Robot Source Search Method in Multi-Gas Diffusion Source Scenarios Based on Information Path Planning

Abstract

The present application relates to a robot source seeking method in a multi-gas diffusion source scenario based on information path planning. The method includes: using a sensing drone to perform particle filter sampling on the gas concentration information of the scenario along a sensing path, constructing an optimization objective function based on free energy according to the sampling results, then optimizing the sensing path according to the optimization objective function and the information path planning algorithm to obtain an optimized sensing path, the drone obtaining concentration sensing information in the scenario along the optimized sensing path, and fitting a multi-source concentration field in the scenario based on a Gaussian process regression model, and finally using a sequential confirmation mechanism to sequentially confirm the positions of gas sources in the multi-source concentration field to obtain the determined positions of all gas sources, completing the robot source seeking task in the multi-gas diffusion source scenario. Using this method can improve the efficiency and success rate of the robot in searching for gas sources in an unknown multi-source scenario, and has the advantage of strong robustness under various source seeking conditions.

Description

Technical Field

[0001] The present application relates to the technical field of robot control, and particularly to a method for a robot to search for a source in a multi-gas diffusion source scenario based on information path planning. Background Art

[0002] The problem of source search is a problem faced by both nature and human society. In this problem, the searcher searches for the location of the source in a scenario where the source continuously releases chemical or other signals through a predetermined source search strategy. With the development of science and technology and human society, more and more fields have begun to study this problem. In the field of public security, nuclear, biological, and chemical leakage accidents often cause huge damage and lead to chaos in human society. In the face of such sudden emergency events, quickly and accurately searching for the leakage source and obtaining target-related information (location, intensity, etc.) is conducive to emergency personnel formulating scientific and reasonable emergency response plans, thereby reducing the harmfulness of leakage events.

[0003] Therefore, how to provide a robot gas source search solution in a multi-gas diffusion source scenario has become an urgent technical problem to be solved. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method for a robot to search for a source in a multi-gas diffusion source scenario based on information path planning.

[0005] A method for a robot to search for a source in a multi-gas diffusion source scenario based on information path planning, the method comprising:

[0006] Performing particle filter sampling on the gas concentration information in the multi-gas diffusion source scenario along the sensing path of the sensing unmanned aerial vehicle, obtaining the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution, calculating according to the spatial entropy and the potential energy to obtain the free energy of the sensing path, and constructing an optimization objective function of the sensing path according to the free energy;

[0007] Performing parameterization processing on the sensing path according to the information path planning algorithm, updating the optimization objective function according to the parameterized sensing path to obtain an updated optimization objective function, calculating the updated optimization objective function, and performing piecewise continuous optimization on the parameterized sensing path according to the calculated objective function value to obtain an optimized sensing path;

[0008] Obtaining concentration sensing information in the multi-gas diffusion source scenario along the optimized sensing path, and fitting the multi-gas diffusion source scenario according to the concentration sensing information and the Gaussian process regression model to obtain a multi-source concentration field;

[0009] Sequential multi-source confirmation mechanism is adopted to sequentially confirm the positions of multi-gas diffusion sources in the multi-source concentration field until the determined positions of all multi-gas diffusion sources are obtained, completing the robot source search task in the multi-gas diffusion source scenario.

[0010] In one embodiment, the free energy of the sensing path is calculated based on spatial entropy and potential energy, including:

[0011] The free energy of the sensing path is calculated based on the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution, expressed as

[0012] ;

[0013] Where represents the number of samplings, represents the spatial entropy, represents the th sampling potential energy, represents the temperature that controls the relative value of potential energy and spatial entropy, represents the trace of the matrix, represents the number of particle filter samples, represents the current position of the sensing UAV, represents the th particle filter sample position, represents the th particle filter sample weight, represents the intensity of the gravitational force of the gas diffusion source on the sensing UAV, represents the weighted covariance matrix of the particle filter sample distribution, represents the scaling factor, represents the exponent that determines the temperature change rate, represents the covariance of the concentration field on the sensing path.

