A method for dynamic planning of floor machine routes

Abstract

The present invention discloses a method for dynamically planning the route of a floor machine, which belongs to the technical field of floor machine control. An electronic map of an area to be operated is obtained, and a target operation route is planned based on the electronic map, so that the initial route planning of the floor machine can be realized. Then, a pre-deployed obstacle recognition model is used to recognize the operation image to determine the obstacles in the operation process, and the obstacles are marked in the electronic map, so that the floor machine can effectively avoid the obstacles to work, and even if the collected map is different from the actual situation, the floor machine can still work effectively.

Description

Technical Field

[0001] The invention belongs to the technical field of floor machine control, and in particular relates to a floor machine route dynamic planning method. Background Art

[0002] Floor machine, whose full name is floor construction machinery, is a kind of professional mechanical equipment used for ground construction and maintenance. It is mainly used for a series of processing work on hard floors such as concrete, asphalt, stone, etc., including grinding, grinding, polishing, cleaning, cutting, paving and other operations. Floor machines are widely used in the fields of construction, decoration, road construction, etc. In the prior art, fully automatic floor machines have appeared, but the existing floor machines usually use preset routes to operate during the operation process. However, the actual working environment is complex and changeable, and the preset routes are difficult to adapt to various scenarios, resulting in low working efficiency and high energy consumption. To this end, it is necessary to study a method that can dynamically plan the operation route of the floor machine according to the actual environment. Summary of the invention

[0003] The invention provides an intelligent control method for building water supply and drainage, which is used to solve the problem that the existing floor machine is difficult to be applied to complex or changeable environments.

[0004] A method for dynamically planning a route for a floor machine, comprising:

[0005] At the initial moment, an electronic map of the area to be operated is obtained, and a target operation route is planned based on the electronic map;

[0006] Based on the target operation route, the floor machine is controlled to operate, and operation images are collected during the operation to obtain operation images;

[0007] Use the pre-deployed obstacle recognition model to identify the operation image to determine the obstacles in the operation process and mark the obstacles in the electronic map;

[0008] When the floor machine collides with an obstacle while operating along the target operating route, the target operating route is revised and the actual route of the floor machine is recorded at the same time;

[0009] When a deviation occurs between the actual travel route and the target operation route, the floor machine is controlled to return to the target operation route to continue operation, thereby realizing dynamic planning of the floor machine route.

[0010] Furthermore, planning a target operation route based on the electronic map includes:

[0011] The electronic map of the area to be operated is gridded, and the straight edge closest to the edge is selected as the initial starting direction. All grids are connected in a serpentine shape to obtain the target operation route.

[0012] Furthermore, the pre-deployment method of the obstacle recognition model includes:

[0013] The obstacle recognition model is constructed using the YOLO v3 algorithm, and the hyperparameters of the obstacle recognition model are randomly generated. The hyperparameters are encoded into vectors to obtain a hyperparameter vector, and multiple different hyperparameter vectors are repeatedly obtained.

[0014] Obtain the loss function value corresponding to each hyperparameter vector, and determine the hyperparameter vector with the smallest loss function value as the optimal hyperparameter vector;

[0015] According to the optimal hyperparameter vector, an adaptive local area search is performed on the hyperparameter vector using a population information full interaction strategy to obtain a hyperparameter vector after the adaptive local area search;

[0016] For the hyperparameter vector after the adaptive local area search, the individual information interaction strategy is used to perform a local area expansion search on the hyperparameter vector to obtain the hyperparameter vector after the local area expansion search;

[0017] For the hyperparameter vector after the local area expansion search, according to the optimal hyperparameter vector, the optimal direction search is performed on the hyperparameter vector using the optimal position guidance strategy to obtain the hyperparameter vector after the optimal direction search;

[0018] For the hyperparameter vector after the optimal direction search, a greedy distributed estimation strategy is used to perform a global region search on the hyperparameter vector to obtain the hyperparameter vector after the global region search;

[0019] Determine whether the training end condition is met. If so, deploy the obstacle recognition model according to the hyperparameter vector after the global area search, otherwise return to the adaptive local area search step.

