CN114326755B - Robot gas source searching method based on local path planning - Google Patents

Robot gas source searching method based on local path planning Download PDF

Info

Publication number
CN114326755B
CN114326755B CN202210251365.6A CN202210251365A CN114326755B CN 114326755 B CN114326755 B CN 114326755B CN 202210251365 A CN202210251365 A CN 202210251365A CN 114326755 B CN114326755 B CN 114326755B
Authority
CN
China
Prior art keywords
source
scene
robot
gas
establishing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210251365.6A
Other languages
Chinese (zh)
Other versions
CN114326755A (en
Inventor
陈彬
季雅泰
王翔汉
秦龙
何华
尹全军
赵勇
朱正秋
艾川
邱思航
谢旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202210251365.6A priority Critical patent/CN114326755B/en
Publication of CN114326755A publication Critical patent/CN114326755A/en
Application granted granted Critical
Publication of CN114326755B publication Critical patent/CN114326755B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Feedback Control In General (AREA)
  • Manipulator (AREA)

Abstract

The invention relates to the field of robot control, and discloses a robot gas source searching method based on local path planning.A grid model of an unknown obstacle scene is established, the unknown obstacle scene is used as a source searching scene, and a convection equation model of gas source diffusion is established; establishing a perception model of the robot in the source search scene, forming a cognitive map according to obstacle information detected by the robot, and establishing a sensor model to convert the contact times of gas concentration and gas molecules; extracting sample information estimated by particle filtering according to a Gaussian mixture model, fitting a weighted sample to obtain the most effective Gaussian distribution, and taking the central point of the most effective Gaussian distribution as a target point of the robot; planning a search path of the source search scene of the robot by using a local path planning algorithm; and generating an unknown obstacle scene robot gas source searching scheme based on the local path planning.