[0014] In one embodiment, the optimization objective function of the sensing path is constructed based on the free energy, including:

[0015] The optimization objective function of the sensing path is constructed based on the free energy, expressed as

[0016] ;

[0017] Where represents the difference in free energy before and after particle filter sampling along the sensing path , and respectively represent the free energy before and after particle filter sampling along the sensing path , represents the path cost constraint on the optimization objective function, Indicates the calculation execution perception path Required cost of the function Indicates the budget

[0018] In one embodiment, the perception path is parameterized according to the information path planning algorithm, and the optimization objective function is updated according to the parameterized perception path to obtain the updated optimization objective function, including:

[0019] The perception path is parameterized into a B-spline curve according to the information path planning algorithm, and the parameterized perception path is constructed by adding boundary constraints to the B-spline curve;

[0020] The optimization objective function is updated according to the parameterized perception path to obtain the updated optimization objective function

[0021] In one embodiment, the perception path is parameterized into a B-spline curve according to the information path planning algorithm, and the parameterized perception path is constructed by adding boundary constraints to the B-spline curve, including:

[0022] The perception path is parameterized according to the information path planning algorithm Parameterized into order with control points of the B-spline curve, and the parameterized perception path is constructed by adding boundary constraints to the B-spline curve; wherein, the boundary constraint means that the starting point of the B-spline curve coincides with the first control point coincides, and the end point of the B-spline curve coincides with the th control point coincides

[0023] In one embodiment, the optimization objective function is updated according to the parameterized perception path to obtain the updated optimization objective function, including:

[0024] The optimization objective function is updated according to the parameterized perception path to obtain the parameterized optimization objective function, expressed as

[0025] ;

[0026] ;

[0027] Wherein, represents the parameterized perception path, represents the path cost constraint received by the parameterized optimization objective function, represents the boundary constraint received by the parameterized optimization objective function, represents the starting point constraint received by the parameterized optimization objective function represents a constant, Represents the concentration field All feasible paths in

[0028] The path cost constraint and boundary constraint in the parameterized optimization objective function are converted into budget multiplication penalty factors and boundary multiplication penalty factors respectively, and the updated optimization objective function is constructed and expressed as

[0029] ;

[0030] in, represents the budget multiplication penalty factor, represents the boundary multiplication penalty factor, represents the updated optimization objective function value, Represents the difference in free energy before and after particle filter sampling along the B-spline curve.

[0031] In one embodiment, the updated optimization objective function is calculated by a covariance evolution algorithm, and the parameterized perception path is segmented and continuously optimized according to the calculated objective function value to obtain the optimized perception path, including:

[0032] The updated optimization objective function is calculated by the covariance evolution algorithm, and the parameterized perception path is continuously optimized in segments according to the calculated objective function value. After each segment of the optimized perception path is obtained and the concentration perception information therein is acquired, the free energy of the segment of the optimized perception path is updated according to the concentration perception information, and a new objective function value is calculated based on the updated free energy. Then, the next segment of the parameterized perception path is optimized according to the new objective function value until the optimization of all segmented paths is completed and a complete optimized perception path is obtained.

[0033] In one embodiment, concentration sensing information in a multi-gas diffusion source scenario is acquired along an optimized sensing path, and the multi-gas diffusion source scenario is fitted based on the concentration sensing information and a Gaussian process regression model to obtain a multi-source concentration field, including:

[0034] The concentration perception information in the multi-gas diffusion source scene is obtained along the optimized perception path, and the multi-gas diffusion source scene is fitted based on the concentration perception information and the Gaussian process regression model to obtain the concentration distribution of the multi-source concentration field, which is expressed as

[0035] ;

[0036] in, represents the mean value function, represents the covariance function, represents a Gaussian process, Indicates a multi-gas diffusion source scenario in the concentration perception information, indicating a multi-source concentration field.