[0020] Furthermore, according to the optimal hyperparameter vector, an adaptive local area search is performed on the hyperparameter vector using a population information full interaction strategy to obtain a hyperparameter vector after the adaptive local area search, including:

[0021] For the i A hyperparameter vector is generated to determine the mutual influence strength and Euclidean distance between the hyperparameter vector and other hyperparameter vectors;

[0022] For the i a hyperparameter vector, and determining the influence vector of other hyperparameter vectors on the hyperparameter vector according to the mutual influence strength and the Euclidean distance;

[0023] Based on the current number of training times, determine the population information full interaction coefficient;

[0024] For thei A hyperparameter vector is generated, and the first search vector corresponding to the hyperparameter vector is determined by incorporating the boundary influence according to the influence vector of other hyperparameter vectors on the hyperparameter vector and the full interaction coefficient of population information;

[0025] According to the first search vector corresponding to each hyperparameter vector, all hyperparameter vectors are searched around the optimal hyperparameter vector to obtain the hyperparameter vector after adaptive local area search.

[0026] Furthermore, for the hyperparameter vector after the adaptive local area search, an individual information interaction strategy is used to perform a local area expansion search on the hyperparameter vector to obtain a hyperparameter vector after the local area expansion search, including:

[0027] For the hyperparameter vector after the nth adaptive local area search, randomly match another hyperparameter vector to the hyperparameter vector to obtain the information interaction vector corresponding to each hyperparameter vector;

[0028] Determine a second search vector corresponding to each hyperparameter vector according to the information interaction vector corresponding to each hyperparameter vector, and perturb the second search vector using a cosine function combined with a time function to obtain a perturbed second search vector;

[0029] According to the second search vector after the disturbance, a local area expansion search is performed on the hyperparameter vector to obtain a hyperparameter vector after the local area expansion search.

[0030] Furthermore, for the hyperparameter vector after the local area expansion search, the optimal direction search is performed on the hyperparameter vector according to the optimal hyperparameter vector and the optimal position guidance strategy, and the hyperparameter vector after the optimal direction search is obtained, including:

[0031] Determine a nonlinear convergence factor for the hyperparameter vector after the m-th local region expansion search;

[0032] According to the optimal hyperparameter vector, a third search vector corresponding to the hyperparameter vector is determined by a random weighting method, and the third search vector is perturbed by a nonlinear convergence factor to obtain a perturbed third search vector;

[0033] According to the third search vector after the disturbance, an optimal direction search is performed on the hyperparameter vector to obtain the hyperparameter vector after the optimal direction search.

[0034] Furthermore, for the hyperparameter vector after the optimal direction search, a greedy distributed estimation strategy is used to perform a global region search on the hyperparameter vector to obtain the hyperparameter vector after the global region search, including:

[0035] For the hyperparameter vector after the optimal direction search, the hyperparameter vector is arranged in descending order according to the loss function value to obtain the arranged hyperparameter vector;

[0036] According to the arranged hyperparameter vectors, the sequence number of each hyperparameter vector is determined, and the weight influence factor of each hyperparameter vector is determined according to the sequence number of each hyperparameter vector;

[0037] The hyperparameter vectors are weighted according to the weight influence factor of each hyperparameter vector to determine the population comprehensive position vector;

[0038] Determine a global search factor corresponding to each hyperparameter vector according to the population comprehensive position vector, and perform a global region search on the hyperparameter vector according to the population comprehensive position vector and the global search factor to obtain a hyperparameter vector after the global region search;

[0039] Determine whether the loss function value corresponding to the hyperparameter vector after the global area search is reduced. If so, accept the global area search; otherwise, reject the global area search.

[0040] Furthermore, based on the target operation route, the process of controlling the floor machine to operate also includes: taking the starting position of the floor machine as the origin and the initial starting direction as the positive X-axis coordinate direction, updating the coordinate points of the travel position in real time to determine the coordinate points of the actual travel route.

[0041] Furthermore, a pre-deployed obstacle recognition model is used to recognize the operation image to determine obstacles in the operation process and mark the obstacles in the electronic map, including:

[0042] Use the pre-deployed obstacle recognition model to identify the operation image to determine the obstacles in the operation process;

[0043] When an obstacle is identified, the coordinates of the obstacle are determined based on the coordinate point of the current floor machine using the binocular vision positioning method;

[0044] The obstacle is marked in the electronic map according to the coordinates of the obstacle.

