero angle direction, the plane angle range around the pallet from 0 degrees to 360 degrees is equally divided by a preset angle step, for example, every several degrees is divided into a bearing section, thereby obtaining a plurality of sequentially adjacent bearing sectors, and the bearing angle range corresponding to the first bearing sector is set as the first interval starting from 0 degrees, and the subsequent bearing sectors are sequentially arranged in the same angle width clockwise or counterclockwise until the entire plane angle range of 0 to 360 degrees is covered; the space region around the pallet from the pallet surface to the maximum height of the goods is divided into a series of sector units divided by height sections and bearing sectors, each sector unit is determined by its belonging height section number and bearing sector number; the stability risk value is calculated by linear weighting, satisfying: ; wherein, is the stability risk value representing the local stability risk, is the height section where the sector unit is located, is the bearing angle of the sector unit, is the offset of the centroid of the goods in the height section relative to the geometric center of the pallet in the direction, is the offset of the centroid of the goods in the height section relative to the geometric center of the pallet in the direction, is the difference between the height of the centroid of the goods in the height section relative to the pallet surface and the reference height, , , is the weight coefficient pre-configured according to the goods category; for each sector unit, a corresponding stability risk threshold is also pre-set, which represents the maximum stability risk level allowed by the sector unit under the current goods category and working condition, and the stability risk threshold is written into the risk map; the target coverage degree is calculated according to the stability risk value and the working condition risk weight, satisfying: ; wherein, is the target coverage degree of each sector unit, is the basic coverage constant, , is the target coverage adjustment coefficient, used to adjust the relative influence weight of stability risk and working condition risk in the calculation of target coverage, The calculated work condition risk weight is determined in advance according to the height of the subsequent stacking, and according to the transportation distance and transportation vibration level belonging to the transportation work condition The target coverage degree in the writing risk map as a sector unit.

[0008] As a preferred technical solution of the present application, the determination of the target winding layer number: by reading the stability risk value and the target coverage degree of the sector unit in the risk map, and combining the pre-set single-layer film coverage contribution, the target coverage degree is converted into the target winding layer number; the target overlap rate is allocated to each sector unit according to the height section, so that the basic overlap rate of the bottom reinforcing layer is higher than that of the middle bearing layer and the top locking layer, and on this basis, the overlap rate is adjusted up or down according to the stability risk value of each sector unit, so that the actual overlap rate in the high-risk direction is higher than that in the low-risk direction; the target tension level is selected from the pre-set multiple tension levels according to the stability risk value and the fragile level of the goods, when the stability risk is high and the goods are not fragile, a high tension level is preferred to be selected to enhance the binding effect, and when the goods are fragile, the tension level is limited to not more than the medium level, and the stability decline caused by the limited tension is compensated by increasing the target winding layer number and the overlap rate, so as to realize the cooperative configuration among the target winding layer number, the target overlap rate and the target tension level; the control parameter trajectory is obtained by extracting the outer contour polygon of the pallet and the goods in the horizontal projection plane, taking the geometric center of the pallet as the coordinate origin, arranging the discrete points on the outer contour in the order of azimuth angle, and according to the stability risk value and the target coverage degree of the corresponding azimuth sector in the risk map, calculating the safety distance matched with each azimuth angle, so that the safety distance in the high-risk direction and the target coverage high direction is greater than that in the low-risk direction and the target coverage low direction; each discrete point on the outer contour is moved outward along the radial direction by a corresponding safety distance and a winding machine shape envelope allowance to obtain a set of winding envelope points, the winding envelope point set is smoothly connected and curve fitted to form a closed winding envelope path, and the target winding layer number, the target overlap rate and the target tension level of each path point are hung according to the corresponding height section and azimuth sector, and the walking speed, lifting speed and film release speed control parameter trajectories varying with the winding angle and the height are generated through time parameterization, so that the self-propelled winding machine can travel along the safety-inflated envelope path during the actual winding process, and execute differentiated winding strategies according to the risk level of different height sections and different azimuth sectors.

[0009] As a preferred technical solution of the present application, the inclination angle of the goods relative to the pallet at each height is estimated by using a visual sensor or a depth sensor during the winding process , and the stability risk value of the corresponding azimuth sector unit is dynamically updated according to the inclination angle , and the updated stability risk value ; wherein is the stability risk value before updating, is the tilt risk gain coefficient, is the tilt angle of the goods at the height with respect to the vertical direction is the indication function of the azimuth sector corresponding to the tilt direction, when the sector azimuth and the tilt direction have an angle less than a preset angle threshold is 1, otherwise 0; during the winding operation, the self-propelled winding machine continuously calculates the tilt angle of the goods at each height with respect to the vertical direction based on the sensing results of the visual sensor or the depth sensor , and combines the angle relationship between the tilt direction and each azimuth sector to online correct the stability risk distribution of the corresponding height section in the risk map, so that the risk map can reflect the evolution characteristics of the goods posture disturbance in the height and azimuth two-dimensional space during the winding process, and provide quantitative basis and decision input for subsequent dynamic improvement of target coverage in corresponding high-risk azimuth, reduction of walking speed and acceleration upper limit, and implementation of directional reinforcement winding and other winding strategies.

[0010] As a preferred technical solution of the present application, the comprehensive quality risk value of each sector unit quality defect is calculated as follows: ; wherein, is the proportion of the area of the sector unit not covered by the film to the visible area of the sector, is the degree of the film coverage layer in the sector being lower than the target coverage , is the normalized severity of quality defects such as slack, wrinkles and film breakage, , , is the quality risk weight coefficient; the target sector unit that needs to be supplemented is determined by combining the stability risk value and the quality risk value, and when the weighted sum of the stability risk value and the quality risk value of a certain sector unit is greater than a safety threshold, the sector unit is added to the supplementary winding set, and the coverage increase of each sector unit in the supplementary winding set is ensured to be not less than a preset coverage compensation when generating a local supplementary winding path.