[0037] In one embodiment, a sequential multi-source confirmation mechanism is adopted to sequentially confirm the positions of multi-gas diffusion sources in the multi-source concentration field until the determined positions of all multi-gas diffusion sources are obtained, completing the robot source search task in the multi-gas diffusion source scenario, including:

[0038] Adopting a sequential confirmation mechanism to sequentially confirm the positions of multi-gas diffusion sources in the multi-source concentration field; wherein, the confirmation steps include an expectation step and a maximum step;

[0039] In the expectation step, according to the peak algorithm, mark the concentration peak points in the multi-source concentration field where the concentration is higher than the marking threshold, and take the concentration peak points as the estimated positions of the multi-gas diffusion sources;

[0040] In the maximum step, use a ground robot equipped with a vision sensor to sequentially go to the estimated positions of all multi-gas diffusion sources for position confirmation, obtain the determined positions of the multi-gas diffusion sources, and update the multi-source concentration field according to the determined positions of the multi-gas diffusion sources to obtain an updated multi-source concentration field;

[0041] Sequentially execute the expectation step and the maximum step until there is no concentration region in the updated multi-source concentration field where the concentration is higher than the termination threshold, obtain the determined positions of all multi-gas diffusion sources in the multi-gas diffusion source scenario, and complete the robot source search task in the multi-gas diffusion source scenario.

[0042] In one embodiment, in the maximum step, use a ground robot equipped with a vision sensor to sequentially go to the estimated positions of all multi-gas diffusion sources for position confirmation, obtain the determined positions of the multi-gas diffusion sources, and update the multi-source concentration field according to the determined positions of the multi-gas diffusion sources to obtain an updated multi-source concentration field, including:

[0043] In the maximum step, use a ground robot equipped with a vision sensor to sequentially go to the estimated positions of all multi-gas diffusion sources, and perform position confirmation on the estimated positions of all multi-gas diffusion sources according to the sensor perception information obtained by the ground robot to obtain the determined positions of the multi-gas diffusion sources;

[0044] Perform concentration perception according to the determined positions of the multi-gas diffusion sources by the ground robot, obtain the source term parameters of the multi-gas diffusion sources, update the multi-source concentration field according to the source term parameters, and eliminate the concentration regions corresponding to the source term parameters in the multi-source concentration field to obtain an updated multi-source concentration field.

[0045] The above-mentioned robot source-seeking method in a multi-gas diffusion source scenario based on information path planning first uses a sensing drone to perform particle filter sampling on the gas concentration information of the scenario along the sensing path, constructs an optimization objective function based on free energy for the sensing path according to the sampling results, then optimizes the sensing path according to the optimization objective function and the information path planning algorithm to obtain an optimized sensing path. The drone moves along the optimized sensing path to sense the concentration sensing information in the scenario, and fits the multi-source concentration field in the scenario based on the Gaussian process regression model. Finally, a sequential confirmation mechanism is used to sequentially confirm the positions of the gas sources in the multi-source concentration field, obtain the determined positions of all gas sources, and complete the robot source-seeking task in the multi-gas diffusion source scenario. Using this method can improve the efficiency and success rate of the robot in searching for gas sources in an unknown multi-source scenario, and has the advantage of strong robustness under various source-seeking conditions. Description of the Drawings

[0046] Figure 1 It is a schematic flowchart of a robot source-seeking method in a multi-gas diffusion source scenario based on information path planning in an embodiment;

[0047] Figure 2 It is a schematic diagram of parameterizing the sensing path in an embodiment. Detailed Embodiments

[0048] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] In one embodiment, as Figure 1 shown, a robot source-seeking method in a multi-gas diffusion source scenario based on information path planning is provided, including the following steps:

[0050] Step S1, perform particle filter sampling on the gas concentration information in the multi-gas diffusion source scenario along the sensing path of the sensing drone, obtain the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution, calculate according to the spatial entropy and the potential energy to obtain the free energy of the sensing path, and construct an optimization objective function for the sensing path according to the free energy.

[0051] It can be understood that in the actual source seeking process, since the drone cannot obtain the positions of multiple sources in the scene, the particle filter method can be used to estimate the position where the source is located. And since each sample of the particle filter represents a point estimate of the source position, the position of the sample can be used to replace the source position to represent the potential energy. Further, in order for the robot to obtain more information about the concentration distribution of the multi-source diffusion scene, it is necessary to combine the spatial entropy and the potential energy of the particle filter sample distribution to design an optimization objective function. The spatial entropy is responsible for making the sensing path cover more detection areas, and the potential energy of the particle filter sample distribution makes the sensing path tend to the area with high concentration.

