Disinfection robot path planning method considering multi-objective optimization

Abstract

The present invention discloses a path planning method for a disinfection robot considering multi-objective optimization, including obtaining a Voronoi diagram V m The improvement point set S, according to the Voronoi diagram V m and the shape of the disinfection robot, and the configuration space is used to represent the feasible sub-area S of the disinfection robot in the area to be planned. O According to the disinfection target, the path s of the disinfection robot is planned, and the path cost function C1(s), the angle cost function C2(s), the energy consumption cost function C3(s), and the path coverage cost function C4(s) are summed or weighted to obtain the total cost function C(x). A learning neural network is further constructed to obtain the path with the minimum total cost function C(x). The present invention solves the problem of how to design a multi-objective path planning method for a disinfection robot; while considering the total travel distance, total rotation angle, total energy consumption, and the overlap rate of the disinfection area, the multi-objective path planning of the disinfection robot is completed, making the multi-objective path planning process of the disinfection robot fast and accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of robot path planning and motion control, and in particular to a disinfection robot path planning method considering multi-objective optimization. Background Art

[0002] Microorganisms harmful to the human body are easily accumulated in environments such as hospitals, infectious disease concentration areas, and hotels. The environment must be disinfected according to regulatory guidelines. The use of disinfection robots can better achieve disinfection operations. In reality, there are many areas with multiple disinfection targets (such as large hospitals, infectious disease outbreak areas, etc.). The use of disinfection robots for disinfection can minimize the risk to humans. In order to survey and disinfect areas with multiple disinfection targets, the disinfection robot must have the ability of multi-target path planning, so that the disinfection robot can plan a path covering all disinfection targets in the area.

[0003] In the existing technology, most multi-objective path planning methods assume that the robot has enough energy to cover the entire area for repeated iterative operation; however, parameters such as path length, rotation angle, energy consumption and path coverage during operation are all issues that must be considered when the disinfection robot performs multi-objective path planning.

[0004] In summary, how to design a reasonable multi-objective path planning method for the disinfection robot while considering parameters such as path length, rotation angle, energy consumption, and path coverage during operation has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide a disinfection robot path planning method considering multi-objective optimization, which can complete the multi-objective path planning of the disinfection robot while considering the total driving distance, total rotation angle, total energy consumption and disinfection area overlap rate of the disinfection robot.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a disinfection robot path planning method considering multi-objective optimization, which comprises the following steps:

[0007] S1. Process the map of the area to be planned into an improved Voronoi map V m , to form several non-overlapping polygonal sub-regions and obtain the Voronoi diagram V m The improvement point set S;

[0008] S2. According to the improved Voronoi diagram V m and the shape of the disinfection robot, and the configuration space is used to represent the feasible sub-area S of the disinfection robot in the area to be planned. O, and use the configuration space to represent the obstacle sub-area S in the area to be planned i ;

[0009] S3. According to the multiple disinfection targets of the disinfection robot, the path s of the disinfection robot is planned, and the path cost function C1(s), angle cost function C2(s), energy consumption cost function C3(s) and path coverage cost function C4(s) of the disinfection robot are constructed respectively, and the path cost function C1(s), the angle cost function C2(s), the energy consumption cost function C3(s) and the path coverage cost function C4(s) are summed or weighted to obtain the total cost function C(x), and the Voronoi diagram V is used. m The improved point set S and the feasible subregion S O To constrain the total cost function C(x);

[0010] Among them, the path cost function C1(s) is about the total travel distance of the disinfection robot in the disinfection target and between each disinfection target, the angle cost function C2(s) is about the total rotation angle of the disinfection robot in the disinfection target and between each disinfection target, the energy consumption cost function C3(s) is about the total energy consumption of the disinfection robot, and the path coverage cost function C4(s) is about the disinfection area overlap rate of the disinfection robot;

[0011] S4. Structuring the total cost function C(x) and further constructing a learning neural network to dynamically obtain a path with the minimum total cost function C(x).

[0012] In the above technical solution, step S1 specifically includes:

[0013] S1.1. Generate a number of random points P on the map of the area to be planned i ;

[0014] S1.2, with the random point P i As the initial point set S0, the improved Voronoi diagram V is generated m , to form several non-overlapping polygonal sub-regions;

[0015] S1.3. Assign weights w to all the polygonal sub-regions i , and using the weight w i Improve the initial point set S0 to obtain the Voronoi diagram V m The improved point set S={w1p1,w2p2,...,w n p n}.