[0011] As a preferred technical solution of the present application, the experience updating rule of the initial parameters of the risk map adopts an exponential weighting method, and a certain risk parameter is updated according to the following formula: ; wherein, is the historical value of the parameter under similar goods and similar working conditions in the last round, is the observation value estimated according to the actual tilt trajectory and the damage condition in this task, The experience update step coefficient is between 0 and 1, and the experience update is only performed when the similarity of the goods category and the working condition feature is not lower than a preset threshold value; the similar goods and the similar working condition are determined by the comprehensive similarity of the goods category feature vector and the working condition feature vector, and when the similarity between two tasks is not lower than a preset task classification threshold value, it is considered that the two tasks belong to the same task, so that the experience update of the risk parameter is only performed between the samples classified into the same task; the control method is suitable for various different specifications of pallets and different sizes of goods, and when it is detected that the pallet size or the goods contour exceeds the preset range, the self-propelled winding machine automatically adjusts the height range and the azimuth resolution of the risk map, and replans the navigation path and the winding strategy, so that the winding control based on the risk map can still be realized under different pallet specifications and goods sizes.

[0012] Compared with the prior art, the application has the following beneficial effects: 1、The application obtains the fusion point cloud data of the pallet and the goods through the visual sensor and the depth sensor, divides the space around the goods into a plurality of sector units according to the height and the azimuth based on the pallet, calculates the stability risk value for each sector unit, sets the target coverage for each sector unit in combination with the goods category and the working condition, and writes the stability and coverage joint risk map uniformly, so that the risk and reinforcement demand of each local area can be finely described in the two-dimensional scale of the height and the azimuth, especially by using a smaller grid division step in the bottom reinforcement layer, the risk resolution of the key areas such as the forklift insertion direction and the unbalanced load direction is improved, and a quantitative basis is provided for subsequent differentiated winding parameter planning.

[0013] 2、The application introduces a simultaneous localization and mapping algorithm based on image sequences and depth information on the self-propelled winding machine, constructs a warehouse environment map and plans a navigation path to the target pallet, automatically completes the alignment of the machine body and the pallet in combination with the pallet outer contour recognition result when approaching the pallet, generates a winding envelope path around the pallet outer contour, and connects the walking speed, the film frame lifting speed and the film release speed control trajectories varying with the height and the azimuth on the path, compared with the existing self-propelled winding equipment relying only on simple obstacle avoidance and manual alignment, the dependence on manual operation experience is significantly reduced, the pallet alignment accuracy and the rationality of the winding trajectory are improved, and a basic guarantee is provided for stable automatic winding operation.

[0014] 3、The application utilizes visual and depth perception to monitor the posture of goods at each height section and the film covering state in real time during the winding process, updates the risk map of each sector unit online, when detecting that the inclination of a certain height section towards a specific direction exceeds the threshold value, only increases the target coverage of the direction and adjacent sectors and reduces the running speed of the corresponding height section, and executes directional reinforcement winding; after completing the preset number of turns, defects such as uncovered, over-thin, loose, wrinkled and broken film are detected, a quality risk evaluation is constructed, the high-risk sectors that need to be supplemented are determined in combination with the stability risk, and a local supplementary winding path with the supplementary winding distance and time as the optimization target is generated, thereby, directional reinforcement and supplementary winding can be implemented for local high-risk areas under the premise of meeting the stability risk threshold and safety threshold, and large-scale redundant winding is avoided, and packaging safety and film utilization rate are considered.

[0015] 4、The application sets a stability risk threshold for each sector unit, and constructs an overall safety determination condition based on the risk map, when the risk map cannot converge below the safety threshold within the predetermined winding time and the number of supplementary windings, the automatic mode is automatically switched to the manual auxiliary mode, the automatic winding is suspended and the operator is prompted to check the goods stack or winding configuration, the risk of continuous operation leading to instability is reduced, and when the risk converges below the safety threshold, the final risk map and winding control parameters of this task are used to correct the initial parameters and control parameters of the risk map under similar goods and similar working conditions according to the experience updating rule, through the above mechanism, not only the safety switching protection between automatic and manual is provided, but also the winding strategy can be continuously optimized in repeated tasks, and the self-adaptability and long-term running robustness for different goods categories and working conditions are improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Fig. 1 The figure is a schematic diagram of the method of the application; Fig. 2 The figure is a schematic diagram of the sector unit division of the application. DETAILED DESCRIPTION

[0017] In order to make the technical means, creative features, purposes and effects realized by the application easy to understand, the application will be further described below in combination with specific embodiments, but the following embodiments are only preferred embodiments of the application, not all. Based on the embodiments in the embodiments, other embodiments obtained by those skilled in the art without creative labor also belong to the protection scope of the application.

[0018] Embodiment: as Figs. 1-2As shown, the self-walking winding machine control method based on visual recognition and autonomous path planning comprises a self-walking winding machine, the self-walking winding machine comprising a machine body, a walking driving mechanism, a winding execution mechanism, a visual sensor, a depth sensor and a control unit, the control unit being electrically connected with the walking driving mechanism, the winding execution mechanism, the visual sensor and the depth sensor, and being used for executing the control method. It should be noted that the machine body, the walking driving mechanism and other structural features are prior known technologies, and will not be described here.