[0052] Step S2: Parametrize the sensing path according to the information path planning algorithm, and update the optimization objective function according to the parametrized sensing path to obtain the updated optimization objective function. Calculate the updated optimization objective function, and perform piecewise continuous optimization on the parametrized sensing path according to the calculated objective function value to obtain the optimized sensing path.

[0053] It can be understood that after designing the optimization objective function, the sensing path needs to be parametrized according to the information path planning algorithm to obtain the parametrized sensing path and the updated optimization objective function. Then, use the covariance evolution algorithm to optimize the sensing path, and based on the continuous optimization strategy, optimize the path segment by segment. Update the spatial entropy and the particle filter after obtaining the sensing information of the previous path segment, and optimize the next path segment based on the new objective function value.

[0054] Step S3: Obtain the concentration sensing information in the multi-gas diffusion source scene along the optimized sensing path, and fit the multi-gas diffusion source scene according to the concentration sensing information and the Gaussian process regression model to obtain the multi-source concentration field.

[0055] It can be understood that since the multi-gas diffusion source scene is a continuous function in space, and the drone can only sample this continuous function at a finite set of positions in space, a regression model is needed to infer the function values at unsampled positions. The Gaussian process regression model provides a probabilistic and non-parametric method to model the multi-source concentration field, thus effectively fitting the concentration distribution under the entire scene.

[0056] Step S4: Adopt a sequential multi-source confirmation mechanism to sequentially confirm the positions of the multi-gas diffusion sources in the multi-source concentration field until the determined positions of all multi-gas diffusion sources are obtained, and complete the source seeking task of the robot in the multi-gas diffusion source scene.

[0057] It can be understood that after the sensing drone executes the optimized sensing path obtained from the information path planning and acquires the Gaussian process fitting results of the multi-source concentration field, the positions of the multi-sources still cannot be confirmed through the concentration distribution map. At the same time, it is impossible to use the drone to process the diffusion sources, such as shutting down and repairing. Therefore, it is necessary to send a ground robot carrying an emergency disposal device and a vision device to the position with the highest possibility of the source in the concentration field to confirm and conduct emergency disposal on the diffusion source. For this purpose, the present invention proposes a sequential multi-source confirmation mechanism to sequentially confirm multiple gas diffusion sources. The sequential multi-source confirmation mechanism is based on the EM (Expectation-Maximization) algorithm framework and divides the confirmation of the gas source into two steps: the expectation step and the maximization step. In the expectation step, the robot determines the possible positions of multiple gas sources in the current multi-source concentration field and goes to confirm them. In the maximization step, the robot senses the relevant information of the successfully confirmed source through the sensor and removes its concentration distribution from the estimated multi-source concentration field (simulating the process of shutting down the diffusion source) to update the expectation information. The expectation step and the maximization step are executed cyclically to complete the confirmation of the multi-source positions and source term information.

[0058] In one embodiment, the free energy of the sensing path is calculated according to the spatial entropy and potential energy, including:

[0059] First, particle filter sampling is performed on the gas concentration information in the multi-gas diffusion source scenario along the sensing path of the sensing drone to obtain the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution.

[0060] Then, calculation is performed according to the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution to obtain the free energy of the sensing path, expressed as

[0061] ;

[0062] Among them, represents the number of samplings, represents the spatial entropy, represents the th potential energy of sampling, represents the temperature that controls the relative value of the potential energy and the spatial entropy, represents the trace of the matrix, The value of will decrease as the particle filter samples change from the initial uniform distribution to the process of aggregating towards the source position; represents the number of particle filter samples, represents the current position of the sensing drone, represents the position of the th particle filter sample, represents the weight of the Represents the intensity of the gas diffusion source on the gravitational force sensed by the drone, Represents the weighted covariance matrix of the particle filter sample distribution, Represents the scaling factor, Represents the exponent determining the rate of temperature change, Represents the covariance of the concentration field on the sensing path.

[0063] It can be understood that since the ratio of potential energy to spatial entropy is regulated by the magnitude of temperature, as the uncertainty of the source position decreases, the proportion of potential energy in the free energy will become larger and larger. In the initial stage of source seeking, a larger proportion of spatial entropy will cause the drone's path to be planned to explore the entire space. As the source seeking process progresses, the drone will go to areas with a higher probability of the source's existence for sampling, which helps to distinguish the positions of multiple sources.