[0045] Furthermore, when the floor machine collides with an obstacle while operating along the target operating route, the target operating route is revised, including:

[0046] When the floor tractor collides with an obstacle while operating along the target operating route, the obstacle is partially bypassed along the left or right side of the straight route and the target operating route is revised.

[0047] The present invention provides a method for dynamic route planning of a floor machine, which obtains an electronic map of an area to be operated, and plans a target operation route based on the electronic map, thereby realizing initial route planning of the floor machine, and then uses a pre-deployed obstacle recognition model to recognize the operation image to determine obstacles in the operation process, and marks the obstacles in the electronic map, so that the floor machine can effectively avoid obstacles to work, and even if the collected map is different from the actual situation, the floor machine can still work effectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0049] Figure 1 A flow chart of a method for dynamically planning a floor machine route provided in an embodiment of the present invention.

[0050] Figure 2 A schematic diagram of planning a target operation route provided by an embodiment of the present invention.

[0051] Figure 3 A schematic diagram of filling the remaining area at the edge to form a complete working area of ​​a floor machine provided in an embodiment of the present invention.

[0052] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0053] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0054] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0055] In most existing construction scenarios, the construction sites are basically rectangular, and there are pillars or other decorative objects in the construction sites. Therefore, the embodiments of the present invention are implemented based on this background.

[0056] like Figure 1 As shown, an embodiment of the present invention provides a method for dynamically planning a route for a floor machine, comprising:

[0057] S11. At an initial moment, an electronic map of the area to be operated is obtained, and a target operation route is planned based on the electronic map;

[0058] The embodiment of the present invention preferably constructs an electronic map of the area to be operated through the RBPF-SLAM algorithm, but problems such as incomplete obstacle recognition or blurred edges may occur. Therefore, the embodiment of the present invention introduces obstacle recognition in the subsequent process, which can help the floor machine effectively avoid obstacles.

[0059] Optionally, since the embodiment of the present invention introduces obstacle recognition, the electronic map may be imported in other ways, and then route planning may be performed based on the imported electronic map.

[0060] S12, based on the target operation route, controlling the floor machine to operate, and collecting operation images during the operation process to obtain operation images;

[0061] Optionally, binocular vision can be used to collect and identify the working image. After the obstacle is identified, the coordinates of the obstacle can be determined by binocular vision, thereby achieving obstacle avoidance.

[0062] S13, using a pre-deployed obstacle recognition model to recognize the operation image to determine obstacles in the operation process, and marking the obstacles in the electronic map;

[0063] For example, the YOLO algorithm can be used to pre-deploy an obstacle recognition model, so that various obstacles can be effectively identified and marked.

[0064] S14. When the floor machine collides with an obstacle while operating along the target operation route, the target operation route is revised, and the actual travel route of the floor machine is recorded at the same time;

[0065] For example, the target working route can be made to bypass obstacles to the left or right, and the actual travel route of the floor machine can be recorded at the same time, which can effectively determine which areas have been worked on and which areas have not been worked on.

[0066] S15. When a deviation occurs between the actual travel route and the target operation route, the floor machine is controlled to return to the target operation route to continue operation, thereby realizing dynamic planning of the floor machine route.

[0067] For example, when deviation is found, the leveler can be driven promptly to approach the planned walking route through a lateral reciprocating mechanism or other mechanisms.

[0068] The present invention provides a method for dynamic route planning of a floor machine, which obtains an electronic map of an area to be operated, and plans a target operation route based on the electronic map, thereby realizing initial route planning of the floor machine, and then uses a pre-deployed obstacle recognition model to recognize the operation image to determine obstacles in the operation process, and marks the obstacles in the electronic map, so that the floor machine can effectively avoid obstacles to work, and even if the collected map is different from the actual situation, the floor machine can still work effectively.

[0069] In an embodiment of the present invention, planning a target operation route based on the electronic map includes:

[0070] The electronic map of the area to be operated is gridded, and the straight edge closest to the edge is selected as the initial starting direction. All grids are connected in a serpentine shape to obtain the target operation route.