[0019] The control method comprises the following steps: S1, obtaining three-dimensional perception data of a target pallet and goods by a visual sensor or a depth sensor of the self-walking winding machine, determining the boundary of the pallet and the outer contour of the goods, the visual sensor comprising an RGB-D camera installed at the front end of the machine body, and the depth sensor comprising a laser radar sensor, when obtaining the three-dimensional perception data, the self-walking winding machine adopts fused point cloud data of the RGB-D camera and the laser radar sensor to construct the three-dimensional perception data of the target pallet and goods, divides the space around the goods into multiple sector units according to height and azimuth sector, calculates a stability risk value of each sector unit based on the goods centroid, local offset and goods category, and sets a target coverage of each sector unit in combination with subsequent stacking and transportation conditions of the pallet, thereby constructing a risk map based on the sector unit, which jointly represents stability and coverage, the risk map representing a collection data structure recording corresponding stability risk value, target coverage, actual coverage and other information in each sector unit, through the risk map, the local stability risk of each height section and each azimuth sector and the film coverage after winding can be directly described, and a unified risk assessment basis is provided for subsequent winding strategy generation, the risk map can be reused between multiple pallets within the same batch of goods, when detecting that the pallet stacking form and the goods type are the same as or similar to the previous pallet, the self-walking winding machine preferentially calls the risk map initial parameters and the winding strategy of the last task, thereby reducing the time of re-modeling and parameter searching.

[0020] S2, the self-propelled winding machine performs a simultaneous localization and mapping algorithm based on a camera image sequence acquired by a visual sensor, the simultaneous localization and mapping algorithm comprising: a visual odometry submodule and a graph optimization submodule; wherein the visual odometry submodule performs feature extraction and matching on the camera image sequence continuously collected by the visual sensor during the movement of the self-propelled winding machine, estimates the relative pose between adjacent image frames to obtain the incremental pose of the robot; the graph optimization submodule identifies the repeated loop scene through loop detection on the camera image sequence, adds the corresponding loop constraint to the pose graph, and performs global optimization on the robot pose and the environment map based on the loop constraint and the incremental pose, thereby reducing the cumulative positioning error and improving the accuracy of the environment map for navigation path planning; when planning the navigation path and the approaching direction of the self-propelled winding machine from the current parking position to the target pallet, under the premise of meeting the environmental obstacle avoidance and passage constraints, the stability risk value of the sector unit corresponding to the approaching direction in the risk map is taken as a reference, the side close to the high-risk direction is preferentially selected as the turning start side in order to perform directional reinforcement winding, and the self-propelled winding machine is driven to move along the navigation path to the vicinity of the target pallet, aligns its own pose by recognizing the pallet corner points and the cargo edges, determines the turning start pose and the first circle winding direction; specifically, the self-propelled winding machine takes the two adjacent corner points of the pallet as a reference when determining the turning start pose, detects the relative position deviation and heading deviation between the pallet corner points and the geometric center of the walking driving mechanism, and gradually reduces the position deviation and heading deviation in an iterative adjustment manner until it is below a preset alignment error threshold.

[0021] S3, based on the risk map and the cargo outer contour, the control unit divides the winding area into height segments in the height direction, merges and defines several height segments close to the pallet support surface as a bottom reinforcement layer, merges and defines several height segments in the middle height range as a middle bearing layer, and merges and defines several height segments close to the top of the cargo as a top locking layer, and configures a basic coverage constant for different height layers, so that the target coverage of the bottom reinforcement layer is higher than that of the middle bearing layer, and the target coverage of the middle bearing layer is higher than that of the top locking layer, and in each height segment, the sector unit is divided according to the azimuth angle, and the target winding layer number, the target overlap rate and the target tension level are determined for each sector unit according to the corresponding stability risk value and the target coverage; on this basis, the pallet outer contour is inflated with a safety distance to generate a turning envelope path, and the control parameter trajectory of the walking speed, the lifting speed and the film release speed varying with the turning angle and the height is constructed.

[0022] S4. While driving the self-walking wrapping machine to perform the wrapping operation around the goods according to the control parameter trajectory, periodically detect the displacement of the goods relative to the pallet, the tilting posture and the film covering state based on the visual sensor or the depth sensor, update the actual coverage of each sector unit in the risk map, when detecting that the displacement or tilting of the goods to a certain direction exceeds the preset threshold, increase the stability risk value of the sector unit in the corresponding direction in the risk map and increase the target coverage, limit the maximum walking speed and acceleration of the self-walking wrapping machine, and perform directional reinforcement wrapping on the direction and adjacent sector units in the subsequent turns.

[0023] S5. After reaching the preset wrapping turns and high coverage conditions, control the self-walking wrapping machine to walk around the periphery of the goods in a detection mode, identify the quality defects such as coverage missing, coverage being too thin, relaxation, wrinkles and film breakage of the film of each sector unit through the visual sensor, write the defect results into the quality state in the risk map, obtain a target sector unit set that needs to be supplemented based on the joint determination of the stability risk and quality risk of each sector unit, generate a local supplementary wrapping path according to the current position and the position of the target sector unit, and drive the self-walking wrapping machine to supplement the target sector unit along the local supplementary wrapping path.

[0024] S6. After the wrapping and supplementary wrapping are completed, store the final risk map, the control parameter trajectory used in the wrapping process, the goods category and the corresponding subsequent stacking and transportation working conditions in association, and based on the preset experience updating rule, update the initial parameters and wrapping strategies of the risk map for subsequent similar goods and similar working conditions, the self-walking wrapping machine is provided with a wireless communication module to interact with a remote monitoring platform, the remote monitoring platform can issue working condition parameters, wrapping strategy templates and risk threshold configurations, and the self-walking wrapping machine adjusts the parameters of the risk map according to the latest configurations during operation.