[0064] In one of the embodiments, an optimization objective function for the sensing path is constructed based on the free energy, expressed as

[0065] ;

[0066] where, Represents the difference in free energy before and after particle filter sampling along the sensing path and Represents the free energy before and after particle filter sampling along the sensing path Represents the path cost constraint on the optimization objective function, Represents the function for calculating the cost required to execute the sensing path Represents the budget.

[0067] It can be understood that in the field of multi-source source seeking, more attention needs to be paid to the concentration distribution in the source area, that is, the distribution of areas with higher concentration, which helps to distinguish the positions of multiple sources. Therefore, when constructing the optimization objective function in this application, by introducing free energy, the attention to high-concentration areas is increased, and most of the path budget is consumed in areas with higher concentration, making the drone sampling path more likely to cover the areas where multiple sources are located.

[0068] In one of the embodiments, the sensing path is parameterized according to the information path planning algorithm, and the optimization objective function is updated according to the parameterized sensing path to obtain the updated optimization objective function, including:

[0069] First, the sensing path is parameterized into a B-spline curve according to the information path planning algorithm, and the parameterized sensing path is constructed by adding boundary constraints to the B-spline curve. ​​​

[0070] Specifically, as Figure 2 shown, the sensing path is parameterized into a -order B-spline curve with control points through the information path planning algorithm. The parameterized sensing path is constructed by adding boundary constraints to the B-spline curve. Among them, as can be seen, the boundary constraint means that the starting point of the B-spline curve coincides with the first control point Figure 2 , and the ending point of the B-spline curve coincides with the th control point ; the parameterized sensing path is defined by a second-order B-spline curve defined by to five control points, and the length of the path is ; the positions of to six sampling points on the sensing path are determined by the sampling frequency .

[0071] Then, the optimization objective function is updated according to the parameterized sensing path to obtain the parameterized optimization objective function, denoted as

[0072] ;

[0073] ;

[0074] Among them, represents the parameterized sensing path, represents the path cost constraint on the parameterized optimization objective function, represents the boundary constraint on the parameterized optimization objective function, represents the starting point constraint on the parameterized optimization objective function, represents a constant, represents the concentration field all feasible paths in. It can be understood that in addition to the initial path cost constraint, boundary constraints need to be added to prevent the planned path from exceeding the range of the environment to be measured. At the same time, the starting point constraint is also necessary, and this constraint can be achieved by omitting the starting point in the vector form of the control points, that is , where represents the vector form.

[0075] Finally, the path cost constraint and boundary constraint in the parameterized optimization objective function are respectively converted into a budget multiplication penalty factor and a boundary multiplication penalty factor, and the updated optimization objective function is constructed, denoted as

[0076] ;

[0077] Among them, represents the budget multiplication penalty factor, represents the boundary multiplication penalty factor, represents the updated optimized objective function value, represents the difference in free energy before and after particle filter sampling along the B-spline curve. Specifically, the budget multiplication penalty factor and the boundary multiplication penalty factor are respectively expressed as

[0078] ;

[0079] Among them, defines a smoothed single-sided step function, expressed as , represents the natural constant; defines a distance function, expressed as , represents a control point to the concentration field region the minimum distance of, represents the concentration field region another control point within.

[0080] It can be understood that when defining the boundary constraint, it is not possible to simply limit the control points of the path to not exceed the edge of the concentration field. It is also necessary to perform boundary constraint checks on the curve generated by the B-spline. After converting the constraint into a multiplication penalty factor, the form of the original path optimization problem becomes an unconstrained optimization problem, and the updated optimized objective function is constructed. The covariance matrix adaptation evolution strategy algorithm (CMA-ES) can be used to solve this function.