Description

Robot gas source searching method based on local path planning
Technical Field
The application relates to the field of robot control, in particular to a robot gas source searching method based on local path planning.
Background
The problem of searching for gas diffusion sources is an important issue in the field of robotic source searching. The first time, people put forward bionics algorithms of source search problems, such as Chemotaxis strategies moving along the direction of concentration gradient, silk algorithms simulating moths to locate odor sources, and the like, by studying the process of searching gas sources through olfaction or other senses of animals. As research further advances, the background of the source searching problem becomes more complex, which puts higher demands on the source search of the robot. The appearance of the cognitive search strategy greatly improves the source searching effect of the robot. Vergasola et al proposed the earliest cognitive search strategy, known as the Infotaxis algorithm, which uses a grid-based approach to maintain information state. Ristic refines the Infotaxi algorithm by using sparse sensing cues in the form of sporadic non-zero sensor measurements. Hutchinson et al propose another cognitive search algorithm, Entrotaxi, which designs a reward function based on maximum entropy sampling. These search strategies express the source search problem as a particle filter-based Partially Observable Markov Decision Process (POMDP), replacing the grid-based approach. In the research of robot source search problems in obstacle scenes, Zhaoyao proposes an Entrotaxi-Jump algorithm to control the source search action of a robot in the obstacle scenes, and the algorithm leads the robot to Jump out of a blocking area in the source search process by introducing a Jump mechanism. Meanwhile, the cognitive strategy with the forbidden region is provided for solving the complex road network constraint problem, and the algorithm prevents the robot from entering the 'dead end' in the source searching process by setting the forbidden region.
Therefore, how to provide a robot gas source search action scheme under an unknown obstacle scene becomes a technical problem to be solved urgently.
The above is only for the purpose of assisting understanding of the technical aspects of the present invention, and does not represent an admission that the above is prior art.
Disclosure of Invention
The invention mainly aims to provide a robot gas source searching method based on local path planning, and aims to solve the technical problem that the prior art cannot provide a robot gas source searching action scheme under an unknown scene.
In order to achieve the above object, the present invention provides a robot gas source searching method based on local path planning, the method comprising:
establishing a grid model of an unknown obstacle scene, taking the unknown obstacle scene as a source search scene, and establishing a convection equation model of gas source diffusion, wherein a gas diffusion source is randomly generated in the source search scene;
establishing a perception model of the robot in the source search scene, forming a cognitive map according to obstacle information detected by the robot, and establishing a sensor model to convert the contact times of gas concentration and gas molecules;
extracting sample information estimated by particle filtering according to a Gaussian mixture model, fitting a weighted sample to obtain the most effective Gaussian distribution, and taking the central point of the most effective Gaussian distribution as a target point of the robot;
planning a search path of the source search scene of the robot by using a local path planning algorithm;
and generating an unknown obstacle scene robot gas source searching scheme based on the local path planning.
Optionally, the establishing a grid model of an unknown obstacle scene, taking the unknown obstacle scene as a source search scene, and establishing a convection equation model of gas source diffusion, where the step of randomly generating a gas diffusion source in the source search scene includes:
establishing an unknown obstacle scene model based on a grid method, dividing a corresponding map in the unknown obstacle scene into square grids with the same size, taking all the square grids as a source search scene, wherein the square grids are occupied with probability;
the grid to be occupied constitutes a barrier;
and randomly generating a gas diffusion source in the source search scene, and establishing a gas diffusion source diffusion model based on a convection diffusion equation.
Optionally, the step of establishing a perception model of the robot in the source search scene, forming a cognitive map according to the obstacle information detected by the robot, and establishing a sensor model to convert the number of times of contact between the gas concentration and the gas molecules includes:
establishing a cognitive map based on the grid map to collect the obstacle distribution condition of the peripheral area sensed by the robot, wherein the sensed grid around the new position is changed from an unknown grid to a known grid;
establishing a sensor perception model, and converting gas concentration data perceived by a sensor into average contact times of the sensor and gas molecules in unit time by using a Simouhol-Fucky formula:
Figure 735212DEST_PATH_IMAGE001
wherein the radius of the sensor is a, the position of the searcher is r,
Figure 777117DEST_PATH_IMAGE002
the position of the gas diffusion source in the source search scene is
Figure 198872DEST_PATH_IMAGE003
，
Figure 402451DEST_PATH_IMAGE004
Is its diffusion strength in the scene.
Optionally, after the step of establishing a sensor sensing model and converting the gas concentration data sensed by the sensor into the average number of times of contact between the sensor and the gas molecules in a unit time by using the schumohwisky formula, the method further includes:
simulating the effects of turbulence on the gas concentration field using the Poisson process, define
Figure 184856DEST_PATH_IMAGE005
The intensity of the poisson process is such that the sensor contacts the gas molecules per unit time at the location r
Figure 889506DEST_PATH_IMAGE006
The probability of the second order is:
Figure 306712DEST_PATH_IMAGE007
and generating the gas contact times of any position of the source search scene according to a Poisson-distributed random number generation method.
Optionally, the extracting, according to the gaussian mixture model, sample information estimated by particle filtering, fitting a weighted sample to obtain a most effective gaussian distribution, and using a central point of the most effective gaussian distribution as a target point of the robot includes:
estimating source item parameters by adopting a Bayes framework, defining information states in a partial observable Markov decision process as posterior probability density functions related to the source item parameters, expressing the estimation of the source item parameters by the posterior probability density functions, and expressing the posterior probability density functions in the k step as follows:
Figure 805827DEST_PATH_IMAGE008
wherein
Figure 308483DEST_PATH_IMAGE009
Representing all the information collected for the first k steps,
Figure 551246DEST_PATH_IMAGE010
representing the source item parameters of the estimation of the k step;
when the sensor senses new information, updating the posterior probability distribution by using a Bayesian formula:
Figure 88538DEST_PATH_IMAGE011
wherein the content of the first and second substances,
Figure 758553DEST_PATH_IMAGE012
；
the Bayesian estimation of the source term parameters is realized by utilizing a particle filtering method,
Figure 247041DEST_PATH_IMAGE013
is approximated as N weighted samples
Figure 27915DEST_PATH_IMAGE014
；
Figure 154134DEST_PATH_IMAGE015
Wherein the content of the first and second substances,
Figure 995052DEST_PATH_IMAGE016
is shown as
Figure 472300DEST_PATH_IMAGE017
Step (1) of
Figure 322445DEST_PATH_IMAGE018
A point estimate of the source term parameters for a sample,
Figure 568749DEST_PATH_IMAGE019
is shown as
Figure 222978DEST_PATH_IMAGE020
Step (1) of
Figure 46578DEST_PATH_IMAGE021
Normalized weight of individual sample, satisfy
Figure 310200DEST_PATH_IMAGE022
Is a dirac
Figure 535645DEST_PATH_IMAGE023
A function;
the weighted samples are updated using a sequential importance sampling approach:
Figure 452785DEST_PATH_IMAGE024
after normalization, a new approximate Bayesian estimation is obtained:
Figure 904626DEST_PATH_IMAGE025
a resampling step is used to increase the particle diversity to determine the target point for the source search.
Optionally, the step of using a resampling step to increase the particle diversity to determine the target point of the source search comprises:
based on approximate Bayesian estimation of source term parameters in the step of particle filtering, i.e.
Figure 830994DEST_PATH_IMAGE026
A set of weighted samples, each weighted sample representing a point estimate of a source position, for the source position using a Gaussian mixture model
Figure 786312DEST_PATH_IMAGE027
Fitting the weighted samples to obtain samples
Figure 139933DEST_PATH_IMAGE028
Sub-distribution, which guides the action of the searcher by using the reward function of the clustering information of the samples;
the posterior probability distribution of the source term parameters can be fitted by a gaussian mixture model as:
Figure 312026DEST_PATH_IMAGE029
wherein the content of the first and second substances,
Figure 42084DEST_PATH_IMAGE030
represents a mean value of
Figure 117488DEST_PATH_IMAGE031
Covariance of
Figure 907589DEST_PATH_IMAGE032
Of a gaussian distribution, coefficients of a gaussian distribution
Figure 68443DEST_PATH_IMAGE033
Satisfy the requirement of
Figure 336613DEST_PATH_IMAGE034
Due to the fact that
Figure 944DEST_PATH_IMAGE035
Defining the most effective Gaussian distribution of the Gaussian distribution, and taking the mean point of the Gaussian distribution
Figure 961947DEST_PATH_IMAGE036
As the target point of the source search.
Optionally, the step of randomly generating a gas diffusion source in the source search scenario and establishing a gas diffusion source diffusion model based on a convection diffusion equation includes:
defining boundary nodes of a known area and an unknown area as n, an evaluation function h (n) is the distance from the n nodes to a target node e, and g (n) is the distance from the current position s of a searcher to the boundary node n;
the total cost of the searcher to go to the target node is:
Figure 845983DEST_PATH_IMAGE037
the optimal cost under the optimal path is as follows:
Figure 917844DEST_PATH_IMAGE038
with respect to the known area(s),
Figure 967839DEST_PATH_IMAGE039
directly obtained by the existing path planning algorithm, and for an unknown area, defining
Figure 834164DEST_PATH_IMAGE040
I.e. the manhattan distance of the boundary node n to the target node eBecause of the existence of obstacles in the unknown area, then
Figure 969611DEST_PATH_IMAGE041
And when the Manhattan distance is established, taking the unknown area as a blank area in the searching process, and simplifying the searching process into a path planning process from the initial node to the target node so as to obtain a corresponding searching path.
The method comprises the steps of establishing a grid model of an unknown obstacle scene, taking the unknown obstacle scene as a source search scene, and establishing a convection equation model of gas source diffusion, wherein a gas diffusion source is randomly generated in the source search scene; establishing a perception model of the robot in the source search scene, forming a cognitive map according to obstacle information detected by the robot, and establishing a sensor model to convert the contact times of gas concentration and gas molecules; extracting sample information estimated by particle filtering according to a Gaussian mixture model, fitting a weighted sample to obtain the most effective Gaussian distribution, and taking the central point of the most effective Gaussian distribution as a target point of the robot; planning a search path of the source search scene of the robot by using a local path planning algorithm; generating an unknown obstacle scene robot gas source searching scheme based on local path planning, and finally forming a robot gas source searching action scheme under the unknown obstacle scene based on the local path planning by introducing a local path planning algorithm.
Drawings
Fig. 1 is a schematic flow chart of a robot gas source searching method based on local path planning according to a first embodiment of the present invention;
FIG. 2 is a schematic view of a grid of a source search scene according to a first embodiment of the method for searching a gas source of a robot based on local path planning;
FIG. 3 is a schematic diagram of a searching method of a most effective Gaussian distribution region according to a first embodiment of the robot gas source searching method based on local path planning;
fig. 4 is a flowchart of a robot gas source search method based on local path planning according to a first embodiment of the present invention, wherein the robot gas source search method is based on local path planning and is used in an unknown obstacle scene.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
An embodiment of the present invention provides a robot gas source searching method based on local path planning, and referring to fig. 1, fig. 1 is a schematic flow diagram of a first embodiment of a robot gas source searching method based on local path planning according to the present invention.
In this embodiment, the robot gas source searching method based on local path planning includes the following steps:
step S10: establishing a grid model of an unknown obstacle scene, taking the unknown obstacle scene as a source search scene, and establishing a convection equation model of gas source diffusion, wherein a gas diffusion source is randomly generated in the source search scene.
In specific implementation, the present embodiment divides the environment information into grids for storing the passable areas of the map and the distribution of the obstacles, as shown in fig. 2. Defining a source search area as having a side length of
Figure 579583DEST_PATH_IMAGE042
Square area of
Figure 343140DEST_PATH_IMAGE043
Inside which a limited number of static obstacles are distributed
Figure 255732DEST_PATH_IMAGE044
To do so by
Figure 3109DEST_PATH_IMAGE045
Are respectively
Figure 790674DEST_PATH_IMAGE046
A shaft,
Figure 408737DEST_PATH_IMAGE047
Axis with the point of intersection of the two boundaries as origin of coordinates
Figure 757810DEST_PATH_IMAGE048
And establishing a rectangular coordinate system. Assume that the searcher step size is a fixed value
Figure 726903DEST_PATH_IMAGE049
To do so by
Figure 554044DEST_PATH_IMAGE050
To form a grid for a side length
Figure 292193DEST_PATH_IMAGE051
Become into
Figure 671222DEST_PATH_IMAGE052
Is rasterized area of
Figure 2977DEST_PATH_IMAGE053
Wherein
Figure 492865DEST_PATH_IMAGE054
Figure 716211DEST_PATH_IMAGE055
Consisting of occupied grids in the map, each grid having the same probability
Figure 531721DEST_PATH_IMAGE056
The number of occupied, then,
Figure 350772DEST_PATH_IMAGE057
wherein
Figure 378771DEST_PATH_IMAGE058
Is composed of
Figure 91512DEST_PATH_IMAGE059
The number of the occupied grids.
In the process of source search, a searcher cannot move to the grid where the obstacle is located, and can only move among the passable grids. Probability of being
Figure 953289DEST_PATH_IMAGE060
Too large, then
Figure 118691DEST_PATH_IMAGE061
Will be divided by the occupied grid, i.e., there may not be a navigable path between the searcher and the location of the source. Therefore, in order to satisfy the connectivity, the present embodiment provides
Figure 91326DEST_PATH_IMAGE062
。
It can be understood that the present embodiment continues to use the modeling of the diffusion process in the cognitive search strategy to establish a convective diffusion equation model of the gas diffusion in space. It is assumed that the height of the obstacles in the source search scene is low, so the effect of the obstacles on the diffusion of gas molecules is neglected. The position of the gas diffusion source in the source search scenario is
Figure 658574DEST_PATH_IMAGE063
，
Figure 189787DEST_PATH_IMAGE064
The source item parameter that the present embodiment needs to estimate is its diffusion strength in the scene
Figure 373644DEST_PATH_IMAGE065
. Considering the wind speed of the gas in the diffusion process
Figure 149970DEST_PATH_IMAGE066
Influence, obtainable from steady state convective diffusion equations, location in source search scene
Figure 712669DEST_PATH_IMAGE067
The gas concentrations were:
Figure 775303DEST_PATH_IMAGE068
wherein, the first and the second end of the pipe are connected with each other,
Figure 915298DEST_PATH_IMAGE069
and D is the effective diffusion coefficient of the gas,
Figure 229735DEST_PATH_IMAGE070
is the gas molecular lifetime. Assuming effective diffusion coefficient D of gas, gas molecular lifetime
Figure 771575DEST_PATH_IMAGE071
And wind speed
Figure 647521DEST_PATH_IMAGE072
It is known that the quantity to be estimated in the gas source searching process is the source item parameter
Figure 274811DEST_PATH_IMAGE073
。
Further, the step of establishing a grid model of an unknown obstacle scene, taking the unknown obstacle scene as a source search scene, and establishing a convection equation model of gas source diffusion, wherein a gas diffusion source is randomly generated in the source search scene, includes: establishing an unknown obstacle scene model based on a grid method, dividing a corresponding map in the unknown obstacle scene into square grids with the same size, taking the unknown obstacle scene as a source search scene, wherein all the square grids are occupied with probability; the grid to be occupied constitutes a gas barrier; and randomly generating a gas diffusion source in the source search scene, and establishing a gas diffusion source diffusion model based on a convection diffusion equation.
Step S20: and establishing a perception model of the robot in the source search scene, forming a cognitive map according to the obstacle information detected by the robot, and establishing a sensor model to convert the contact times of the gas concentration and the gas molecules.
It will be appreciated that the seeker makes decisions based on the perceived environment as they move. A general searcher such as a ground robot has a limited sensing radius and can only sense the distribution of surrounding obstacles. If the surrounding environment is complex, the searcher is easily trapped in a local loop. Therefore, a cognitive map needs to be constructed for the searcher, the perceived environment is integrated into the cognitive map of the searcher, and cognitive map support is provided for subsequent source search activities.
In a particular implementation, a cognitive map of a searcher is constructed based on a grid map. Assuming a perceived radius of
Figure 986415DEST_PATH_IMAGE074
To search the position of the person
Figure 523707DEST_PATH_IMAGE075
Is the center and the periphery
Figure 193723DEST_PATH_IMAGE074
The layer grids are perceivable grids. Cognitive maps will remember the state of all the perceivable grids as the searcher position
Figure 449255DEST_PATH_IMAGE075
In the change, the perceptible grid around the new location changes from the unknown grid to the known grid (occupied and passable). When a searcher moves, the searcher can sense the gas concentration of the current position through the sensor, and because the precision of the existing spherical sensor (the radius is a) is low, the gas concentration data of any position needs to be converted into the average contact frequency of the sensor and gas molecules in unit time by using a Schumohov formula:
Figure 105495DEST_PATH_IMAGE076
in a real environment, gas is influenced by a turbulence effect when being diffused, so that a concentration field is disturbed, a sensor can only obtain sporadic and discontinuous effective readings when sensing, and in a cognitive search strategy, the embodiment simulates the turbulence effect by using a Poisson processIn response to the influence of the gas concentration field, define
Figure 621927DEST_PATH_IMAGE077
The intensity of the poisson process is such that the sensor contacts the gas molecules per unit time at the location r
Figure 836745DEST_PATH_IMAGE078
The probability of the next order is
Figure 438628DEST_PATH_IMAGE079
According to the Poisson distributed random number generation method, the gas contact times of any position of the source search scene can be generated.