[0025] It should be noted that S1 includes the following steps: S1.1, the construction of the fusion point cloud data includes: the self-walking wrapping machine respectively collects the point cloud data of the target pallet and the goods by using the laser radar sensor and the RGB-D camera, converts the point cloud data to the wrapping machine body coordinate system through the pre-established external parameter calibration relationship and time synchronization mechanism, and performs filtering, de-duplication and merging processing on the converted point cloud data to obtain the fusion point cloud data for constructing the three-dimensional perception data, so as to improve the perception accuracy and robustness of the boundaries of the pallet and the outer contour of the goods.

[0026] S1.2, the division of the sector unit includes: first establishing a space rectangular coordinate system with the geometric center of the pallet as the origin, letting the X-axis and the Y-axis be parallel to the two mutually perpendicular edges of the pallet, and the Z-axis be perpendicular to the upper direction, thereby obtaining the pallet coordinate system, determining the pallet surface height and the maximum height of the goods according to the three-dimensional perception data And determine the radius of the circumcircle of the outer edge of the tray in the XY plane. In the height direction, the range from the pallet surface to the maximum height of the goods [0, Divide into several height segments, for example, into There are several height segments, with each segment spaced apart by [number]. In the horizontal direction, with the X-axis of the pallet coordinate system as the zero angle direction, the planar angle range around the pallet from 0 degrees to 360 degrees is divided at equal intervals according to a preset angle step size. For example, an azimuth interval is divided every certain number of degrees, thus obtaining several sequentially adjacent azimuth sectors. The azimuth angle range corresponding to the first azimuth sector is set as the first interval starting from 0 degrees. Subsequent azimuth sectors are arranged sequentially clockwise or counterclockwise with the same angle width until the entire planar angle range from 0 to 360 degrees is covered. In this way, the spatial area around the pallet from the pallet surface to the maximum height of the goods is divided into a series of sector units divided by height segment and azimuth sector. The sector unit is determined by its height segment number and azimuth sector number. In practice, the self-propelled winding machine traverses and merges the point cloud data point by point. For each point in the point cloud data, the height value of the point in the pallet coordinate system is read first, and it is determined which height segment it falls into. Then, based on the angle between the projection of the point on the horizontal plane and the X-axis of the pallet coordinate system, the azimuth sector to which the point belongs is determined, and the point is assigned to the corresponding sector unit. After completing the traversal of the merged point cloud data, the local centroid position, local outer edge contour, local coverage, and local quality defect characteristics of the point cloud data are statistically analyzed in each sector unit, and the above statistical results are used as the basic data for constructing the risk map.

[0027] S1.3. The stability risk value of each sector unit is calculated using a linear weighted method, satisfying: .

[0028] in, The stability risk value, used to characterize local stability risk, reflects the degree of local instability in a sector cell caused by cargo center offset and height superelevation. The height range where the sector unit is located. The azimuth angle of the sector unit represents the horizontal angle relative to the zero-angle direction of the tray. For the height segment, the center of gravity of the cargo relative to the geometric center of the pallet is at The offset in direction. For the height segment, the center of gravity of the cargo relative to the geometric center of the pallet is at The offset in direction. This is the difference between the height of the cargo's center of gravity relative to the pallet surface and the reference height within the height range. , , The weight coefficient is pre-configured according to the cargo category, used to adjust the relative importance of the horizontal deviation factor and the height factor in the stability risk value calculation, and the stability risk threshold corresponding to each sector unit is also pre-set, used to represent the maximum stability risk level allowed for the sector unit under the current cargo category and working condition, and the stability risk threshold is written into the risk map, and in other embodiments, the weight coefficient can also be adjusted in combination with the pallet specification, subsequent stacking and transportation working conditions; through the above calculation method, when the local cargo centroid in a certain height section and a certain azimuth sector deviates from the pallet geometric center in the horizontal plane is larger and the overall height of the cargo is higher, the corresponding stability risk value is larger, thereby directly reflecting that the sector unit is more likely to overturn or displace, so that the winding strategy can focus on reinforcing the high-risk areas with high stability risk values; when calculating, the stability risk value is also corrected according to the fragile level of the cargo in the cargo category, the stability risk value of the sector unit corresponding to fragile cargo is increased, and the stability risk value of the sector unit corresponding to impact-resistant cargo is reduced, so as to reflect the difference in tolerance ability of different cargos.

[0029] S1.4, the target coverage of each sector unit is calculated according to the stability risk value and the working condition risk weight, and satisfies: .

[0030] wherein, is the target coverage of each sector unit, is a basic coverage constant, , is a target coverage adjustment coefficient, used to adjust the relative influence weight of the stability risk and the working condition risk in the target coverage calculation, is a working condition risk weight pre-determined according to the height of the subsequent stacking, and according to the transportation distance and transportation vibration level belonging to the transportation working condition, and the calculated is written into the risk map as the target coverage of the sector unit; through the above calculation method, when the stability risk value of a certain sector unit is higher, the corresponding working condition risk weight is larger, the target coverage of the sector unit is correspondingly increased, so as to allocate more winding layers or higher overlap rate in the subsequent winding strategy, so as to enhance the local stability; otherwise, the target coverage is kept at a level close to the basic coverage, so as to avoid excessive winding and film waste; by introducing the target coverage calculation mechanism based on the stability risk and the working condition risk, the winding machine can implement differentiated winding strategies for different heights and different azimuths of the cargo area on the premise of meeting the overall stability requirement, and realize the comprehensive optimization between the film consumption and the packaging safety.