[0081] In one embodiment, the updated optimized objective function is calculated by the covariance evolution algorithm, and the parameterized perception path is continuously optimized in segments according to the calculated objective function value to obtain the optimized perception path, including:

[0082] The updated optimized objective function is calculated by the covariance evolution algorithm, and the parameterized perception path is continuously optimized in segments according to the calculated objective function value. After obtaining each segment of the optimized perception path and acquiring the concentration perception information therein, the free energy of this segment of the optimized perception path is updated according to the concentration perception information, and the new objective function value is calculated according to the updated free energy. Then, the next segment of the parameterized perception path is optimized according to the new objective function value until the optimization of all segmented paths is completed to obtain the complete optimized perception path.

[0083] In one embodiment, concentration perception information in a multi-gas diffusion source scenario is obtained along an optimized perception path, and the multi-gas diffusion source scenario is fitted according to the concentration perception information and a Gaussian process regression model to obtain a multi-source concentration field, including:

[0084] The concentration distribution of the multi-source concentration field is obtained by obtaining concentration perception information in the multi-gas diffusion source scenario along the optimized perception path and fitting the multi-gas diffusion source scenario according to the concentration perception information and the Gaussian process regression model, and is expressed as

[0085] ;

[0086] wherein, represents the mean function, represents the covariance function, represents the Gaussian process, represents the multi-gas diffusion source scenario in the concentration perception information, represents the multi-source concentration field.

[0087] Furthermore, the present invention uses the Matern kernel function to calculate the covariance function , and is expressed as

[0088] ;

[0089] wherein, represents the gamma function, represents the variable, ]>represents the concentration perception information and another concentration perception information the distance between, represents the modified Bessel function, and the hyperparameter in the function is obtained through prior information.

[0090] In one embodiment, a sequential multi-source confirmation mechanism is adopted to sequentially confirm the positions of multi-gas diffusion sources in the multi-source concentration field; wherein, the confirmation steps include an expectation step and a maximization step, and the specific steps are as follows:

[0091] First, in the expectation step, according to the peak algorithm, the concentration peak points in the multi-source concentration field with a concentration higher than the marking threshold are marked, and the concentration peak points are used as the estimated positions of the multi-gas diffusion sources. Specifically, assuming that there are sources in the concentration field, and their positions are , according to the multi-source concentration field fitted by the sensing unmanned aerial vehicle, it can be estimated that the most likely positions of the multi-sources are , and the peak algorithm is used to find the peak points in the multi-source concentration field higher than the marking threshold , taking as the estimated position of the source,

[0092] Then, in the maximum step, the ground robot carrying a vision sensor is used to sequentially go to the estimated positions of all multi-gas diffusion sources for position confirmation, obtain the determined positions of the multi-gas diffusion sources, and update the multi-source concentration field according to the determined positions of the multi-gas diffusion sources to obtain the updated multi-source concentration field.

[0093] Specifically, after obtaining the estimated position of the source , the ground robot carrying a vision sensor is used to sequentially go to the estimated positions of all multi-gas diffusion sources, and the estimated positions of all multi-gas diffusion sources are confirmed according to the sensor perception information obtained by the ground robot to obtain the determined positions of the multi-gas diffusion sources. At the same time, according to the ground robot's concentration perception at the determined positions of the multi-gas diffusion sources, the source term parameters of the multi-gas diffusion sources are obtained , according to the source term parameters to update the multi-source concentration field , and the concentration regions corresponding to the source term parameters in the multi-source concentration field are removed to obtain the updated multi-source concentration field to reduce interference for the desired step.

[0094] Finally, the desired step and the maximum step are sequentially executed until there is no concentration region with a concentration higher than the termination threshold in the updated multi-source concentration field , and the determined positions of all multi-gas diffusion sources in the multi-gas diffusion source scenario are obtained to complete the robot source search task in the multi-gas diffusion source scenario.

[0095] In summary, the present invention provides a robot source search scheme in a multi-gas diffusion source scenario based on information path planning. First, a perception drone is used to perform particle filter sampling on the gas concentration information of the scenario along the perception path, and an optimization objective function based on free energy of the perception path is constructed according to the sampling results. Then, the perception path is optimized according to the optimization objective function and the information path planning algorithm to obtain an optimized perception path. The drone moves along the optimized perception path to sense the concentration perception information in the scenario, and the multi-source concentration field in the scenario is fitted based on the Gaussian process regression model. Finally, a sequential confirmation mechanism is used to sequentially confirm the gas source positions in the multi-source concentration field to obtain the determined positions of all gas sources and complete the robot source search task in the multi-gas diffusion source scenario.