[0071] like Figure 2 As shown in the figure, the width of the gridded electronic map of the area to be operated is the operating width of the floor machine, and then all the grids are connected in a serpentine shape to obtain the target operation route. However, in actual situations, the operating width is often not enough at the edge, so the adjacent position can be used to fill it up, such as Figure 3 As shown, the remaining area of ​​the edge is connected to the borrowed area to form a complete workable area, allowing the floor machine to work smoothly.

[0072] In an embodiment of the present invention, a method for pre-deploying an obstacle recognition model includes:

[0073] S21. Use the YOLO v3 algorithm to build an obstacle recognition model, randomly generate hyperparameters of the obstacle recognition model, encode the hyperparameters into a vector, obtain a hyperparameter vector, and repeatedly obtain multiple different hyperparameter vectors;

[0074] S22, obtaining the loss function value corresponding to each hyperparameter vector, and determining the hyperparameter vector with the smallest loss function value as the optimal hyperparameter vector;

[0075] For example, a pre-calibrated image can be used to train an obstacle recognition model, so that the loss function value corresponding to each hyperparameter vector can be obtained. The smaller the loss function value corresponding to the hyperparameter vector, the better the hyperparameter vector.

[0076] S23, according to the optimal hyperparameter vector, an adaptive local area search is performed on the hyperparameter vector using a population information full interaction strategy to obtain a hyperparameter vector after the adaptive local area search;

[0077] S24, for the hyperparameter vector after the adaptive local area search, using the individual information interaction strategy to perform a local area expansion search on the hyperparameter vector to obtain the hyperparameter vector after the local area expansion search;

[0078] S25, for the hyperparameter vector after the local area expansion search, according to the optimal hyperparameter vector, and using the optimal position guidance strategy to search the hyperparameter vector in the optimal direction, to obtain the hyperparameter vector after the optimal direction search;

[0079] S26, for the hyperparameter vector after the optimal direction search, a greedy distributed estimation strategy is used to perform a global region search on the hyperparameter vector to obtain the hyperparameter vector after the global region search;

[0080] S27, judging whether the training end condition is met, if so, deploying the obstacle recognition model according to the hyperparameter vector after the global area search, otherwise returning to the adaptive local area search step.

[0081] In the prior art, the algorithm parameters (such as weights) of the YOLO v3 algorithm need to be optimized before the YOLO v3 algorithm can normally perform target recognition. The prior art usually uses optimization algorithms such as particle swarm optimization and genetic algorithm to optimize the algorithm parameters of the YOLO v3 algorithm, which either falls into the local optimum too early or has the problem of poor convergence accuracy. Therefore, the embodiment of the present invention provides a pre-deployment method of an obstacle recognition model to solve the technical problems existing in the prior art, thereby improving the obstacle recognition accuracy.

[0082] Optionally, determining whether the training end condition is met may include: determining whether the current training times are greater than or equal to the maximum training times, and if so, determining that the training end condition is met, otherwise determining that the training end condition is not met.

[0083] In an embodiment of the present invention, according to the optimal hyperparameter vector, an adaptive local area search is performed on the hyperparameter vector using a population information full interaction strategy to obtain a hyperparameter vector after the adaptive local area search, including:

[0084] A1. For i A hyperparameter vector is generated to determine the mutual influence strength and Euclidean distance between the hyperparameter vector and other hyperparameter vectors;

[0085] In this embodiment, the mutual influence strength between the hyperparameter vector and other hyperparameter vectors is:

[0086]

[0087] in, Indicates tDuring the training process i The hyperparameter vector is k The hyperparameter vector is about d The mutual influence strength of the dimensional parameters, represents the attraction strength parameter and is set to 0.5. l represents the attraction scale parameter and is set to 1.5, e represents the natural constant, Indicates t During the training process i The first hyperparameter vector d Dimension parameters, Indicates t During the training process k The first hyperparameter vector k Dimension parameters, i =1,2,…,N, k =1,2,…,N, and i Not equal to k , N represents the total number of hyperparameter vectors, d =1,2,…,D, where D represents the total dimension of the hyperparameter vector.