[0031] It should be further explained that S2 includes the following steps: S2.1, obstacle avoidance and passing constraint includes: the environment map includes static obstacle layer and dynamic obstacle layer, the self-moving winding machine periodically updates the dynamic obstacle layer during navigation, and the detected dynamic targets such as forklifts and personnel are set with time-decaying occupation grid to avoid dynamic obstacles during path planning; during navigation, when a new dynamic obstacle appears in front of the current navigation path and the predicted safety distance is lower than the preset safety threshold, a local re-planning algorithm is triggered to re-plan the navigation path in the local area, and after successful re-planning, it is smoothly switched to the new navigation path to continue moving to the target tray, the self-moving winding machine can periodically report its position, task progress and predicted completion time to the warehouse scheduling system regularly set in the working environment, and the warehouse scheduling system cooperatively schedules the channel occupation and task rhythm according to the state information of multiple mobile devices to reduce the path conflict between the self-moving winding machine and other automatic guided vehicles.

[0032] S2.2, after the self-moving winding machine is initially deployed or the sensor is replaced, a set of calibration procedures is performed, the self-moving winding machine is moved to the vicinity of the preset calibration board, the calibration image and pose data are collected, and the external parameter of the vision sensor coordinate system relative to the machine body coordinate system is calibrated to improve the accuracy of three-dimensional perception data and environment map.

[0033] Next, S3 comprises the following step: S3.1, the sector unit division step size is self-adaptively changed with the height, the step size representing the height and azimuth angle of the sector unit, the smaller the step size, the denser the grid, the higher the risk map resolution, but the larger the calculation amount, on the contrary, the grid is coarser, the calculation is faster, but the resolution is lower, the height step size and azimuth angle step size of the bottom reinforcement layer are smaller than those of the middle bearing layer and the top locking layer, so that the risk resolution of the bottom area is higher than that of the middle and top areas, in order to control the bottom reinforcement winding strategy more finely, specifically, since the bottom reinforcement layer directly bears the weight of the goods and is closely related to the forklift insertion, the pallet slip and the overturning starting position, slight deviation, concave-convex or local vacancy will have a significant impact on the overall stability, therefore, the bottom area is divided into multiple small sector units with small height interval and narrow azimuth angle span, in order to capture the stability difference and film covering condition of different positions of the bottom more finely, for the middle bearing layer and the top locking layer, since their influence on the overall stability is relatively uniform, the influence degree of a single local protrusion is lower than that of the bottom area, a relatively large height interval and azimuth angle span can be used for its division, thereby reducing the number of sector units, reducing the calculation and storage overhead, through the above-mentioned manner, the spatial resolution of the bottom area in the risk map is significantly higher than that of the middle and top areas, when planning the target coverage and executing the winding control, the self-propelled winding machine can implement more fine-grained winding strategies for different sector units of the bottom reinforcement layer, for example, increasing the number of winding layers or improving the overlap rate for the bottom edge, the fork tooth insertion direction and the bottom area with local protrusions, thereby significantly improving the anti-slip and anti-overturning ability of the bottom area under the premise of controllable film consumption.

[0034] S3.2, Determination of target winding layer number: by reading the stability risk value and target coverage of the sector unit in the risk map, combined with the pre-set single-layer film coverage contribution, the target coverage is converted into the target winding layer number, so that the target winding layer number is larger in the sector unit with higher stability risk or more stringent working condition requirement, and smaller in the sector unit with lower stability risk and more relaxed working condition requirement; the target overlap rate is allocated to each sector unit according to the height section, so that the basic overlap rate of the bottom reinforcing layer is higher than that of the middle bearing layer and the top locking layer, and on this basis, the overlap rate is adjusted up or down according to the stability risk value of each sector unit, so that the actual overlap rate of the high-risk direction with high stability risk value is higher than that of the low-risk direction with low stability risk value, thereby forming higher winding density in the high-risk direction; the target tension level is selected from the pre-set multiple tension levels according to the stability risk value and the fragile level of the goods, and the fragile level of the goods includes two levels of fragile and not fragile, when the stability risk is high and the goods are not fragile, a higher tension level is preferentially selected to enhance the binding effect, when the goods are fragile, the tension level is limited to not more than the medium level, and the stability decline caused by the limited tension is compensated by appropriately increasing the target winding layer number and the overlap rate, so as to realize the coordinated configuration among the target winding layer number, the target overlap rate and the target tension level.

[0035] S3.3, Control parameter trajectory: by extracting the outer contour polygon of the tray and the goods in the horizontal projection plane, taking the geometric center of the tray as the coordinate origin, arranging the discrete points on the outer contour in the order of azimuth angle, and according to the stability risk value and target coverage of the corresponding azimuth sector in the risk map, calculating the safety distance matched with each azimuth angle, so that the safety distance of the high-risk direction and the direction with higher target coverage is greater than that of the low-risk direction and the direction with lower target coverage; on this basis, each discrete point on the outer contour is moved outward along the radial direction by a corresponding safety distance and a winding machine shape envelope allowance to obtain a set of circumnavigation envelope points, the circumnavigation envelope points are smoothly connected and curve fitted to form a closed circumnavigation envelope path, and the target winding layer number, the target overlap rate and the target tension level are hung according to the height section and the azimuth sector corresponding to each path point, the walking speed, the lifting speed and the film release speed control parameter trajectory varying with the circumnavigation angle and the height are generated by time parameterization, so that the self-propelled winding machine can travel along the safety-inflated envelope path in the actual circumnavigation process, and execute differentiated winding strategies according to the risk level of different height sections and different azimuth sectors, the control parameter trajectory generated in the time dimension adopts trapezoidal velocity profile or S-curve velocity profile, so that the walking speed of the winding machine and the lifting speed of the film frame change smoothly in the acceleration and deceleration sections, so as to reduce the impact and vibration on the goods.

[0036] Next, S4 includes the following steps: S4.1, during the winding process, use a vision sensor or depth sensor to estimate the tilt angle of the goods relative to the pallet at each height. The stability risk value of the corresponding azimuth sector unit is dynamically updated based on the tilt angle. satisfy: .

[0037] in, The stability risk value before the update. This is the tilt risk gain coefficient. The angle of inclination of the cargo at that height relative to the vertical direction. For the indicator function of the azimuth sector corresponding to the tilt direction, when the angle between the sector azimuth and the tilt direction is less than a preset angle threshold... Select 1 otherwise select 0; during the wrapping operation, the self-propelled wrapping machine continuously estimates the tilt angle of the cargo relative to the vertical direction at each height based on the perception results of vision sensors or depth sensors. Furthermore, by combining the angle relationship between the tilt direction and each directional sector, the stability risk distribution of the corresponding height segment in the risk map is corrected online: when a significant tilt of the cargo in a specific direction is detected at a certain height, only in the directional sectors within that height segment where the angle with the tilt direction is less than a preset angle threshold, the original stability risk value is incrementally superimposed according to the product of the absolute value of the tilt angle and the tilt risk gain coefficient. This results in the stability risk value of the directional sector increasing monotonically with the degree of tilt, while other directional sectors not tilted in the direction of tilt remain unchanged at their original risk level. This allows the risk map to reflect the evolution characteristics of the cargo attitude disturbance in the two-dimensional space of height and directional during the winding process in real time, providing quantitative basis and decision input for subsequent winding strategies such as dynamically increasing target coverage, reducing the upper limit of walking speed and acceleration, and implementing directional reinforcement winding in the corresponding high-risk directional areas.

[0038] S4.2 When the absolute value of the tilt angle at any height exceeds the preset tilt threshold, directional reinforcement winding is triggered. During the directional reinforcement winding, the target coverage of the tilt direction and its adjacent azimuth sector units is increased in the risk map, the upper limit of walking speed and the upper limit of acceleration of the corresponding height segment are reduced, and the number of winding layers of the above sector units is increased preferentially in the subsequent predetermined number of circles.

[0039] S4.3 Monitoring of membrane tension is achieved by setting a tension sensor installed on the pre-stretching mechanism or by indirectly calculating the tension by detecting the current of the drive motor. When a sudden change in the actual tension exceeding the preset tension fluctuation threshold is detected in a short period of time, the walking speed and membrane release speed are temporarily reduced to avoid the membrane breaking or the cargo displacement being aggravated.

[0040] Next, S5 includes the following steps: S5.1, calculating the comprehensive quality risk value of each sector unit quality defect The calculation is as follows: .

[0041] wherein, is the proportion of the area of the sector unit not covered by the film to the visible area of the sector, is a measure of the degree to which the number of film covering layers in the sector is lower than the target coverage , is a normalized severity measure of quality defects such as slack, wrinkles, and broken film, , , is a quality risk weight coefficient; the target sector unit that needs to be supplemented with winding is determined by combining the stability risk value and the quality risk value, and when the weighted sum of the stability risk value and the quality risk value of a sector unit is greater than a safety threshold value, the sector unit is added to the supplementary winding set, and the increase in the coverage of each sector unit in the supplementary winding set is ensured to be not less than the preset coverage compensation amount when generating the local supplementary winding path; the safety threshold value is a judgment threshold value based on the risk map, and is used to determine whether the packaging state after the main winding and the local supplementary winding meets the overall safety requirement. The safety threshold value can be quantitatively represented by a pre-set global risk judgment condition, for example, the stability risk value and the quality risk value of each sector unit are combined in a predetermined manner, and when the comprehensive risk value of all sector units in the risk map is not more than the corresponding stability risk threshold value, and the proportion of the number of sector units whose comprehensive risk value exceeds the stability risk threshold value is lower than a preset proportion, it is considered that the risk map has converged below the safety threshold value; otherwise, when there are still a large number of sector units that are continuously higher than their stability risk threshold value within the predetermined winding time and the number of supplementary windings, it is considered that the risk map cannot converge below the safety threshold value.

[0042] S5.2, the self-propelled winding machine is provided with an automatic mode and a manual auxiliary mode, when the risk map cannot converge below the safety threshold value within the predetermined winding time and the number of supplementary windings, the system automatically switches to the manual auxiliary mode, suspends the automatic winding at the same time, and reports the risk sector position and the risk type to the upper system, prompting the operator to adjust the goods stacking or winding configuration; the upper system can be any one or a combination of a warehouse management system, a production management system, or a logistics scheduling system, used to provide the self-propelled winding machine with configuration data such as goods category, pallet information, and working conditions.

[0043] S5.3, the generation of the local supplementary winding path aims to minimize the supplementary winding movement distance and the supplementary winding time consumption, under the constraint condition that all sector unit coverage increase in the supplementary winding set is not lower than the preset lower limit, a plurality of candidate supplementary winding path schemes are evaluated, and the scheme with the optimal comprehensive path length and supplementary winding time index is selected as the executed supplementary winding path.

[0044] S5.4, the self-propelled winding machine supports multi-stage winding configuration in the same pallet task, after completing the main winding stage and performing at least one local supplementary winding, an additional reinforcement stage can be selectively started according to the change result of the risk map, and additional reinforcement winding is performed on the azimuth sector corresponding to the fork insertion position, the opening direction of the pallet, and the force direction of the fork.

[0045] Finally, S6 includes the following steps: S6.1, when outputting the winding task result, the self-propelled winding machine generates a packaging quality report according to the final risk map, and the packaging quality report at least includes the final coverage distribution of each height section, the reinforcement winding area distribution, the supplementary winding path execution times, and the detected quality defect types and processing results.

[0046] S6.2, the experience updating rule of the initial parameters of the risk map adopts an exponential weighting method, and a certain risk parameter is updated according to the following formula: .

[0047] wherein, is the historical value of the parameter under similar goods and similar working conditions in the last round, is the observation value estimated according to the actual inclination trajectory and the damage condition in this task, The experience update step coefficient is between 0 and 1, and the experience update is only performed when the similarity of the cargo category and the working condition feature is not lower than a preset threshold; the similar cargo and the similar working condition are determined by the comprehensive similarity of the cargo category feature vector and the working condition feature vector, for example, the cargo category similarity and the working condition similarity are obtained by using the cosine similarity or the weighted Euclidean distance, and then the comprehensive similarity is obtained by combining the preset weight coefficients; in addition, the control unit is provided with a preset task classification threshold, and when the similarity between two tasks is not lower than the task classification threshold, it is considered that the two tasks belong to the same type of task, so that the experience update of the risk parameter is only performed between the samples belonging to the same type of task; the cargo category feature vector is generated by the cargo information issued by the upper system or manually input, and the cargo information at least includes the cargo category, the shape size range, the single pallet weight level, the stacking layer number and the like, and each field is converted into a numerical value in a preset order and combined into a fixed-length cargo category feature vector; the working condition feature vector is generated according to the working condition information corresponding to the current task, and the working condition information at least includes the stacking height level of the pallet in the subsequent operation, the transportation distance level, the transportation mode, the road condition or the vibration level and the like, and each field is converted into a numerical value in a preset order and combined into a fixed-length working condition feature vector.

[0048] S6.3, the control method is suitable for various different specifications of pallets and different sizes of cargos, when it is detected that the pallet size or the cargo profile exceeds the preset range, the self-propelled winding machine automatically adjusts the height range and the azimuth resolution of the risk map, and re-plans the navigation path and the winding strategy, so as to ensure that the winding control based on the risk map can be realized under different pallet specifications and cargo sizes.

[0049] The embodiments of the application are described in detail above in combination with the drawings, but the application is not limited thereto, and various changes can be made within the knowledge of those skilled in the art without departing from the purpose of the application.

Claims

1. A control method of a self-propelled winding machine based on visual recognition and autonomous path planning, characterized in that, The method comprises the following steps: S1, obtaining three-dimensional perception data of the target pallet and the goods by a visual sensor or a depth sensor of the self-propelled wrapping machine, determining the boundary of the pallet and the outer contour of the goods, dividing the space around the goods into a plurality of sector units according to height and azimuth sector, calculating the stability risk value of each sector unit, setting the target coverage of each sector unit, and constructing a risk map; S2, performing a simultaneous localization and mapping algorithm based on the camera image sequence obtained by the visual sensor, when planning a navigation path and a close direction, taking the risk map as a reference, preferentially selecting the side close to the high-risk direction as the starting side of the detour, and driving the self-propelled wrapping machine to move along the navigation path to the vicinity of the target pallet to determine the starting pose of the detour and the first wrapping direction; S3, based on the risk map and the outer contour of the goods, the control unit divides the wrapping area into different height layers according to the height direction, and configures a basic coverage constant for each height layer, and determines the target wrapping layer number, the target overlap rate and the target tension level of each sector unit according to the corresponding stability risk value and the target coverage, and constructs a control parameter trajectory; S4, while performing the wrapping operation according to the control parameter trajectory, periodically detecting the state of the goods, updating the actual coverage of each sector unit in the risk map, and when the displacement or inclination of the goods in a certain direction exceeds a preset threshold, increasing the stability risk value of the sector unit in the corresponding direction in the risk map and increasing its target coverage; S5, after reaching the preset wrapping number and height coverage condition, controlling the self-propelled wrapping machine to wrap around the goods in a detection mode, identifying the quality defects of the film in each sector unit through the visual sensor, writing the defect results into the quality state in the risk map, and based on the joint determination of the stability risk and the quality risk of each sector unit, obtaining a target sector unit set that needs to be supplemented, generating a local supplementary wrapping path according to the current position and the position of the target sector unit, and performing supplementary wrapping; S6, after the wrapping and supplementary wrapping are completed, storing the final risk map, the control parameter trajectory used in the wrapping process, the goods category and the corresponding subsequent stacking and transportation working conditions in association, and based on the preset experience updating rule, updating the initial parameters and wrapping strategies of the risk map for subsequent similar goods and similar working conditions.

2. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 1, wherein, The division of the sector unit comprises: first establishing a space rectangular coordinate system with the geometric center of the pallet as the origin, letting the X-axis and the Y-axis be parallel to two mutually perpendicular edges of the pallet, and the Z-axis be perpendicular and point upward, thereby obtaining a pallet coordinate system, determining the pallet surface height and the maximum height of the goods according to the three-dimensional perception data, and determining the circumcircle radius of the pallet outer edge in the XY plane; In the height direction, the interval from the pallet surface to the maximum height of the goods is divided into a plurality of height sections; In the horizontal direction, the X-axis of the pallet coordinate system is taken as the zero angle direction, and the planar angle range of one turn around the pallet is divided into a plurality of equal intervals according to a preset angle step from 0 degrees to 360 degrees. The space region from the tray surface to the maximum height of the goods around the tray is divided into a series of sector units in height sections and azimuth sectors, each sector unit being determined by its height section number and azimuth sector number.

3. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 2, characterized in that, The stability risk value is calculated in a linear weighting manner, satisfying: ； wherein, is a stability risk value for representing the local stability risk, is a height section where the sector unit is located, is an azimuth angle of the sector unit, is an offset amount of the cargo mass center under the height section relative to the pallet geometric center in the direction, is an offset amount of the cargo mass center under the height section relative to the pallet geometric center in the direction, is a difference value of the cargo mass center in the height section exceeding the reference height relative to the pallet surface, , , is a weight coefficient pre-configured according to the cargo category; For each sector unit, a stability risk threshold corresponding thereto is also preset, which represents the maximum stability risk level allowed for the sector unit under the current goods category and working condition, and the stability risk threshold is written into the risk map.

4. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 3, characterized in that, The target coverage degree is calculated according to the stability risk value and the working condition risk weight, satisfying: ； wherein, is a target coverage degree for each sector unit, is a base coverage constant, , is a target coverage degree adjustment coefficient, used to adjust the relative influence weight of stability risk and working condition risk in the calculation of target coverage degree, is a working condition risk weight determined in advance according to the height of the subsequent stack, and according to the transportation distance and transportation vibration level belonging to the transportation working condition, and is calculated as is written into the risk map as the target coverage degree of the sector unit.

5. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 4, wherein, The determination of the target winding layer number: the target coverage degree is converted into the target winding layer number by reading the stability risk value and the target coverage degree of the sector unit in the risk map and combining the preset single-layer film coverage contribution; The target overlap rate is assigned to each sector unit according to the height section, so that the basic overlap rate of the bottom reinforcing layer is higher than that of the middle bearing layer and the top locking layer, and on this basis, the overlap rate is adjusted up or down according to the stability risk value of each sector unit, so that the actual overlap rate in the high-risk direction is higher than that in the low-risk direction; The target tension level is selected from a plurality of preset tension levels according to the stability risk value of each sector unit and the goods fragility level.

6. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 5, wherein, The control parameter trajectory is obtained by extracting the outer contour polygon of the tray and the goods in the horizontal projection plane, arranging the discrete points on the outer contour in the order of azimuth angle with the geometric center of the tray as the coordinate origin, and calculating the safety distance matched with each azimuth angle according to the stability risk value and the target coverage degree of the corresponding azimuth sector in the risk map, so that the safety distance in the high-risk direction and the direction with high target coverage degree is greater than that in the low-risk direction and the direction with low target coverage degree; Each discrete point on the outer contour is moved outward by a corresponding safety distance and a winding machine shape envelope allowance to obtain a set of circumnavigation envelope points, the circumnavigation envelope points are smoothly connected and curve-fitted to form a closed circumnavigation envelope path, and the target winding layer number, the target overlap rate and the target tension level corresponding to each path point are hung according to the height section and the azimuth sector, and the walking speed, the lifting speed and the film release speed control parameter trajectories varying with the circumnavigation angle and the height are generated by time parameterization, so that the self-propelled winding machine can travel along the safety-inflated envelope path during actual circumnavigation and execute differentiated winding strategies according to the risk levels of different height sections and different azimuth sectors.

7. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 6, wherein, In the winding process, the inclination angle of the goods at each height relative to the pallet is calculated by using a visual sensor or a depth sensor And the stability risk value of the corresponding orientation sector unit is dynamically updated according to the inclination angle Satisfies: ； wherein, is the stability risk value before the update, is the tilt risk gain coefficient, is the tilt angle of the cargo at this height with respect to the vertical direction is an indicator function of the azimuth sector corresponding to the tilt direction; During the winding operation, the self-propelled winding machine continuously calculates the inclination angle of the goods at each height relative to the vertical direction based on the sensing results of the visual sensor or the depth sensor and combines the inclination direction with the included angle relationship of each azimuth sector to online correct the stability risk distribution of the corresponding height section in the risk map, so that the risk map can reflect the evolution characteristics of the goods posture disturbance in the height and azimuth two-dimensional space during the winding process, providing quantitative basis and decision input for subsequent winding strategies such as dynamically increasing the target coverage, reducing the upper limit of walking speed and acceleration, and implementing directional reinforcement winding in the corresponding high-risk azimuth.

8. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 7, wherein, a comprehensive quality risk value for quality defects of each sector unit is calculated as follows: ； wherein, is the proportion of the area of the sector not covered by the film to the visible area of the sector, is the degree measure of the extent to which the number of film layers covering the sector is less than the target coverage , is the normalized severity measure of quality defects such as slack, wrinkles, and broken film, , , is the quality risk weight coefficient; The target sector units that need to be supplemented are determined by combining the stability risk value and the quality risk value, and when the weighted sum of the stability risk value and the quality risk value of a sector unit is greater than a safety threshold, the sector unit is added to the supplementary winding set, and the coverage degree of each sector unit in the supplementary winding set is ensured to increase by no less than a preset coverage compensation amount when generating a local supplementary winding path.

9. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 8, wherein, The empirical updating rule for the initial parameters of the risk map adopts an exponential weighting manner, and a certain risk parameter is updated according to the following formula: ； wherein, is the historical value of the parameter under similar goods and similar working conditions in the last round, is the observation value estimated according to the actual tilt trajectory and damage condition of the current task, is an experience update step length coefficient between 0 and 1, and the above experience update is only performed when the similarity of the goods category and the working condition characteristics is not lower than a preset threshold. Similar goods and similar working conditions are determined by the comprehensive similarity of the goods category feature vector and the working condition feature vector. When the similarity between two tasks is not lower than the task classification threshold, it is considered to belong to the same type of task, so that the experience update of the risk parameter is only carried out between the samples classified into the same type of task.

10. The control method of the vision recognition and autonomous path planning based self-moving winding machine according to claim 9, wherein, The control method is suitable for various different specifications of pallets and different sizes of goods. When it is detected that the pallet size or the goods profile exceeds the preset range, the self-propelled winding machine automatically adjusts the height range and azimuth resolution of the risk map, and re-plans the navigation path and winding strategy.

Images