Further, the step of establishing a perception model of the robot in the source search scene, forming a cognitive map according to obstacle information detected by the robot, and establishing a sensor model to convert the number of times of contact between gas concentration and gas molecules includes: establishing a cognitive map based on the grid map to collect the obstacle distribution condition of the peripheral area sensed by the robot, wherein the sensed grid around the new position is changed from an unknown grid to a known grid; establishing a sensor perception model, and converting gas concentration data perceived by a sensor into average contact times of the sensor and gas molecules in unit time by using a Simouhol-Fucky formula:
Figure 898559DEST_PATH_IMAGE080
wherein the radius of the sensor is a, the position of the searcher is r,
Figure 269498DEST_PATH_IMAGE081
the position of the gas diffusion source in the source search scene is
Figure 156682DEST_PATH_IMAGE082
，
Figure 980282DEST_PATH_IMAGE083
Is its diffusion strength in the scene.
Further, after the step of establishing a sensor sensing model and converting the gas concentration data sensed by the sensor into the average number of times of contact between the sensor and the gas molecules in a unit time by using the schumohwisky formula, the method further includes: simulating the effects of turbulence on the gas concentration field using the Poisson process, define
Figure 243904DEST_PATH_IMAGE084
The intensity of the poisson process is such that the sensor contacts the gas molecules per unit time at the location r
Figure 469349DEST_PATH_IMAGE085
The probability of the second order is:
Figure 28900DEST_PATH_IMAGE086
and generating the gas contact times of any position of the source search scene according to a Poisson-distributed random number generation method.
Step S30: and extracting sample information estimated by particle filtering according to the Gaussian mixture model, fitting the weighted sample to obtain the most effective Gaussian distribution, and taking the central point of the most effective Gaussian distribution as a target point of the robot.
In specific implementation, a Bayesian framework is adopted to estimate source item parameters, information states in a partial observable Markov decision process are defined as posterior probability density functions related to the source item parameters, the posterior probability density functions are used for representing the estimation of the source item parameters, and the estimation of the source item parameters is updated by the information sensed at each step, so that the posterior probability density function at the k step can be represented as
Figure 339795DEST_PATH_IMAGE087
Wherein
Figure 266163DEST_PATH_IMAGE088
Representing all the information collected for the first k steps,
Figure 221481DEST_PATH_IMAGE089
and (3) representing the source item parameters of the estimation in the k step, and updating the posterior probability distribution by using a Bayesian formula when the sensor senses new information:
Figure 575102DEST_PATH_IMAGE090
wherein
Figure 248660DEST_PATH_IMAGE091
The initial information state may be obtained by obtaining prior information, and if there is no prior information, the initial probability distribution is represented using a uniform distribution. And the Bayesian estimation of the source term parameters is realized by utilizing a particle filtering method. Since the above expression is difficult to solve by a functional expression,
Figure 713139DEST_PATH_IMAGE092
is approximated as N weighted samples
Figure 54122DEST_PATH_IMAGE093
Figure 578644DEST_PATH_IMAGE094
Wherein the content of the first and second substances,
Figure 972454DEST_PATH_IMAGE095
is shown as
Figure 506203DEST_PATH_IMAGE096
Step (1) of
Figure 436113DEST_PATH_IMAGE097
A point estimate of the source term parameters for a sample,
Figure 865958DEST_PATH_IMAGE098
is shown as
Figure 638742DEST_PATH_IMAGE099
Step (1) of
Figure 320390DEST_PATH_IMAGE100
The normalized weight of each sample, satisfied,
Figure 229440DEST_PATH_IMAGE101
Figure 236710DEST_PATH_IMAGE102
is a dirac
Figure 231211DEST_PATH_IMAGE103
A function. Next, the weighted samples are updated using a sequential importance sampling method:
Figure 841184DEST_PATH_IMAGE104
after normalization, a new approximate Bayesian estimation is obtained:
Figure 247151DEST_PATH_IMAGE105
in order to avoid the occurrence of particle degradation phenomena, a resampling step is used to increase the particle diversity. In the step of particle filtering, an approximate Bayesian estimation of the source term parameters is obtained, i.e.
Figure 284377DEST_PATH_IMAGE106
A set of weighted samples. Each weighted sample represents a point estimate of the source location, so the more concentrated the samples are in the search space, the more likely the source location is to appear in that region. The patent uses a Gaussian mixture model pair
Figure 907119DEST_PATH_IMAGE106
Fitting the weighted samples to obtain samples
Figure 461729DEST_PATH_IMAGE107
And sub-distribution, namely, the clustering information of the samples is used for replacing a reward function in the traditional method to guide the action of the searcher. The posterior probability distribution of the source term parameters can be fit by a gaussian mixture model as:
Figure 345371DEST_PATH_IMAGE108
wherein the content of the first and second substances,
Figure 428865DEST_PATH_IMAGE109
represents a mean value of
Figure 397958DEST_PATH_IMAGE110
Covariance of
Figure 349733DEST_PATH_IMAGE111
A gaussian distribution of (a). Coefficient of Gaussian distribution
Figure 461783DEST_PATH_IMAGE112
And (4) meeting the requirement.
Figure 106391DEST_PATH_IMAGE113
Due to the fact that
Figure 172567DEST_PATH_IMAGE114
The maximum Gaussian distribution occupies most samples, the most effective Gaussian distribution of the Gaussian distribution is defined, and the probability of existence of the area source where the most effective Gaussian distribution exists is highest, so that the mean value of the Gaussian distribution can be calculated
Figure 928034DEST_PATH_IMAGE115
As a target point of the source search, as shown in fig. 3.
Further, the step of extracting sample information estimated by particle filtering according to a gaussian mixture model, fitting a weighted sample to obtain a most effective gaussian distribution, and using a center point of the most effective gaussian distribution as a target point of the robot includes: estimating source item parameters by adopting a Bayes framework, defining information states in a partial observable Markov decision process as posterior probability density functions related to the source item parameters, expressing the estimation of the source item parameters by the posterior probability density functions, and expressing the posterior probability density functions in the k step as follows:
Figure 661635DEST_PATH_IMAGE116
wherein
Figure 477144DEST_PATH_IMAGE117
Representing all the information collected for the first k steps,
Figure 889671DEST_PATH_IMAGE118
representing the source item parameters of the estimation of the k step;
when the sensor senses new information, updating the posterior probability distribution by using a Bayesian formula:
Figure 324194DEST_PATH_IMAGE119
wherein the content of the first and second substances,
Figure 36935DEST_PATH_IMAGE120
；
the Bayesian estimation of the source term parameters is realized by utilizing a particle filtering method,
Figure 146316DEST_PATH_IMAGE121
is approximated as N weighted samples
Figure 577298DEST_PATH_IMAGE122
：
Figure 549933DEST_PATH_IMAGE123
Wherein the content of the first and second substances,
Figure 117181DEST_PATH_IMAGE124
is shown as
Figure 8913DEST_PATH_IMAGE125
Step (1) of
Figure 536978DEST_PATH_IMAGE126
A point estimate of the source term parameters for a sample,
Figure 437937DEST_PATH_IMAGE127
is shown as
Figure 637DEST_PATH_IMAGE128
Step (1) of
Figure 63271DEST_PATH_IMAGE129
Normalized weight of individual sample, satisfy
Figure 577167DEST_PATH_IMAGE130
。
Figure 16238DEST_PATH_IMAGE131
Is a dirac
Figure 433444DEST_PATH_IMAGE132
A function;
the weighted samples are updated using a sequential importance sampling approach:
Figure 666979DEST_PATH_IMAGE133
after normalization, a new approximate Bayesian estimation is obtained:
Figure 294270DEST_PATH_IMAGE134
a resampling step is used to increase the particle diversity to determine the target point for the source search.
Further, the step of using a resampling step to increase the diversity of particles to determine the target point of the source search comprises: based on approximate Bayesian estimation of source term parameters in the step of particle filtering, i.e.
Figure 412399DEST_PATH_IMAGE135
A set of weighted samples, each weighted sample representing a point estimate of a source location, for the source location using a Gaussian mixture model
Figure 808745DEST_PATH_IMAGE135
Fitting the weighted samples to obtain samples
Figure 354127DEST_PATH_IMAGE136
Sub-distribution, which guides the action of the searcher by using the reward function of the clustering information of the samples;
the posterior probability distribution of the source term parameters can be fitted by a gaussian mixture model as:
Figure 203134DEST_PATH_IMAGE137
wherein the content of the first and second substances,
Figure 626419DEST_PATH_IMAGE138
represents a mean value of
Figure 142851DEST_PATH_IMAGE139
Covariance of
Figure 718188DEST_PATH_IMAGE140
Of a gaussian distribution, coefficients of a gaussian distribution
Figure 929858DEST_PATH_IMAGE141
Satisfy the requirement of
Figure 514423DEST_PATH_IMAGE142
Due to the fact that
Figure 495149DEST_PATH_IMAGE143
Defining the most effective Gaussian distribution of the Gaussian distribution, and taking the mean point of the Gaussian distribution
Figure 506967DEST_PATH_IMAGE144
As the target point of the source search.
Step S40: and planning a search path of the source search scene of the robot by using a local path planning algorithm.
Further, the step of randomly generating a gas diffusion source in the source search scenario and establishing a gas diffusion source diffusion model based on a convection diffusion equation includes: defining boundary nodes of a known area and an unknown area as n, an evaluation function h (n) is the distance from the n nodes to a target node e, and g (n) is the distance from the current position s of a searcher to the boundary node n; the total cost of the searcher to go to the target node is:
Figure 471512DEST_PATH_IMAGE145
the optimal cost under the optimal path is as follows:
Figure 594189DEST_PATH_IMAGE146
with respect to the known area(s),
Figure 927956DEST_PATH_IMAGE147
directly obtained by the existing path planning algorithm, and for an unknown area, defining
Figure 986042DEST_PATH_IMAGE148
I.e., the manhattan distance from the boundary node n to the target node e, because an obstacle exists in the unknown area, then
Figure 296937DEST_PATH_IMAGE149
And when the Manhattan distance is established, taking the unknown area as a blank area in the searching process, and simplifying the searching process into a path planning process from the initial node to the target node so as to obtain a corresponding searching path.
Step S50: and generating an unknown obstacle scene robot gas source searching scheme based on the local path planning.
In specific implementation, as shown in fig. 4, a searcher senses the surrounding environment in an unknown obstacle scene, obtains and updates the particle filter estimation of the source position, obtains the most effective gaussian distribution area with the maximum probability of the source position by a cognitive search strategy algorithm based on gaussian mixture distribution, draws a search path to the center of the area by using the existing path planning algorithm, and continuously senses the surrounding environment and the obstacle in the process that a robot moves on the path, updates the map and the particle filter estimation, and obtains a new most effective gaussian distribution area and a new search path until a termination condition is reached, such as finding the source position.
According to the method, an unknown obstacle scene is used as a source search scene according to a grid model for establishing the unknown obstacle scene, a convection equation model of gas source diffusion is established, wherein a gas diffusion source is randomly generated in the source search scene; establishing a perception model of the robot in the source search scene, forming a cognitive map according to obstacle information detected by the robot, and establishing a sensor model to convert the contact times of gas concentration and gas molecules; extracting sample information estimated by particle filtering according to a Gaussian mixture model, fitting a weighted sample to obtain the most effective Gaussian distribution, and taking the central point of the most effective Gaussian distribution as a target point of the robot; planning a search path of the source search scene of the robot by using a local path planning algorithm; generating an unknown obstacle scene robot gas source searching scheme based on local path planning, and finally forming a robot gas source searching action scheme under the unknown obstacle scene based on the local path planning by introducing a local path planning algorithm.
The embodiments or specific implementation manners of the robot gas source searching device based on the local path planning of the present invention may refer to the above method embodiments, and are not described herein again.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., a rom/ram, a magnetic disk, an optical disk) and includes instructions for enabling a terminal device (e.g., a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (5)