[0088] A2. For i a hyperparameter vector, and determining the influence vector of other hyperparameter vectors on the hyperparameter vector according to the mutual influence strength and the Euclidean distance;

[0089] In this embodiment, the influence vector of other hyperparameter vectors on the hyperparameter vector is:

[0090]

[0091] in, Indicates t During the training process i The hyperparameter vector is k The influence vector between the hyperparameter vectors d Dimension parameters, Indicates t During the training process i Hyperparameter vector With k Hyperparameter vector The Euclidean distance between .

[0092] A3. Determine the total interaction coefficient of population information based on the current number of training times;

[0093] In this embodiment, the population information full interaction coefficient is:

[0094]

[0095] in, Indicates the preset maximum value of the total interaction coefficient of population information, which is set to 1; It represents the preset minimum value of the full interaction coefficient of population information, which is set to 0.00001; Indicates the preset maximum number of training times.

[0096] A4. For i A hyperparameter vector is generated, and the first search vector corresponding to the hyperparameter vector is determined by incorporating the boundary influence according to the influence vector of other hyperparameter vectors on the hyperparameter vector and the full interaction coefficient of population information;

[0097] In this embodiment, the first search vector is:

[0098]

[0099] in, The first search vector d Dimension parameters, Indicates d The upper bound of the dimension parameter, Indicates d The lower bound of the dimension parameter.

[0100] A5. According to the first search vector corresponding to each hyperparameter vector, all hyperparameter vectors are searched around the optimal hyperparameter vector to obtain the hyperparameter vector after the adaptive local area search.

[0101] In this embodiment, the hyperparameter vector after the adaptive local area search is:

[0102]

[0103] in, Represents the hyperparameter vector after adaptive local region search No. d Dimension parameters, represents the optimal hyperparameter vector d Dimension parameters.

[0104] The population information full interaction strategy provided by the embodiment of the present invention can enable the hyperparameter vector to learn the information of other hyperparameter vectors and search around the optimal hyperparameter vector. While ensuring the efficiency of the algorithm, it can improve the search in unfamiliar areas and greatly avoid repeated searches. The early algorithm should conduct large-scale searches on the entire search domain, and the later local exploration based on the optimal individual can also avoid falling into the local optimal value to a certain extent.

[0105] In an embodiment of the present invention, for a hyperparameter vector after an adaptive local area search, an individual information interaction strategy is used to perform a local area expansion search on the hyperparameter vector to obtain a hyperparameter vector after the local area expansion search, including:

[0106] B1. For the hyperparameter vector after the nth adaptive local area search, randomly match another hyperparameter vector to the hyperparameter vector to obtain the information interaction vector corresponding to each hyperparameter vector;

[0107] B2. Determine a second search vector corresponding to each hyperparameter vector according to the information interaction vector corresponding to each hyperparameter vector, and perturb the second search vector using a cosine function combined with a time function to obtain a perturbed second search vector;

[0108] In this embodiment, the second search vector after the disturbance is:

[0109]

[0110] in, represents the second search vector after perturbation, represents the maximum number of training times, Indicates t During the training process n A hyperparameter vector, Represents the hyperparameter vector The corresponding information interaction vector, |*| means taking the absolute value of each element, represents pi, represents the first random number between (0,1), represents the second random number between (0,1), n =1,2,…,N.

[0111] B3. Perform a local area expansion search on the hyperparameter vector according to the second search vector after the disturbance to obtain a hyperparameter vector after the local area expansion search.

[0112]

[0113] in, Indicates t During the training process n The first hyperparameter vector d Dimension parameters, Represents the hyperparameter vector after the local region expansion search No. d Dimension parameters, represents the second search vector after perturbation d Dimension parameters.

[0114] The local area extended search provided by the embodiment of the present invention can enable the hyperparameter vector to search to other positions in the form of a fluctuation curve, and can explore different areas more extensively, balancing the local and global development capabilities of the algorithm.

[0115] In an embodiment of the present invention, for a hyperparameter vector after a local area expansion search, an optimal direction search is performed on the hyperparameter vector according to an optimal hyperparameter vector and an optimal position guidance strategy is adopted to obtain a hyperparameter vector after the optimal direction search, including:

[0116] C1, determining the nonlinear convergence factor for the hyperparameter vector after the m-th local area expansion search;

[0117] In this embodiment, the nonlinear convergence factor is:

[0118]

[0119] in, Indicates t The nonlinear convergence factor during the training process, represents the starting value of the nonlinear convergence factor and is set to 2.5; represents the final value of the nonlinear convergence factor and is set to 0.0001.

[0120] C2. According to the optimal hyperparameter vector, a third search vector corresponding to the hyperparameter vector is determined by a random weighting method, and the third search vector is perturbed by a nonlinear convergence factor to obtain a perturbed third search vector;

[0121] In this embodiment, the third search vector after the disturbance is:

[0122]

[0123]

[0124] in, represents the third search vector after perturbation, represents the perturbation search factor, represents the hyperparameter vector after the mth local region expansion search during the tth training process, represents the optimal hyperparameter vector, Represents a third random number between (0,1).

[0125] C3. According to the third search vector after the disturbance, the optimal direction search is performed on the hyperparameter vector to obtain the hyperparameter vector after the optimal direction search.

[0126] In this embodiment, the hyperparameter vector after the optimal direction search is:

[0127]

[0128] in, The third search vector d Dimension parameters, Represents the hyperparameter vector No. d Dimension parameters, Represents the hyperparameter vector after optimal direction search No. d Dimension parameters.

[0129] The optimal direction search provided by the embodiment of the present invention can enable the algorithm to continuously search for the optimal position based on its own position and the position corresponding to the optimal individual, and a nonlinear convergence factor is set, which can have a higher search speed in the early stage of the algorithm and a higher search accuracy in the later stage of the algorithm, thereby improving the algorithm's search capability.

[0130] In an embodiment of the present invention, for the hyperparameter vector after the optimal direction search, a greedy distributed estimation strategy is used to perform a global area search on the hyperparameter vector to obtain the hyperparameter vector after the global area search, including:

[0131] D1. For the hyperparameter vector after searching the optimal direction, arrange the hyperparameter vector in descending order according to the loss function value to obtain the arranged hyperparameter vector; after arranging the hyperparameter vector, the later the arrangement, the better the position of the hyperparameter vector.

[0132] D2. Determine the sequence number of each hyperparameter vector according to the arranged hyperparameter vectors, and determine the weight influence factor of each hyperparameter vector according to the sequence number of each hyperparameter vector;

[0133] In this embodiment, the weight influencing factor is:

[0134]

[0135] in, Indicates j The weight influence factor corresponding to the hyperparameter vector after permutation, Represents the total number of hyperparameter vectors.

[0136] D3. Weight the hyperparameter vectors according to the weight influence factor of each hyperparameter vector to determine the population comprehensive position vector;

[0137] In this embodiment, the population comprehensive position vector is:

[0138]

[0139] in, Indicates t The population comprehensive position vector during the training process, Indicates t During the training process j The hyperparameter vector after searching for optimal directions.

[0140] D4. Determine a global search factor corresponding to each hyperparameter vector according to the population comprehensive position vector, and perform a global region search on the hyperparameter vector according to the population comprehensive position vector and the global search factor to obtain a hyperparameter vector after the global region search;

[0141] In this embodiment, the global search factor corresponding to each hyperparameter vector is determined according to the population comprehensive position vector as follows:

[0142]

[0143] in, Represents the hyperparameter vector The corresponding global search factor, T represents transposition;

[0144] A global region search is performed on the hyperparameter vector according to the population comprehensive position vector and the global search factor, and the hyperparameter vector obtained after the global region search is:

[0145]

[0146] in, The first d Dimension parameters, represents the optimal hyperparameter vector d Dimension parameters, Represents the hyperparameter vector No. d Dimension parameters, Represents the hyperparameter vector after global region search No. d Dimension parameters, represents a random number, and Obey (0, ) has a normal distribution.

[0147] D5. Determine whether the loss function value corresponding to the hyperparameter vector after the global area search is reduced. If so, accept the global area search; otherwise, reject the global area search.

[0148] The greedy distributed estimation strategy provided by the embodiment of the present invention can effectively assist the entire population to escape from the local optimum, effectively improve the global search capability of the algorithm, and ultimately improve the obstacle recognition effect.

[0149] In an embodiment of the present invention, the process of controlling the floor machine to operate based on the target operating route also includes: taking the starting position of the floor machine as the origin and the initial starting direction as the positive X-axis coordinate direction, updating the coordinate points of the travel position in real time to determine the coordinate points of the actual travel route.

[0150] In an embodiment of the present invention, a pre-deployed obstacle recognition model is used to recognize the operation image to determine obstacles in the operation process and mark the obstacles in the electronic map, including:

[0151] Use the pre-deployed obstacle recognition model to identify the operation image to determine the obstacles in the operation process;

[0152] When an obstacle is identified, the coordinates of the obstacle are determined based on the coordinate point of the current floor machine using the binocular vision positioning method;

[0153] The obstacle is marked in the electronic map according to the coordinates of the obstacle.

[0154] In an embodiment of the present invention, when the floor machine collides with an obstacle during operation along the target operation route, the target operation route is revised, including:

[0155] When the floor tractor collides with an obstacle while operating along the target operating route, the obstacle is partially bypassed along the left or right side of the straight route and the target operating route is revised.

[0156] Optionally, if the edge coordinates of the obstacle are complete, after bypassing the obstacle, return to the target operation route to continue working, otherwise continue to move forward along the bypassed route, and then repeat the part that should be walked (that is, the target operation route after the obstacle), and then return to the target operation route to work. It is also possible to walk directly to the target operation route to work, and finally return to the unworked part to work, so as to achieve full leveling.

[0157] It is worth noting that the algorithm for bypassing obstacles can be adjusted using existing algorithms, or obstacles can be measured in real time through binocular vision when bypassing, and after bypassing the obstacles, the original route can be returned to work.