[0016] In the above technical solution, the feasible sub-region S in step S2 O , specifically:

[0017]

[0018] In the configuration space, p r is the mass point of the disinfection robot, S R For all points p r The set composed of S C is the point set of the entire area to be planned, S W is the point set of the working space of the disinfection robot in the area to be planned, and Sc=S i +S W .

[0019] In the above technical solution, the path cost function C1(s) in step S3 is specifically:

[0020]

[0021] Among them, l i is the travel distance of the disinfection robot from the end point of the previous disinfection target to the starting point of the current disinfection target, d j is the travel distance of the disinfection robot in the jth disinfection target, q+1 is the total number of disinfection targets, and q is l i The total number of segments.

[0022] In the above technical solution, the angle cost function C2(s) in step S3 is specifically:

[0023]

[0024] Among them, (Q i ,Q i+1 ) is the sequence point set of path s, θ(Q i ,Q i+1 ) is the disinfection robot in Q i With Q i+1 The rotation angle between these two points, h, is the total number of rotations of the disinfection robot.

[0025] In the above technical solution, the energy consumption cost function C3(s) in step S3 is specifically:

[0026]

[0027] Among them, E1 is the energy consumption coefficient of the disinfection robot during the driving process from the end point of the previous disinfection target to the starting point of the current disinfection target, E2 is the energy consumption coefficient of the disinfection robot during the driving and disinfection process in the disinfection target, and E3 is the energy consumption coefficient of the disinfection robot during rotation.

[0028] In the above technical solution, the path coverage cost function C4(s) in step S3 is specifically:

[0029]

[0030] Among them, p is the point set of disinfection target, Q i is the point set of the disinfection range of the disinfection robot, and dA is the differential of the disinfection target.

[0031] In the above technical solution, the path coverage cost function C4(s) is simplified by the following steps:

[0032] Sa, define the Voronoi diagram V m The centroid of any polygonal subregion of is point g, then:

[0033]

[0034]

[0035] Among them, g x is the x-axis coordinate of point g, g y is the y-axis coordinate of point g, n a The Voronoi diagram V m The number of vertices in A g is the area of ​​the polygonal subregion where point g is located, and:

[0036]

[0037] Sb, the Voronoi diagram V m For all polygonal sub-regions, the path coverage cost function C4(s) is simplified to:

[0038]

[0039] In the above technical solution, the total cost function C(x) in step S3 is:

[0040] min C(x)=(C1(s),C2(s),C3(s),C4(s)).

[0041] In the above technical solution, step S4 specifically includes:

[0042] S4.1. Iteratively plan the path s of the disinfection robot using a Markov decision process;

[0043] In the Markov decision process, there is a reward function r t =c i ×r i +b s ×r c -μ i ×c u ;

[0044] Among them, c i The Voronoi diagram V of the disinfection robot m The reward weight of the polygonal sub-region, r i is the reward value of the polygonal sub-region; b s is the reward weight at the end of the Markov decision process, r c is the reward value at the end of the Markov decision process, μ i The Voronoi diagram V of the disinfection robot m The penalty weight of the polygonal sub-region, c i is the penalty value of the polygonal sub-region;

[0045] S4.2. Construct a learning neural network and use the learning neural network to train the disinfection robot's path s iterative planning process, using an experience buffer to record the disinfection robot's state, actions, and rewards during training;

[0046] The activation function of the learning neural network at point p is:

[0047]

[0048] Among them, p x is the x-axis coordinate of point p, p y is the y-axis coordinate of point p, and α is the learning parameter;

[0049] S4.3. During the training of the disinfection robot, each node of the learning neural network uses a marginal value function to fit the policy gradient;

[0050] Moreover, the marginal value function G at time t is r About parameter τ, function Status t and the response ψ t The relationship is:

[0051]

[0052] in, To adjust the expected coefficient, w t is the importance weight coefficient, Q τ (s t |ψ t ) is a retrospective evaluation value, and

[0053] Among them, η t is the reward value in the Markov decision process, λ t is the reward coefficient, κ is the weight factor; w t+1 is the dynamic adjustment weight at time t+1, Q(s t+1 |ψ t+1 ) is the estimated value at time t+1, Q τ (s t+1 |ψ t+1 ) is the retrospective evaluation value at time t+1;

[0054] S4.4, use the mean square method to calculate the marginal value function G r Optimize to convert the marginal value function G r The weights are adjusted to the following relationship:

[0055] M t =εM t-1 +(1-ε)|g t-1 |;

[0056]

[0057] w t+1 =w t +Δw;

[0058] Among them, M t is the marginal value function G r The exponential mean of g t-1 is the marginal value function G r The gradient vector at time t-1, ε and ρ are both adjustment parameters, ξ is the learning rate of the learning neural network; w t+1 is the dynamic adjustment weight at time t+1, w t is the adjustment weight value at time t, and Δw is the weight adjustment amount at time t.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. This multi-objective optimization disinfection robot path planning method completes the multi-objective path planning of the disinfection robot in the form of a total cost function C(x), taking into account the total driving distance, total rotation angle, total energy consumption and the overlap rate of the disinfection area of ​​the disinfection robot, and according to the improved Voronoi diagram V mand the shape of the disinfection robot to obtain the feasible subregion S O , to constrain the total cost function C(x), making the multi-objective path planning process of the disinfection robot fast and accurate.

[0061] 2. This disinfection robot path planning method considering multi-objective optimization, the simplified path coverage cost function C4(s), can be used for the Voronoi diagram V m All polygonal sub-areas are used to represent the overlap rate of the disinfection area of ​​the disinfection robot, which reduces the amount of calculation when calculating the minimum value of the total cost function C(x).

[0062] 3. This disinfection robot path planning method considering multi-objective optimization, the learning neural network automatically trains the disinfection robot through the Markov decision process, and gradually approaches the maximum value of the reward function through a finite number of Markov decision process iterations and training, and through the marginal value function G r and its optimization method, which can further approach the maximum value of the reward function and reduce the number of iterations and training, so that the performance of the disinfection robot can be rapidly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flowchart of the steps of embodiment 1 of the present invention.

[0064] Figure 2 This is a flowchart of step S1 in embodiment 1 of the present invention.

[0065] Figure 3 The Voronoi diagram V constructed in step S1 of the first embodiment of the present invention is m Schematic view of .

[0066] Figure 4 This is a schematic diagram of the operation of the disinfection robot of the present invention within the disinfection target and between various disinfection targets.

[0067] Figure 5 Schematic diagram of the disinfection range of the disinfection robot of the present invention within the disinfection target.

[0068] Figure 6 This is a flowchart of the steps of embodiment 2 of the present invention.

[0069] Figure 7 This is a flowchart of the steps of embodiment 3 of the present invention. DETAILED DESCRIPTION

[0070] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0071] Example 1:

[0072] See also Figure 1 This embodiment provides a disinfection robot path planning method considering multi-objective optimization, which includes the following steps:

[0073] S1. Process the map of the area to be planned into an improved Voronoi map V m , to form several non-overlapping polygonal sub-regions and obtain the Voronoi diagram V m The improvement point set S.

[0074] Using the improved Voronoi diagram V m The improved point set S provides quantifiable and computable discrete map data for the path planning of the disinfection robot, which simplifies the path planning process of the disinfection robot.

[0075] S2. According to the improved Voronoi diagram V m and the shape of the disinfection robot, and the configuration space is used to represent the feasible sub-area S of the disinfection robot in the area to be planned. O , and use the configuration space to represent the obstacle sub-area S in the area to be planned i .

[0076] Feasible subregion S O That is the free space of the disinfection robot, in the feasible sub-area S O In this way, the disinfection robot does not intersect with any obstacle; the feasible subregion S represented in this way O In the process, the shape influence of the disinfection robot can be ignored and the disinfection robot can be regarded as a mass point, which simplifies the path planning process of the disinfection robot. In the subsequent path planning process, the point set of the planned path is from the feasible subregion S O Obtained from.

[0077] S3. According to the multiple disinfection targets of the disinfection robot, the path s of the disinfection robot is planned, and the path cost function C1(s), angle cost function C2(s), energy consumption cost function C3(s) and path coverage cost function C4(s) of the disinfection robot are constructed respectively. The path cost function C1(s), angle cost function C2(s), energy consumption cost function C3(s) and path coverage cost function C4(s) are summed or weighted to obtain the total cost function C(x), and the Voronoi diagram V is used.m Improved point set S and feasible subregion S O To constrain the total cost function C(x).

[0078] Among them, the path cost function C1(s) is about the total driving distance of the disinfection robot in the disinfection target and between each disinfection target, the angle cost function C2(s) is about the total rotation angle of the disinfection robot in the disinfection target and between each disinfection target, the energy consumption cost function C3(s) is about the total energy consumption of the disinfection robot, and the path coverage cost function C4(s) is about the overlap rate of the disinfection area of ​​the disinfection robot.

[0079] In this step, the disinfection targets refer to a number of block areas that need to be disinfected and are distributed in the area to be planned. The disinfection targets are adjacent or separated in the area to be planned, but do not overlap.

[0080] According to the total cost function C(x), the minimum value of the total cost function C(x) is calculated to obtain the conditions, and the optimal multi-objective planning path can be achieved.

[0081] S4. Structuring the total cost function C(x) and further constructing a learning neural network to dynamically obtain the path with the minimum total cost function C(x).

[0082] See also Figure 2 Specifically, step S1 includes:

[0083] S1.1. Generate a number of random points P on the map of the area to be planned i .

[0084] In fact, the random point P i The generation of has nothing to do with the parameters of the area to be planned, the parameters of the disinfection robot itself, and the external parameters.

[0085] S1.2, with random point P i As the initial point set S0, the improved Voronoi diagram V is generated m , to form several non-overlapping polygonal sub-regions.

[0086] S1.3. Assign weights w to all polygonal sub-regions i , and use the weight w i Improve the initial point set S0 to obtain the Voronoi diagram V m The improved point set S={w1p1,w2p2,...,w n p n}.

[0087] See also Figure 3 , the Voronoi diagram V generated in this way m It has the following properties:

[0088]

[0089] Among them, p m and p n are any points in the improved point set S, and p is the Voronoi diagram V m For any point on the Voronoi diagram V, this property shows that for every m The polygonal subregion of the boundary, in the Voronoi diagram V m The vertices outside the boundary will be replaced by the Voronoi diagram V m The boundaries or vertices of the area to be planned can be correctly represented.

[0090] Specifically, the feasible sub-region S in step S2 O , specifically:

[0091]

[0092] In the configuration space, p r is the mass point of the disinfection robot, S R For all points p r The set composed of S C is the point set of the entire area to be planned, S W is the point set of the disinfection robot's workspace in the area to be planned, and Sc=S i +S W .

[0093] In the point set S C In the above example, all the particles p of the disinfection robot are considered and offset. r The set S R Then, in the feasible subregion S O In the proposed method, the shape influence of the disinfection robot can be ignored and the disinfection robot can be regarded as a point mass, which simplifies the path planning process of the disinfection robot.

[0094] Specifically, the path cost function C1(s) in step S3 is:

[0095]

[0096] See also Figure 4 , where l i is the travel distance of the disinfection robot from the end point of the previous disinfection target to the starting point of the current disinfection target, i represents the travel distance of the i-th segment; d j is the travel distance of the disinfection robot in the jth disinfection target, q+1 is the total number of disinfection targets, and q is l i The total number of segments.

[0097] The path cost function C1(s) can accurately represent the total driving distance of the disinfection robot in the disinfection target and between each disinfection target in the form of a discrete mathematical model, thereby serving as one of the parameters of the total cost function C(x).

[0098] Specifically, the angle cost function C2(s) in step S3 is:

[0099]

[0100] Among them, (Q i ,Q i+1 ) is the sequence point set of path s, θ(Q i ,Q i+1 ) for disinfection robots in Q i With Q i+1 The rotation angle between these two points, h is the total number of rotations of the disinfection robot.

[0101] The angle cost function C2(s) can accurately represent the total rotation angle of the disinfection robot in the disinfection target and between each disinfection target in the form of a discrete mathematical model, thereby serving as one of the parameters of the total cost function C(x).