[0096] It should be understood that although Figure 1 the steps in the flowchart ofFigure 1 At least some of the steps may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed and completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least some of the sub-steps or stages of other steps or other steps.

[0097] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0098] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A robot source seeking method in a multi-gas diffusion source scenario based on information path planning, characterized in that, The method includes: Performing particle filter sampling on the gas concentration information in the multi-gas diffusion source scenario along the sensing path of the sensing drone, obtaining the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution, calculating according to the spatial entropy and the potential energy to obtain the free energy of the sensing path, and constructing an optimization objective function of the sensing path according to the free energy; Performing parameterization processing on the sensing path according to the information path planning algorithm, updating the optimization objective function according to the parameterized sensing path to obtain an updated optimization objective function, calculating the updated optimization objective function by the covariance evolution algorithm, and performing piecewise continuous optimization on the parameterized sensing path according to the calculated objective function value to obtain an optimized sensing path; Obtaining the concentration sensing information in the multi-gas diffusion source scenario along the optimized sensing path, and fitting the multi-gas diffusion source scenario according to the concentration sensing information and the Gaussian process regression model to obtain a multi-source concentration field; Adopting a sequential multi-source confirmation mechanism to sequentially confirm the positions of the multi-gas diffusion sources in the multi-source concentration field until the determined positions of all the multi-gas diffusion sources are obtained, and completing the robot source search task in the multi-gas diffusion source scenario; Among them, calculating according to the spatial entropy and the potential energy to obtain the free energy of the sensing path includes: Calculating according to the spatial entropy of the sensing path and the potential energy of the particle filter sample distribution to obtain the free energy of the sensing path, expressed as: ； Among them, represents the number of samplings, represents the spatial entropy, represents the potential energy of the th sampling, represents the temperature that controls the relative values of the potential energy and the spatial entropy, represents the trace of the matrix, represents the number of particle filter samples, represents the current position of the sensing UAV, represents the position of the th particle filter sample, represents the weight of the th particle filter sample, represents the intensity of the gravitational force of the gas diffusion source on the sensing UAV, <00,00031>represents the weighted covariance matrix of the particle filter sample distribution, represents the scale factor, represents the exponent that determines the rate of temperature change, represents the covariance of the concentration field on the sensing path; Among them, constructing an optimization objective function of the sensing path according to the free energy includes: Constructing an optimization objective function of the sensing path according to the free energy, expressed as: ； Among them, represents the difference in free energy before and after particle filtering sampling along the sensing path and and respectively represent the free energy before and after particle filtering sampling along the sensing path , represents the path cost constraint imposed on the optimization objective function, represents calculating the cost required to execute the sensing path , represents the budget.

2. The method according to claim 1, wherein Performing parameterization processing on the sensing path according to the information path planning algorithm, and updating the optimization objective function according to the parameterized sensing path to obtain an updated optimization objective function, including: Parameterizing the sensing path into a B-spline curve according to the information path planning algorithm, and constructing the parameterized sensing path by adding boundary constraints to the B-spline curve; Updating the optimization objective function according to the parameterized sensing path to obtain an updated optimization objective function.

3. The method according to claim 2, wherein Parameterizing the sensing path into a B-spline curve according to the information path planning algorithm, and constructing the parameterized sensing path by adding boundary constraints to the B-spline curve, including: Parameterize the perception path according to the information path planning algorithm as an order of B-spline curve with control points ; construct the parameterized perception path by adding boundary constraints to the B-spline curve, where the boundary constraints refer to constraining the starting point of the B-spline curve to coincide with the first control point and the ending point of the B-spline curve to coincide with the th control point .

4. The method according to claim 2, wherein Updating the optimization objective function according to the parameterized sensing path to obtain an updated optimization objective function, including: Updating the optimization objective function according to the parameterized sensing path to obtain a parameterized optimization objective function, expressed as: ； ； Among them, represents the parameterized perception path, represents the path cost constraint on the parameterized optimization objective function, represents the boundary constraint on the parameterized optimization objective function, represents the starting point constraint on the parameterized optimization objective function, represents a constant, represents the concentration field all feasible paths in; Converting the path cost constraint and the boundary constraint in the parameterized optimization objective function into a budget multiplication penalty factor and a boundary multiplication penalty factor respectively, and constructing an updated optimization objective function, expressed as: ； Among them, represents the budget multiplication penalty factor, represents the boundary multiplication penalty factor, represents the updated optimal objective function value, represents the difference in free energy before and after particle filter sampling along the B-spline curve.

5. The method according to claim 1, characterized in that, Calculating the updated optimization objective function by the covariance evolution algorithm, and performing piecewise continuous optimization on the parameterized sensing path according to the calculated objective function value to obtain an optimized sensing path, including: Calculate the updated optimized objective function through the covariance evolution algorithm, and perform piecewise continuous optimization on the parameterized sensing path according to the calculated objective function value. After obtaining each optimized sensing path segment and acquiring the concentration sensing information therein, update the free energy of this optimized sensing path segment according to the concentration sensing information, calculate a new objective function value based on the updated free energy, and then optimize the next parameterized sensing path segment according to the new objective function value until the optimization of all path segments is completed to obtain a complete optimized sensing path.

6. The method according to claim 1, characterized in that, Obtain the concentration sensing information in the multi-gas diffusion source scenario along the optimized sensing path, and fit the multi-gas diffusion source scenario according to the concentration sensing information and the Gaussian process regression model to obtain a multi-source concentration field, including: Obtain the concentration sensing information in the multi-gas diffusion source scenario along the optimized sensing path, and fit the multi-gas diffusion source scenario according to the concentration sensing information and the Gaussian process regression model to obtain the concentration distribution of the multi-source concentration field, expressed as: ； Among them, represents the mean value function, represents the covariance function, represents the Gaussian process, represents the multi-gas diffusion source scenario the concentration perception information in, represents the multi-source concentration field.

7. The method according to claim 1, characterized in that, Adopt a sequential multi-source confirmation mechanism to sequentially confirm the positions of multi-gas diffusion sources in the multi-source concentration field until the determined positions of all multi-gas diffusion sources are obtained, and complete the robot source-seeking task in the multi-gas diffusion source scenario, including: Adopt a sequential confirmation mechanism to sequentially confirm the positions of multi-gas diffusion sources in the multi-source concentration field; among them, the confirmation steps include an expectation step and a maximization step; In the expectation step, mark the concentration peak points in the multi-source concentration field with a concentration higher than the marking threshold according to the peak algorithm, and use the concentration peak points as the estimated positions of multi-gas diffusion sources; In the maximization step, use a ground robot equipped with a vision sensor to sequentially go to the estimated positions of all multi-gas diffusion sources for position confirmation, obtain the determined positions of multi-gas diffusion sources, and update the multi-source concentration field according to the determined positions of multi-gas diffusion sources to obtain an updated multi-source concentration field; Sequentially execute the expectation step and the maximization step until there is no concentration region with a concentration higher than the termination threshold in the updated multi-source concentration field, obtain the determined positions of all multi-gas diffusion sources in the multi-gas diffusion source scenario, and complete the robot source-seeking task in the multi-gas diffusion source scenario.

8. The method according to claim 7, wherein In the maximization step, use a ground robot equipped with a vision sensor to sequentially go to the estimated positions of all multi-gas diffusion sources for position confirmation, obtain the determined positions of multi-gas diffusion sources, and update the multi-source concentration field according to the determined positions of multi-gas diffusion sources to obtain an updated multi-source concentration field, including: In the maximization step, use a ground robot equipped with a vision sensor to sequentially go to the estimated positions of all multi-gas diffusion sources, and perform position confirmation on the estimated positions of all multi-gas diffusion sources according to the sensor sensing information obtained by the ground robot to obtain the determined positions of multi-gas diffusion sources; The ground robot performs concentration sensing at the determined position of the multi-gas diffusion source, obtains source term parameters of the multi-gas diffusion source, updates the multi-source concentration field according to the source term parameters, eliminates the concentration area corresponding to the source term parameters in the multi-source concentration field, and obtains an updated multi-source concentration field.

Images