1. A robot gas source searching method based on local path planning is characterized by comprising the following steps:
establishing a grid model of an unknown obstacle scene, taking the unknown obstacle scene as a source search scene, and establishing a convection diffusion equation model of gas source diffusion, wherein a gas diffusion source is randomly generated in the source search scene;
establishing a perception model of the robot in the source search scene, forming a cognitive map according to obstacle information detected by the robot, and establishing a sensor model to convert the contact times of gas concentration and gas molecules;
extracting sample information estimated by particle filtering according to a Gaussian mixture model, fitting a weighted sample to obtain the most effective Gaussian distribution, and taking the central point of the most effective Gaussian distribution as a target point of the robot;
planning a search path of the source search scene of the robot by using a local path planning algorithm;
generating an unknown obstacle scene robot gas source searching scheme based on local path planning;
the establishing of the grid model of the unknown obstacle scene, taking the unknown obstacle scene as a source searching scene, and establishing a convection equation model of gas source diffusion, wherein the step of randomly generating a gas diffusion source in the source searching scene comprises the following steps:
establishing an unknown obstacle scene model based on a grid method, dividing a corresponding map in the unknown obstacle scene into square grids with the same size, taking all the square grids as a source search scene, wherein the square grids are occupied with probability;
the grid to be occupied constitutes a barrier;
randomly generating a gas diffusion source in the source search scene, and establishing a gas diffusion source diffusion model based on a convection diffusion equation;
the step of randomly generating a gas diffusion source in the source search scene and establishing a gas diffusion source diffusion model based on a convection diffusion equation comprises the following steps:
defining boundary nodes of a known area and an unknown area as n, an evaluation function h (n) is the distance from the n nodes to a target node e, and g (n) is the distance from the current position s of a searcher to the boundary node n;
the total cost of the searcher to go to the target node is:
Figure 659577DEST_PATH_IMAGE001
the optimal cost under the optimal path is as follows:
Figure 387492DEST_PATH_IMAGE002
with respect to the known area(s),
Figure 783839DEST_PATH_IMAGE003
directly obtained by the existing path planning algorithm, and for an unknown area, defining
Figure 453854DEST_PATH_IMAGE004
I.e., the manhattan distance from the boundary node n to the target node e, because an obstacle exists in the unknown area, then
Figure 302862DEST_PATH_IMAGE005
And when the Manhattan distance is established, taking the unknown area as a blank area in the searching process, and simplifying the searching process into a path planning process from the initial node to the target node so as to obtain a corresponding searching path.
2. The method of claim 1, wherein the step of modeling the robot's perception in the source search scenario, forming a cognitive map based on obstacle information detected by the robot, and modeling the sensors to convert the number of times the gas concentration has been in contact with the gas molecules comprises:
establishing a cognitive map based on the grid map to collect the obstacle distribution condition of the peripheral area sensed by the robot, wherein the sensed grid around the new position is changed from an unknown grid to a known grid;
establishing a sensor perception model, and converting gas concentration data perceived by a sensor into average contact times of the sensor and gas molecules in unit time by using a Simouhol-Fucky formula:
Figure 83736DEST_PATH_IMAGE006
wherein the radius of the sensor is a, the position of the searcher is r,
Figure 334589DEST_PATH_IMAGE007
the position of the gas diffusion source in the source search scene is
Figure 926238DEST_PATH_IMAGE008
，
Figure 262542DEST_PATH_IMAGE009
Is its diffusion strength in the scene;
Figure 847107DEST_PATH_IMAGE010
representing gas concentration data at searcher location r.
3. The method of claim 2, wherein the step of establishing a sensor-aware model to convert the gas concentration data sensed by the sensor into an average number of contacts the sensor has with gas molecules per unit time using the schumohwisky equation further comprises:
simulating the effects of turbulence on the gas concentration field using the Poisson process, define
Figure 952466DEST_PATH_IMAGE011
For the intensity of the poisson process, the sensor is then in contact with air per unit time at position rBody molecules
Figure 964284DEST_PATH_IMAGE012
The probability of the second order is:
Figure 273037DEST_PATH_IMAGE013
and generating the gas contact times of any position of the source search scene according to a Poisson-distributed random number generation method.
4. The method of claim 1, wherein the extracting of the sample information of the particle filter estimation according to the gaussian mixture model, fitting the weighted samples to obtain the most efficient gaussian distribution, and using the center point of the most efficient gaussian distribution as the target point of the robot comprises:
estimating source item parameters by adopting a Bayes framework, defining information states in a partial observable Markov decision process as posterior probability density functions related to the source item parameters, expressing the estimation of the source item parameters by the posterior probability density functions, and expressing the posterior probability density functions in the k step as follows:
Figure 661293DEST_PATH_IMAGE014
wherein
Figure 621159DEST_PATH_IMAGE015
Representing all the information collected for the first k steps,
Figure 803878DEST_PATH_IMAGE016
representing the source item parameters of the estimation of the k step;
when the sensor senses new information, updating the posterior probability distribution by using a Bayesian formula:
Figure 114774DEST_PATH_IMAGE017
wherein the content of the first and second substances,
Figure 791874DEST_PATH_IMAGE018
；
the Bayesian estimation of the source term parameters is realized by utilizing a particle filtering method,
Figure 606246DEST_PATH_IMAGE019
is approximated as N weighted samples
Figure 225447DEST_PATH_IMAGE020
：
Figure 758059DEST_PATH_IMAGE021
Wherein the content of the first and second substances,
Figure 4
is shown as
Figure 720330DEST_PATH_IMAGE026
Step (1) of
Figure 774874DEST_PATH_IMAGE027
A point estimate of the source term parameters for a sample,
Figure 452160DEST_PATH_IMAGE025
is shown as
Figure 720330DEST_PATH_IMAGE026
Step (1) of
Figure 774874DEST_PATH_IMAGE027
Normalized weight of individual sample, satisfy
Figure 204718DEST_PATH_IMAGE028
，
Figure 728234DEST_PATH_IMAGE029
Is a dirac
Figure 534516DEST_PATH_IMAGE030
A function;
the weighted samples are updated using a sequential importance sampling approach:
Figure 443567DEST_PATH_IMAGE031
after normalization, a new approximate Bayesian estimation is obtained:
Figure 309891DEST_PATH_IMAGE032
a resampling step is used to increase the particle diversity to determine the target point for the source search.
5. The method of claim 4, wherein the step of using the resampling step to increase the particle diversity to determine the target point for the source search comprises:
based on approximate Bayesian estimation of source term parameters in the step of particle filtering, i.e.
Figure 304392DEST_PATH_IMAGE034
A set of weighted samples, each weighted sample representing a point estimate of a source position, for the source position using a Gaussian mixture model
Figure 914365DEST_PATH_IMAGE034
Fitting the weighted samples to obtain samples
Figure 428654DEST_PATH_IMAGE036
Sub-distribution, which guides the action of the searcher by using the reward function of the clustering information of the samples;
the posterior probability distribution of the source term parameters can be fitted by a gaussian mixture model as:
Figure 465880DEST_PATH_IMAGE037
wherein the content of the first and second substances,
Figure 947677DEST_PATH_IMAGE038
represents a mean value of
Figure 361341DEST_PATH_IMAGE039
Covariance of
Figure 979404DEST_PATH_IMAGE040
Of a gaussian distribution, coefficients of a gaussian distribution
Figure 938264DEST_PATH_IMAGE041
Satisfy the requirement of
Figure 641778DEST_PATH_IMAGE042
Due to the fact that
Figure 593553DEST_PATH_IMAGE043
Defining the most effective Gaussian distribution of the Gaussian distribution, and taking the mean point of the Gaussian distribution
Figure 331702DEST_PATH_IMAGE044
As the target point of the source search.
CN202210251365.6A 2022-03-15 2022-03-15 Robot gas source searching method based on local path planning Active CN114326755B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210251365.6A CN114326755B (en) 2022-03-15 2022-03-15 Robot gas source searching method based on local path planning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210251365.6A CN114326755B (en) 2022-03-15 2022-03-15 Robot gas source searching method based on local path planning