[0158] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for dynamic planning of floor machine routes, characterized in that: include: At the initial moment, an electronic map of the area to be operated is obtained, and a target operation route is planned based on the electronic map; Based on the target operation route, the floor machine is controlled to operate, and operation images are collected during the operation to obtain operation images; Use the pre-deployed obstacle recognition model to identify the operation image to determine the obstacles in the operation process and mark the obstacles in the electronic map; When the floor machine collides with an obstacle while operating along the target operating route, the target operating route is revised and the actual route of the floor machine is recorded at the same time; When a deviation occurs between the actual travel route and the target operation route, the floor machine is controlled to return to the target operation route to continue operation, thereby realizing dynamic planning of the floor machine route; The pre-deployment method of the obstacle recognition model includes: The obstacle recognition model is constructed using the YOLO v3 algorithm, and the hyperparameters of the obstacle recognition model are randomly generated. The hyperparameters are encoded into vectors to obtain a hyperparameter vector, and multiple different hyperparameter vectors are repeatedly obtained. Obtain the loss function value corresponding to each hyperparameter vector, and determine the hyperparameter vector with the smallest loss function value as the optimal hyperparameter vector; According to the optimal hyperparameter vector, an adaptive local area search is performed on the hyperparameter vector using a population information full interaction strategy to obtain a hyperparameter vector after the adaptive local area search; For the hyperparameter vector after the adaptive local area search, the individual information interaction strategy is used to perform a local area expansion search on the hyperparameter vector to obtain the hyperparameter vector after the local area expansion search; For the hyperparameter vector after the local area expansion search, according to the optimal hyperparameter vector, the optimal direction search is performed on the hyperparameter vector using the optimal position guidance strategy to obtain the hyperparameter vector after the optimal direction search; For the hyperparameter vector after the optimal direction search, a greedy distributed estimation strategy is used to perform a global region search on the hyperparameter vector to obtain the hyperparameter vector after the global region search; Determine whether the training end condition is met. If so, deploy the obstacle recognition model according to the hyperparameter vector after the global area search, otherwise return to the adaptive local area search step; According to the optimal hyperparameter vector, the population information full interaction strategy is used to perform an adaptive local area search on the hyperparameter vector, and the hyperparameter vector after the adaptive local area search is obtained, including: For the i A hyperparameter vector is generated to determine the mutual influence strength and Euclidean distance between the hyperparameter vector and other hyperparameter vectors; For the i a hyperparameter vector, and determining the influence vector of other hyperparameter vectors on the hyperparameter vector according to the mutual influence strength and the Euclidean distance; Based on the current number of training times, determine the population information full interaction coefficient; For the i A hyperparameter vector is generated, and the first search vector corresponding to the hyperparameter vector is determined by incorporating the boundary influence according to the influence vector of other hyperparameter vectors on the hyperparameter vector and the full interaction coefficient of population information; According to the first search vector corresponding to each hyperparameter vector, all hyperparameter vectors are searched around the optimal hyperparameter vector to obtain the hyperparameter vector after the adaptive local area search; For the hyperparameter vector after the adaptive local area search, the individual information interaction strategy is used to perform a local area expansion search on the hyperparameter vector to obtain the hyperparameter vector after the local area expansion search, including: For the hyperparameter vector after the nth adaptive local area search, randomly match another hyperparameter vector to the hyperparameter vector to obtain the information interaction vector corresponding to each hyperparameter vector; Determine a second search vector corresponding to each hyperparameter vector according to the information interaction vector corresponding to each hyperparameter vector, and perturb the second search vector using a cosine function combined with a time function to obtain a perturbed second search vector; Performing a local area expansion search on the hyperparameter vector according to the second search vector after the disturbance, and obtaining a hyperparameter vector after the local area expansion search; For the hyperparameter vector after the local area expansion search, according to the optimal hyperparameter vector, the optimal direction search is performed on the hyperparameter vector using the optimal position guidance strategy to obtain the hyperparameter vector after the optimal direction search, including: Determine a nonlinear convergence factor for the hyperparameter vector after the m-th local region expansion search; According to the optimal hyperparameter vector, a third search vector corresponding to the hyperparameter vector is determined by a random weighting method, and the third search vector is perturbed by a nonlinear convergence factor to obtain a perturbed third search vector; According to the third search vector after the disturbance, the hyperparameter vector is searched in the optimal direction to obtain the hyperparameter vector after the optimal direction search; For the hyperparameter vector after the optimal direction search, a greedy distributed estimation strategy is used to perform a global region search on the hyperparameter vector to obtain the hyperparameter vector after the global region search, including: For the hyperparameter vector after the optimal direction search, the hyperparameter vector is arranged in descending order according to the loss function value to obtain the arranged hyperparameter vector; According to the arranged hyperparameter vectors, the sequence number of each hyperparameter vector is determined, and the weight influence factor of each hyperparameter vector is determined according to the sequence number of each hyperparameter vector; The hyperparameter vectors are weighted according to the weight influence factor of each hyperparameter vector to determine the population comprehensive position vector; Determine a global search factor corresponding to each hyperparameter vector according to the population comprehensive position vector, and perform a global region search on the hyperparameter vector according to the population comprehensive position vector and the global search factor to obtain a hyperparameter vector after the global region search; Determine whether the loss function value corresponding to the hyperparameter vector after the global area search is reduced. If so, accept the global area search; otherwise, reject the global area search.

2. The method for dynamic planning of floor machine routes according to claim 1, characterized in that: Planning a target operation route based on the electronic map includes: The electronic map of the area to be operated is gridded, and the straight edge closest to the edge is selected as the initial starting direction. All grids are connected in a serpentine shape to obtain the target operation route.

3. The method for dynamic planning of floor machine routes according to claim 1, characterized in that: Based on the target operation route, the process of controlling the floor machine to operate also includes: taking the starting position of the floor machine as the origin and the initial starting direction as the positive X-axis coordinate direction, updating the coordinate points of the travel position in real time, and determining the coordinate points of the actual travel route.

4. The method for dynamic planning of floor machine routes according to claim 3 is characterized in that: Use the pre-deployed obstacle recognition model to identify the operation image to determine the obstacles in the operation process and mark the obstacles in the electronic map, including: Use the pre-deployed obstacle recognition model to identify the operation image to determine the obstacles in the operation process; When an obstacle is identified, the coordinates of the obstacle are determined based on the coordinate point of the current floor machine using the binocular vision positioning method; The obstacle is marked in the electronic map according to the coordinates of the obstacle.

5. The method for dynamic planning of floor machine routes according to claim 4, characterized in that: When the tractor collides with an obstacle while operating along the target operating route, the target operating route shall be revised, including: When the floor tractor collides with an obstacle while operating along the target operating route, the obstacle is partially bypassed along the left or right side of the straight route and the target operating route is revised.

Images