[0102] Specifically, the energy consumption cost function C3(s) in step S3 is:

[0103]

[0104] Among them, E1 is the energy consumption coefficient of the disinfection robot during the driving process from the end point of the previous disinfection target to the starting point of the current disinfection target, E2 is the energy consumption coefficient of the disinfection robot during the driving and disinfection process in the disinfection target, and E3 is the energy consumption coefficient of the disinfection robot during rotation.

[0105] The energy consumption cost function C3(s) can accurately represent the total energy consumption of the disinfection robot in the form of a discrete mathematical model, and thus serve as one of the parameters of the total cost function C(x).

[0106] Specifically, the path coverage cost function C4(s) in step S3 is:

[0107]

[0108] Among them, p is the point set of disinfection target, Q i is the point set of the disinfection range of the disinfection robot, and dA is the differential of the disinfection target.

[0109] See also Figure 5The disinfection range of the disinfection robot refers to the effective range of the disinfection robot that can disinfect the environment when it is at a certain point in space. This is related to the structural design of the disinfection robot's motion mechanism and disinfection mechanism; during the actual disinfection process, the disinfection robot will travel back and forth in the disinfection target. During the back-and-forth driving process, the disinfection range of the disinfection robot will overlap. The purpose of designing the path coverage cost function C4(s) is to minimize the overlapping area of ​​the disinfection range during the disinfection process of the disinfection robot; the path coverage cost function C4(s) can accurately represent the disinfection area overlap rate of the disinfection robot in the form of a discrete mathematical model, thereby serving as one of the parameters of the total cost function C(x).

[0110] In summary, the total cost function C(x) in step S3 is:

[0111] minC(x)=(C1(s),C2(s),C3(s),C4(s)).

[0112] According to the total cost function C(x), the minimum value of the total cost function C(x) is calculated to obtain the conditions, and the optimal multi-objective planning path can be achieved.

[0113] This method of disinfection robot path planning considering multi-objective optimization completes the multi-objective path planning of the disinfection robot in the form of a total cost function C(x), taking into account the total driving distance, total rotation angle, total energy consumption and overlap rate of the disinfection area of ​​the disinfection robot, and according to the improved Voronoi diagram V m and the shape of the disinfection robot to obtain the feasible subregion S O , to constrain the total cost function C(x), making the multi-objective path planning process of the disinfection robot fast and accurate.

[0114] Example 2:

[0115] See also Figure 6 This embodiment provides a disinfection robot path planning method considering multi-objective optimization. Based on the disinfection robot path planning method considering multi-objective optimization provided in the first embodiment, it further includes:

[0116] The path coverage cost function C4(s) is simplified by the following steps:

[0117] Sa, define the Voronoi diagram V m The centroid of any polygonal subregion of is point g, then:

[0118]

[0119]

[0120] Among them, g xis the x-axis coordinate of point g, g y is the y-axis coordinate of point g, n a Voronoi diagram V m The number of vertices in A g is the area of ​​the polygonal subregion where point g is located, and:

[0121]

[0122] In the Voronoi diagram V m In any polygonal sub-region of , the point where the path coverage cost function C4(s) is minimized is point g. Therefore, it is necessary to express the x-axis and y-axis coordinates of point g and calculate the Voronoi diagram V. m Area A g .

[0123] Sb, comprehensive Voronoi diagram V m For all polygonal sub-regions of , the path coverage cost function C4(s) is simplified to:

[0124]

[0125] This disinfection robot path planning method considering multi-objective optimization, the simplified path coverage cost function C4(s), can be used for the Voronoi diagram V m All polygonal sub-areas are used to represent the overlap rate of the disinfection area of ​​the disinfection robot, which reduces the amount of calculation when calculating the minimum value of the total cost function C(x).

[0126] Example 3:

[0127] See also Figure 7 This embodiment provides a disinfection robot path planning method considering multi-objective optimization. Based on the disinfection robot path planning method considering multi-objective optimization provided in the first embodiment, it further includes:

[0128] Step S4 specifically includes:

[0129] S4.1. Use the Markov decision process to iteratively plan the path s of the disinfection robot.

[0130] In the Markov decision process, there is a reward function r t =c i ×r i +b s ×r c -μ i ×c u ;

[0131] Among them, c i Voronoi diagram for the disinfection robot mThe reward weight of the triangular area, r i is the reward value of the triangle area; b s is the reward weight at the end of the Markov decision process, r c is the reward value at the end of the Markov decision process, μ i Voronoi diagram for the disinfection robot m The penalty weight of the triangular area, c i is the penalty value of the triangle area;

[0132] S4.2. Construct a learning neural network and use it to train the disinfection robot's path s iterative planning process, using an experience buffer to record the disinfection robot's state, actions, and rewards during training;

[0133] In a Markov decision process, we have:

[0134] The activation function of the learning neural network at point p is:

[0135]

[0136] Among them, p x is the x-axis coordinate of point p, p y is the y-axis coordinate of point p, and α is the learning parameter;

[0137] S4.3. During the training of the disinfection robot, each node of the learning neural network uses the marginal value function to fit the policy gradient;

[0138] Moreover, the marginal value function G at time t is r About parameter τ, function Status t and the response ψ t The relationship is:

[0139]

[0140] in, To adjust the expected coefficient, w t is the importance weight coefficient, Q τ (s t |ψ t ) is a retrospective evaluation value, and

[0141] Among them, η t is the reward value in the Markov decision process, and the initial value is 1; t is the reward coefficient, the initial value is selected as 0.9; κ is the weight factor, the initial value is selected as 0.8; w t+1 is the dynamic adjustment weight at time t+1, Q(st+1 |ψ t+1 ) is the estimated value at time t+1, Q τ (s t+1 |ψ t+1 ) is the retrospective evaluation value at time t+1;

[0142] S4.4, use the mean square method to calculate the marginal value function G r Optimize to transform the marginal value function G r The weights are adjusted to the following relationship:

[0143] M t =εM t-1 +(1-ε)|g t-1 |;

[0144]

[0145] w t+1 =w t +Δw;

[0146] Among them, M t is the marginal value function G r The exponential mean of g t-1 is the marginal value function G r The gradient vector at time t-1, ε and ρ are both adjustment parameters, ξ is the learning rate of the learning neural network, and the initial value is selected as 0.8; w t+1 is the dynamic adjustment weight at time t+1, w t is the adjustment weight value at time t, and Δw is the weight adjustment amount at time t.

[0147] This disinfection robot path planning method considering multi-objective optimization, the learning neural network automatically trains the disinfection robot through the Markov decision process, and gradually approaches the maximum value of the reward function through a finite number of Markov decision process iterations and training, and through the marginal value function G r and its optimization method, which can further approach the maximum value of the reward function and reduce the number of iterations and training, so that the performance of the disinfection robot can be rapidly improved.

[0148] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A disinfection robot path planning method considering multi-objective optimization, characterized in that: The steps include: S1. Process the map of the area to be planned into an improved Voronoi map V m , to form several non-overlapping polygonal sub-regions and obtain the Voronoi diagram V m The improvement point set S; S2. According to the improved Voronoi diagram V m and the shape of the disinfection robot, and the configuration space is used to represent the feasible sub-area S of the disinfection robot in the area to be planned. O , and use the configuration space to represent the obstacle sub-area S in the area to be planned i ; S3. According to the multiple disinfection targets of the disinfection robot, the path s of the disinfection robot is planned, and the path cost function C1(s), angle cost function C2(s), energy consumption cost function C3(s) and path coverage cost function C4(s) of the disinfection robot are constructed respectively, and the path cost function C1(s), the angle cost function C2(s), the energy consumption cost function C3(s) and the path coverage cost function C4(s) are summed or weighted to obtain the total cost function C(x), and the Voronoi diagram V is used. m The improved point set S and the feasible subregion S O To constrain the total cost function C(x); the path cost function C1(s), specifically: Among them, l i is the travel distance of the disinfection robot from the end point of the previous disinfection target to the starting point of the current disinfection target, d j is the travel distance of the disinfection robot in the jth disinfection target, q+1 is the total number of disinfection targets, and q is l i The total number of segments; The angle cost function C2(s) is specifically: Among them, (Q i ,Q i+1 ) is the sequence point set of path s, θ(Q i ,Q i+1 ) is the disinfection robot in Q i With Q i+1 The rotation angle between these two points, h is the total number of rotations of the disinfection robot; The energy consumption cost function C3(s) is specifically: Among them, E1 is the energy consumption coefficient of the disinfection robot during the driving process from the end point of the previous disinfection target to the starting point of the current disinfection target, E2 is the energy consumption coefficient of the disinfection robot during the driving and disinfection process in the disinfection target, and E3 is the energy consumption coefficient of the disinfection robot during rotation; The path coverage cost function C4(s) is simplified to: Among them, p is the point set of disinfection target, Q i is the point set of the disinfection range of the disinfection robot, dA is the differential of the disinfection target; Among them, the path cost function C1(s) is about the total travel distance of the disinfection robot in the disinfection target and between each disinfection target, the angle cost function C2(s) is about the total rotation angle of the disinfection robot in the disinfection target and between each disinfection target, the energy consumption cost function C3(s) is about the total energy consumption of the disinfection robot, and the path coverage cost function C4(s) is about the disinfection area overlap rate of the disinfection robot; S4. Structuring the total cost function C(x) and further constructing a learning neural network to dynamically obtain a path with the minimum total cost function C(x).

2. The disinfection robot path planning method considering multi-objective optimization according to claim 1 is characterized in that: The step S1 specifically includes: S1.

1. Generate a number of random points P on the map of the area to be planned i ; S1.2, with the random point P i As the initial point set S0, the improved Voronoi diagram V is generated m , to form several non-overlapping polygonal sub-regions; S1.

3. Assign weights w to all the polygonal sub-regions i , and using the weight w i Improve the initial point set S0 to obtain the Voronoi diagram V m The improved point set S={w1P1,w2P2,...,w n P n }.

3. The disinfection robot path planning method considering multi-objective optimization according to claim 2 is characterized in that: The feasible sub-region S in step S2 O , specifically: In the configuration space, p r is the mass point of the disinfection robot, S R For all points p r The set composed of S C is the point set of the entire area to be planned, S W is the point set of the working space of the disinfection robot in the area to be planned, and Sc=S i +S W .

4. The disinfection robot path planning method considering multi-objective optimization according to claim 1 is characterized in that: The total cost function C(x) in step S3 is: minC(x)=(C1(s),C2(s),C3(s),C4(s)).

5. The disinfection robot path planning method considering multi-objective optimization according to claim 1 is characterized in that: The step S4 specifically includes: S4.

1. Iteratively plan the path s of the disinfection robot using a Markov decision process; In the Markov decision process, there is a reward function r t =c i ×r i +b s ×r c -μ i ×c u ; Among them, c i The Voronoi diagram V of the disinfection robot m The reward weight of the polygonal sub-region, r i is the reward value of the polygonal sub-region; b s is the reward weight at the end of the Markov decision process, r c is the reward value at the end of the Markov decision process, μ i The Voronoi diagram V of the disinfection robot m The penalty weight of the polygonal sub-region, c u is the penalty value of the polygonal sub-region; S4.

2. Construct a learning neural network and use the learning neural network to train the disinfection robot's path s iterative planning process, using an experience buffer to record the disinfection robot's state, actions, and rewards during training; The activation function of the learning neural network at point p is: Among them, p x is the x-axis coordinate of point p, p y is the y-axis coordinate of point p, and α is the learning parameter; S4.

3. During the training of the disinfection robot, each node of the learning neural network uses a marginal value function to fit the policy gradient; Moreover, the marginal value function G at time t is r About parameter τ, function Status t and the response ψ t The relationship is: in, To adjust the expected coefficient, w t is the importance weight coefficient, Q τ (s t |ψ t ) is a retrospective evaluation value, and Among them, η t is the reward value in the Markov decision process, λ t is the reward coefficient, κ is the weight factor; w t+1 is the dynamic adjustment weight at time t+1, Q(s t+1 |ψ t+1 ) is the estimated value at time t+1, Q τ (s t+1 |ψ t+1 ) is the retrospective evaluation value at time t+1; S4.4, use the mean square method to calculate the marginal value function G r Optimize to convert the marginal value function G r The weights are adjusted to the following relationship: M t =εM t-1 +(1-e)|g t-1 |; In t+1 =in t +Δw; Among them, M t is the marginal value function G r The exponential mean of g t-1 is the marginal value function G r The gradient vector at time t-1, ε and ρ are both adjustment parameters, ξ is the learning rate of the learning neural network; w t+1 is the dynamic adjustment weight at time t+1, w t is the adjustment weight value at time t, and Δw is the weight adjustment amount at time t.

Images