Publications (2)

Publication Number Publication Date
CN114326755A CN114326755A (en) 2022-04-12
CN114326755B true CN114326755B (en) 2022-06-07

Family

ID=81033570

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210251365.6A Active CN114326755B (en) 2022-03-15 2022-03-15 Robot gas source searching method based on local path planning

Country Status (1)

Country Link
CN (1) CN114326755B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117148872B (en) * 2023-11-01 2024-01-26 中国人民解放军国防科技大学 Robot collaborative source searching method, device and equipment under multi-gas diffusion source scene

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100585332C (en) * 2007-12-24 2010-01-27 河北工业大学 Method for robot independently searching odor source in indoor environment
CN112488153B (en) * 2020-11-06 2022-07-05 中国人民解放军国防科技大学 Autonomous searching and positioning method for odor sources based on clustering tendency principle
CN112231964B (en) * 2020-11-06 2022-07-05 中国人民解放军国防科技大学 Gas leakage source autonomous searching and positioning method based on deep reinforcement learning
CN114131600A (en) * 2021-12-01 2022-03-04 中国人民解放军国防科技大学 Method and system for generating robot source search scheme based on Gaussian mixture model
CN114154383A (en) * 2021-12-01 2022-03-08 中国人民解放军国防科技大学 Multi-robot-source search scheme generation method and system based on cognitive search strategy

Also Published As

Publication number Publication date
CN114326755A (en) 2022-04-12

Similar Documents

Publication Publication Date Title
US10984532B2 (en) Joint deep learning for land cover and land use classification
CN112734775B (en) Image labeling, image semantic segmentation and model training methods and devices
CN111541943B (en) Video processing method, video operation method, device, storage medium and equipment
CN114326755B (en) Robot gas source searching method based on local path planning
CN110222767A (en) Three-dimensional point cloud classification method based on nested neural and grating map
KR20190028242A (en) Method and device for learning neural network
CN111709468B (en) Training method and device for directional artificial intelligence and storage medium
JP7114082B2 (en) Information processing device, information processing method and program
Gleason et al. A fusion approach for tree crown delineation from lidar data.
CN110276363A (en) A kind of birds small target detecting method based on density map estimation
CN111783716A (en) Pedestrian detection method, system and device based on attitude information
Gélard et al. Leaves segmentation in 3d point cloud
JP5531643B2 (en) Passage detection method, apparatus, and program
Datta et al. Skeletonization by a topology-adaptive self-organizing neural network
JP5407897B2 (en) Image classification method, apparatus, and program
Cabrita et al. Odor Guided Exploration and Plume Tracking-Particle Plume Explorer.
CN107462247A (en) A kind of indoor orientation method, device and computer-readable recording medium
JP7114081B2 (en) Information processing device, information processing method and program
Mewes et al. An agent-based extension for object-based image analysis for the delineation of irrigated agriculture from remote sensing data
JP2022053460A (en) Tree discrimination program
Xu et al. Identification of street trees’ main nonphotosynthetic components from mobile laser scanning data
CN110245326A (en) Data estimation method, equipment, storage medium and device neural network based
Humayun et al. Representing vague places: Determining a suitable method
Yu et al. Urban land use classification using street view images based on deep transfer network
Cong et al. Salient man-made object detection based on saliency potential energy for unmanned aerial vehicles remote sensing image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant