1. An unknown environment-oriented multi-radioactive-source online searching method is characterized by comprising the following steps:

the method comprises the following steps that firstly, the position and the corner posture of a robot are initially set according to the range of an unknown area to be detected, the number of radioactive sources in the unknown area to be detected is estimated according to priori knowledge, and the number of layers of multilayer particle swarms is set;

initializing the RRT search tree according to the position of the current robot;

step three, random sampling is carried out in an unknown area to be detected, and the initialized RRT search tree is expanded in the unknown area to be detected through a cost map-based obstacle detection method, so that a sub-node set and branches of the expanded RRT search tree are obtained;

sampling the radiation intensity of the current robot position by adopting an omnidirectional radiation sensor, predicting the state of a radioactive source in an unknown area to be detected by combining a self-adaptive differential evolution-peak suppression particle filter algorithm, and simultaneously calculating the optimal confidence probability corresponding to all layers of particle swarm states;

step five, judging whether the confidence probability calculated in the step four is larger than or equal to the termination confidence probability threshold THR_confAnd searched for area S_expOccupying the area S of the unknown region to be measured_susRatio S of_exp/S_susWhether or not it is greater than or equal to the end-of-search area ratio THR_expIf so, completing the on-line search of the multiple radioactive sources in the unknown environment, otherwise, executing a sixth step;

step six, according to the predicted value of the state of the radioactive source of the unknown area to be detected in the step four, calculating the radiation field information Gain of each sub-node in the sub-node set obtained in the step three_src；

Step seven, the radiation field information Gain of each sub-node is searched through the radiation Gain correction term and the repeated exploration correction term_srcCorrecting;

step eight, screening the radiation field information gains of all the sub-nodes corrected in the step seven according to a sub-node gain maximization criterion to obtain a branch with the maximum radiation field information gain of the sub-nodes, and taking a first sub-node of the branch with the maximum radiation field information gain as a target sampling point of the exploration;

step nine, navigating the robot by using a DWA algorithm, and moving the robot to the searched target sampling point in the step eight; and returning to execute the step two.

2. The unknown environment-oriented online searching method for multiple radioactive sources as claimed in claim 1, wherein in step two, the specific method for initializing the RRT search tree according to the current position of the robot is as follows:

taking the position of the robot as a center, uniformly extending q branches outwards along the circumferential direction, and expanding the RRT search tree to obtain q branches and q sub-nodes; completing initialization of the RRT search tree; wherein q is a positive integer greater than or equal to 12.

3. The unknown-environment-oriented online multi-radiation-source searching method of claim 1, wherein in the third step, random sampling is performed in an unknown area, and the method for obtaining the sub-node set of the expanded RRT search tree by expanding the initialized RRT search tree through an obstacle detection method comprises:

step three, random sampling is carried out in an unknown area to be detected, a sub-node which is most adjacent to a sampling point is searched in an RRT search tree, the sampling point is connected with the nearest sub-node, and a branch path is obtained;

step two, judging whether the branch path has an obstacle or not by adopting an obstacle detection method based on a cost map, if so, giving up the sampling point, and returning to execute the step one; otherwise, executing the third step;

and thirdly, intercepting a fixed step length on the branch path, taking an intercepted point as a child node, realizing one-step expansion of the RRT search tree, judging whether the number of the current child nodes reaches a child node number threshold value, if so, obtaining a child node set, and if not, returning to execute the third step.

4. The unknown-environment-oriented online multi-radiation-source searching method as claimed in claim 1, wherein in step four, the omnidirectional radiation sensor is used to sample the radiation intensity of the current robot position, and the specific method for predicting the radiation source state of the unknown area by combining the adaptive differential evolution-peak suppression particle filter algorithm is as follows:

respectively updating particles of each layer of particle swarm by using the radiation intensity of the current robot position, respectively carrying out center clustering on each layer of particle swarm by using a mean shift clustering algorithm, judging whether the clustering center in each layer of particle swarm meets a central intensity limiting condition, and taking the state of the clustering center meeting the central intensity limiting condition as the predicted state of the radioactive source;

the method for obtaining the state prediction of the radioactive source of the kth layer of particle swarm specifically comprises the following steps:

fourthly, establishing a fitness function by using the radiation intensity of the current robot position according to a differential evolution algorithm, and calculating the fitness of each initial particle in the K-th layer of initial particle swarm by using the fitness function;

performing variation treatment on each initial particle in the particle swarm of the K-th layer through a variation strategy to obtain corresponding variation particles;

step four, performing cross operation on each variation particle to obtain a corresponding test particle:

fourthly, calculating the fitness of each test particle, judging whether the fitness of each test particle is greater than the fitness of the corresponding initial particle, and if so, updating the initial particles in the layer of particle swarm into the corresponding test particles; completing one updating of the particle swarm state of the layer;

step four, judging whether the updating times of the particle swarm state reach an iteration time threshold value, if so, executing step four and step six, otherwise, taking the particle swarm updated in step four as an initial particle swarm, and returning to execute step four;

fourthly, carrying out mean shift clustering on the updated particle swarm, and judging whether a clustering center meets a center strength limiting condition; if so, taking the position of the clustering center and the radiation intensity of the position as the predicted value of the radiation source state; otherwise, no radioactive source exists in the K particle swarm.

5. The unknown environment-oriented multi-radiation-source online searching method according to claim 4, wherein in the fourth step, in the first step, the fitness function is:

w(p_i)＝w_obs(m(S_h),p_i,C_-k)·w_ps(p_i,θ_ps)·w_dist(p_i,C_-k) Formula one

Wherein, w (p)_i) Is a particle p_iThe corresponding integrated particle weight is the initial particle p_iFitness of w_obs(m(S_h),p_i,C_-k) Is an observation weight term of the particle, w_ps(p_i,θ_ps) For peak suppression correction terms of the particles, w_dist(p_i,C_-k) Correction of the inter-cluster distance of the particles by a term m (S)_h) For the intensity of the radiation, p, sampled at the h-th time_iIs the ith initial particle, C in the particle swarm_-kAs a cluster center not containing the k-th particle group, θ_psIs the peak suppression center;

observation weight term of particle:

wherein p (-) represents the probability function of the Poisson 'S observation model, I' (S)_h,p_i,C_-k) At the h-th sampling point S_hThe predicted radiation intensity of the superposition is processed,is a rounding operation.

6. The method of claim 5, wherein in the second step, the variant particle is obtained by calculating according to the following formula:

wherein the content of the first and second substances,updating variation particles of the current-layer particle swarm in the iteration process of the g +1 th update time; k is the number of layers of the particles,is the initial particle in the current layer particle group for the g-th iteration,for the g-th iteration, the maximum fitness particles in the current layer particle swarm,random particles in the current layer of particle swarm are taken;F₁and F₂All are adaptive variation rates, alpha is the moving scale of the elite,is composed ofCorresponding particle weight, beta is a random movement scale,is the particle weight average.

7. The method of claim 6, wherein in step four or three, each variant particle is cross-operated to obtain a test particle by calculating according to the following formula:

wherein the content of the first and second substances,the kth layer grain in the g +1 iteration processTest particles of a sub-population;

wherein CR is a cross rate adaptively adjusted based on particle weight, sigmaⁱWhite Gaussian noise, CR, with mean value of 0_baseFor the base value of the crossing rate, CR_scaleFor the cross-rate coefficient based on the weight of the particles,is the particle weight average.

8. The unknown-environment-oriented online multi-radiation-source searching method according to claim 7, wherein in the fourth step, the optimal confidence probability corresponding to the current particle swarm state is:

in the formula (I), the compound is shown in the specification,for the optimal confidence probability corresponding to the current particle swarm state,is the cluster center of all layer particle groups,as a cluster centerAt the sampling point S_hConfidence probability of (C), N_kIs the total number of particle layers, N_hP (-) represents the probability function of the Poisson observation model for the total number of times the robot samples the radiation,is at the same timeThe predicted radiation intensity of the superposition is processed,for the h-th sampling point S_hThe position of the corresponding position is determined,is a rounding operation.

9. The unknown-environment-oriented online searching method for multiple radioactive sources as claimed in claim 8, wherein in step six, the radiation field information Gain of each sub-node in the sub-node set_srcComprises the following steps:

wherein the content of the first and second substances,single point radiation gain function for the jth radiation source for the child node, N_jThe total number of radiation sources in the prediction of radiation source status for the unknown region,an inter-source correction factor of the jth radioactive source is given to the child node, and is used for relieving the influence caused by multi-source benefit superposition;

wherein the content of the first and second substances,searching the mth child node n in the tree for the RRT_mTo the jth radiation source A_jNormalized distance of (d);

wherein the content of the first and second substances,searching for child node n in tree for RRT_mThe location information of (a) is stored in the storage unit,to predict the radiation source A_jThe location information of (a);normalizing the distance scale factor for radiation gain, H_srcThe distance offset is normalized for the radiation gain,

gain superposition effect correction factor

Wherein the content of the first and second substances,the maximum distance ratio from the mth child node to the nth and jth radioactive sources is defined, and the minimum value is 1; lambda [ alpha ]²Is a scale factor for adjusting the multi-source benefit superposition effect;

wherein the content of the first and second substances,the normalized distance from the mth child node to the nth radiation source in the RRT is searched.

10. The on-line multi-radiation-source searching method for unknown environment of claim 9 wherein in step seven, the radiation gain correction term C is used_radAnd repeatedly exploring the correction term C_odomRadiation field information Gain for each sub-node_srcThe correction method comprises the following steps:

computing a repeat search correction term C_odom:

Wherein, C_odom(n_m) Is a node n_mRepeated exploration of correction term, N_odomTotal number of sampling points for the robot history, n_mSearching the mth child node in the tree for the RRT,is the m-th sub-node n_mNormalized distance to the ith sample point;

wherein the content of the first and second substances,is the position of the ith sample point, ξ_odomCorrection of the scale factor of the normalized distance for repeated exploration, H_odomNormalized distance offset corrected for repeated exploration;

calculating a radiation gain correction term C_rad；

Wherein the content of the first and second substances,is equal to n_mThe radiation sample value at the node closest to the sample point,for ambient background radiation, theta_radFor horizontal offset of radiation gain correction curve, b_radThe radiation gain correction scale parameter is used for controlling the change speed of the correction effect, eta is the background value after the radiation gain correction, and the lower limit value after the correction is controlled;

modified radiation information gainComprises the following steps:

Background

The radioactive source is a generic term for a radiation source composed of radioactive substances, and the intensity of the radiation source decreases in space according to the rule of distance quadratic decrease. Because the radioactive source has the characteristics of high carrying energy, obvious field superposition, statistical fluctuation and the like, the sampling and prediction of the radioactive source under the condition of unknown quantity and intensity become the difficulty of exploring the radiation environment. Compared with a manual detection mode, the method has remarkable safety advantages of completing radiation detection and prediction on line by the robot. In addition, compared with the traditional identification method using a sensor array or a special sensor combination (gamma camera and radiation spectrometer), the method has higher intelligence and practical value based on the problems of sparse sampling, online prediction, number non-parameter estimation and the like of the omnidirectional radiation sensor carried by the robot. The method is a key problem of the robot for autonomously searching for multiple radioactive source tasks by performing multi-point radioactive source state estimation and making an effective exploration strategy on a suspicious environment only by means of sensor data carried by the robot.

Compared with the search of a single radioactive source in a known environment, the current research for searching a multi-point radioactive source on line comprises the following difficulties:

(1) due to the accumulated radiation effect in the multi-peak radiation field and the measurement characteristics of the GM counter, the airborne sensor can only obtain the accumulated radiation dose rate of the current position containing background radiation, but based on the coupled radiation data, the traditional nonparametric prediction algorithm is easy to obtain an error estimation result.

(2) Because the task scene is an unknown environment with multiple radioactive sources, the method which only depends on the search of a single radioactive source cannot finish high-efficiency sampling on the superposed field.

(3) Considering that the number of radioactive sources in a local suspected area is unknown, the source neighborhood needs to be heavily sampled, so that the method is not applicable to finish exploration by simply using the area ratio of an explored area.

In conclusion, the detection of the unknown space multi-radiation source is poor in accuracy and low in efficiency only by taking the area ratio of the searched area as the basis for finishing the search.

Disclosure of Invention

The invention aims to solve the problems of poor accuracy and low efficiency of multi-radioactive source detection in the existing unknown space, and provides an unknown environment-oriented multi-radioactive source online searching method.

The invention relates to an unknown environment-oriented multi-radioactive-source online searching method, which comprises the following steps:

the method comprises the following steps:

the method comprises the following steps that firstly, the position and the corner posture of a robot are initially set according to the range of an unknown area to be detected, the number of radioactive sources in the unknown area to be detected is estimated according to priori knowledge, and the number of layers of multilayer particle swarms is set;

initializing the RRT search tree according to the position of the current robot;

step three, random sampling is carried out in an unknown area to be detected, and the initialized RRT search tree is expanded in the unknown area to be detected through a cost map-based obstacle detection method, so that a sub-node set and branches of the expanded RRT search tree are obtained;

sampling the radiation intensity of the current robot position by adopting an omnidirectional radiation sensor, predicting the state of a radioactive source in an unknown area to be detected by combining a self-adaptive differential evolution-peak suppression particle filter algorithm, and simultaneously calculating the optimal confidence probability corresponding to all layers of particle swarm states;

step five, judging whether the confidence probability calculated in the step four is larger than or equal to the termination confidence probability threshold THR_confAnd searched for area S_expOccupying the area S of the unknown region to be measured_susRatio S of_exp/S_susWhether or not it is greater than or equal to the end-of-search area ratio THR_expIf so, completing the on-line search of the multiple radioactive sources in the unknown environment, otherwise, executing a sixth step;

step six, according to step four to be measuredAnd C, calculating the predicted value of the state of the radioactive source of the unknown area, and calculating the radiation field information Gain of each sub-node in the sub-node set obtained in the step three_src；

Step seven, the radiation field information Gain of each sub-node is searched through the radiation Gain correction term and the repeated exploration correction term_srcCorrecting;

step eight, screening the radiation field information gains of all the sub-nodes corrected in the step seven according to a sub-node gain maximization criterion to obtain a branch with the maximum radiation field information gain of the sub-nodes, and taking a first sub-node of the branch with the maximum radiation field information gain as a target sampling point of the exploration;

step nine, navigating the robot by using a DWA algorithm, and moving the robot to the searched target sampling point in the step eight; and returning to execute the step two.

Further, in the second step of the present invention, a specific method for initializing the RRT search tree according to the current position of the robot is as follows:

taking the position of the robot as a center, uniformly extending q branches outwards along the circumferential direction, and expanding the RRT search tree to obtain q branches and q sub-nodes; completing initialization of the RRT search tree; wherein q is a positive integer greater than or equal to 12.

Further, in the third step of the present invention, random sampling is performed in an unknown area, and the initialized RRT search tree is expanded by an obstacle detection method, and a specific method for obtaining a sub-node set of the expanded RRT search tree is as follows:

step three, random sampling is carried out in an unknown area to be detected, a sub-node which is most adjacent to a sampling point is searched in an RRT search tree, the sampling point is connected with the nearest sub-node, and a branch path is obtained;

step two, judging whether the branch path has an obstacle or not by adopting an obstacle detection method based on a cost map, if so, giving up the sampling point, and returning to execute the step one; otherwise, executing the third step;

and thirdly, intercepting a fixed step length on the branch path, taking an intercepted point as a child node, realizing one-step expansion of the RRT search tree, judging whether the number of the current child nodes reaches a child node number threshold value, if so, obtaining a child node set, and if not, returning to execute the third step.

Furthermore, in the fourth step of the present invention, the specific method for predicting the radiation source state of the unknown area by using the omnidirectional radiation sensor to sample the radiation intensity of the current robot position and combining the adaptive differential evolution-peak suppression particle filter algorithm is as follows:

respectively and simultaneously updating particles of each layer of particle swarm by using the radiation intensity of the current robot position, respectively carrying out center clustering on each layer of particle swarm by using a mean shift clustering algorithm, judging whether the clustering center in each layer of particle swarm meets a central intensity limiting condition, and taking the state of the clustering center meeting the central intensity limiting condition as the predicted state of the radioactive source;

the method for obtaining the state prediction of the radioactive source of the kth layer of particle swarm specifically comprises the following steps:

fourthly, establishing a fitness function by using the radiation intensity of the current robot position according to a differential evolution algorithm, and calculating the fitness of each initial particle in the K-th layer of initial particle swarm by using the fitness function;

performing variation treatment on each initial particle in the particle swarm of the K-th layer through a variation strategy to obtain corresponding variation particles;

step four, performing cross operation on each variation particle to obtain a corresponding test particle:

fourthly, calculating the fitness of each test particle, judging whether the fitness of each test particle is greater than the fitness of the corresponding initial particle, and if so, updating the initial particles in the layer of particle swarm into the corresponding test particles; completing one updating of the particle swarm state of the layer;

step four, judging whether the updating times of the particle swarm state reach an iteration time threshold value or not, if so, executing step four and step six, otherwise, taking the particle swarm updated in step four as an initial particle swarm, and returning to execute step four;

fourthly, carrying out mean shift clustering on the updated particle swarm, and judging whether a clustering center meets a center strength limiting condition; if so, taking the position of the clustering center and the radiation intensity of the position as the predicted value of the radiation source state; otherwise, no radioactive source exists in the K particle swarm.

Further, in the fourth step, the fitness function is:

w(p_i)＝w_obs(m(S_h),p_i,C_-k)·w_ps(p_i,θ_ps)·w_dist(p_i,C_-k) Formula one

Wherein, w (p)_i) Is a particle p_iThe corresponding integrated particle weight is the initial particle p_iThe degree of fitness of (a) to (b),

w_obs(m(S_h),p_i,C_-k) Is an observation weight term of the particle, w_ps(p_i,θ_ps) For peak suppression correction terms of the particles, w_dist(p_i,C_-k) Correction of the inter-cluster distance of the particles by a term m (S)_h) For the intensity of the radiation, p, sampled at the h-th time_iIs the ith initial particle, C in the particle swarm_-kIs a cluster center, theta, not containing the k-th particle group_psIs the peak suppression center;

observation weight term of particle:

where p (-) represents the probability function of the Poisson's observation model, I' (p)_i,C_-k) According to the initial particle p_iAnd C_-kThe calculated cumulative predicted radiation intensity is,is a rounding operation.

Further, in the present invention, in the second step, obtaining the variant particle is calculated by the following formula:

wherein the content of the first and second substances,updating variation particles of the current-layer particle swarm in the iteration process of the g +1 th update time; k is the number of layers of the particles,is the initial particle in the current layer particle group for the g-th iteration,for the g-th iteration, the maximum fitness particles in the current layer particle swarm,random particles in the current layer of particle swarm are taken; F₁and F₂All are adaptive variation rates, alpha is the movement scale of the elite and the distance of the movement to the optimal particle,is composed ofThe corresponding weight of the particle, beta is the random movement scale, the distance moved to the random particle,is the particle weight average.

Further, in the invention, in the third step, each variant particle is subjected to cross operation, and the obtained test particle is calculated by the following formula:

wherein the content of the first and second substances,test particles of the kth layer of particle swarm in the g +1 iteration process;

CR is the crossing rate, sigma, adaptively adjusted based on the weight of the particlesⁱWhite Gaussian noise, CR, with mean value of 0_baseFor the base value of the crossing rate, CR_scaleFor the cross-rate coefficient based on the weight of the particles,is the particle weight average.

Further, in the fourth step of the present invention, the optimal confidence probability corresponding to the current particle swarm state is:

in the formula (I), the compound is shown in the specification,for the optimal confidence probability corresponding to the current particle swarm state,is the cluster center of all layer particle groups,as a cluster centerIn the samplingPoint S_hConfidence probability of (C), N_kIs the total number of particle layers, N_hP (-) represents the probability function of the Poisson observation model for the total number of times the robot samples the radiation,is at the same timePredicted radiation intensity of the superposition, from multi-layer clustering centerIt is determined that,for the h-th sampling point S_hThe position of the corresponding position is determined,is a rounding operation.

Furthermore, in the sixth step, in the present invention, the radiation field information Gain of each sub-node in the sub-node set_srcComprises the following steps:

wherein the content of the first and second substances,single point radiation gain function for the jth radiation source, N, for a node_jThe total number of radiation sources in the prediction of radiation source status for the unknown region,an inter-source correction factor of the jth radioactive source is calculated for the node, and is used for relieving the influence caused by multi-source benefit superposition;

wherein the content of the first and second substances,for the mth node n in RRT tree_mTo the jth radiation source A_jNormalized distance of (d);

wherein the content of the first and second substances,searching node n in tree for RRT_mIncluding x and y axis coordinates (m);to predict the radiation source A_jIncluding x and y axis coordinates (m);normalizing the distance scale factor for the radiation gain, and determining according to the exploration area range and the positioning precision of the radioactive source; h_srcNormalizing the distance offset for the radiation gain to prevent the situation of infinite numerical value caused by over-small distance;

gain superposition effect correction factor

Wherein the content of the first and second substances,the maximum distance ratio from the mth node to the nth and jth radioactive sources is defined, and the minimum value is 1; lambda [ alpha ]²Is a scale factor for adjusting the multi-source benefit superposition effect;

wherein the content of the first and second substances,the normalized distance from the mth child node to the nth radiation source in the RRT is searched.

Further, in the present invention, in step seven, the radiation gain correction term C is used_radAnd repeatedly exploring the correction term C_odomRadiation field information Gain for each sub-node_srcThe correction method comprises the following steps:

computing a repeat search correction term C_odom:

Wherein N is_odomTotal number of sampling points for the robot history, n_mFor the mth node in the RRT spanning tree,for the mth node n_mNormalized distance to the ith sample point;

wherein n is_mFor the mth node in the RRT spanning tree,is the position of the ith sample point, ξ_odomCorrection of the scale factor of the normalized distance for repeated exploration, H_odomCorrecting the offset of the normalized distance for repeated exploration;

calculating a radiation gain correction term C_rad；

Wherein the content of the first and second substances,is equal to n_mThe radiation sample value at the node closest to the sample point,for ambient background radiation, theta_radFor horizontal offset of radiation gain correction curve, b_radThe radiation gain correction scale parameter is used for controlling the change speed of the correction effect, eta is the background value after the radiation gain correction, and the lower limit value after the correction is controlled;

modified radiation information gainComprises the following steps:

compared with the existing method for estimating the parameters of the radioactive source, the method has the advantages that the consumed time and the particle swarm number are in a linear relation in each iteration of the measurement set, so that the problem of dimension disaster dilemma caused by the increase of the number of the radioactive sources is avoided, and the high efficiency of the online prediction method is reflected. In addition, due to the fact that a self-adaptive differential evolution idea is introduced on the basis of an original PSPF algorithm, the method not only reduces the number of particles of each group, but also improves the accuracy of prediction. Moreover, the portable radiation sensor of the robot adopts the Geiger-Muller tube which has light weight, low cost, small volume, high efficiency and wide use as the gas ionization detector, is easy to be applied to a small unmanned exploration platform and can be widely applied to various mobile detection tasks aiming at multiple radioactive sources.

Drawings

FIG. 1(a) is a schematic diagram of the state and sampled position of each layer of particle swarm after 1 st exploration sampling;

FIG. 1(b) is a schematic diagram of the state and sampled position of each layer of particle swarm after 5 th exploration sampling;

FIG. 1(c) is a schematic diagram illustrating the states and sampled positions of the particle groups in each layer after the 15 th exploration sampling;

FIG. 1(d) is a schematic diagram illustrating the states and sampled positions of the particle groups in each layer after the 25 th exploration sampling;

FIG. 1(e) is a schematic diagram illustrating states and sampled positions of particle groups in each layer after 35 th exploration sampling;

FIG. 1(f) is a schematic diagram of the state and sampled position of each layer of particle swarm after 45 th exploration sampling;

FIG. 1(g) is a schematic diagram of the states and sampled positions of the particle groups in each layer after the 55 th exploration sampling;

FIG. 1(h) is a diagram illustrating states and sampled positions of particle groups in each layer after the 65 th exploration sampling;

FIG. 1(i) is a schematic diagram of the state and sampled position of each layer of particle group after 75 th exploration sampling;

FIG. 1(j) is a schematic diagram of the state and sampled position of each layer of particle swarm after 85 th exploration sampling;

FIG. 1(k) is a schematic diagram of the state and sampled position of each layer of particle swarm after the 95 th exploration sampling;

FIG. 1(l) is a diagram illustrating the states and sampled positions of particle groups in each layer after the 105 th exploration sampling;

FIG. 2(a) trajectory of the robot and constructed map after 1 st exploration;

FIG. 2(b) the trajectory of the robot and the constructed map after the 5 th exploration;

FIG. 2(c) trajectory of the robot and constructed map after the 15 th exploration;

FIG. 2(d) trajectory of the robot and constructed map after the 25 th exploration;

FIG. 2(e) the trajectory of the robot and the constructed map after the 35 th exploration;

FIG. 2(f) trajectory of the robot and constructed map after the 45 th exploration;

FIG. 2(g) trajectory of the robot and constructed map after the 55 th exploration;

FIG. 2(h) trajectory of the robot and constructed map after the 65 th exploration;

FIG. 2(i) trajectory of the robot and constructed map after 75 th exploration;

FIG. 2(j) trajectory of the robot and constructed map after the 85 th exploration;

FIG. 2(k) trajectory of the robot and constructed map after the 95 th exploration;

FIG. 2(l) trajectory of the robot and constructed map after the 105 th exploration;

FIG. 3 is a diagram illustrating a robot exploration trajectory and multi-source prediction results;

FIG. 4 is a diagram of a predicted position error curve of a radioactive source under a four-point radioactive source scene;

FIG. 5 is a graph of the error in the predicted intensity of a radioactive source under a four-point radioactive source scenario;

FIG. 6 is a graph of the run-time distribution of the search process for a four-point radioactive source scenario.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.

The first embodiment is as follows: the embodiment provides an unknown environment-oriented multi-radioactive-source online searching method, which comprises the following steps:

the method comprises the following steps that firstly, the position and the corner posture of a robot are initially set according to the range of an unknown area to be detected, the number of radioactive sources in the unknown area to be detected is estimated according to priori knowledge, and the number of layers of multilayer particle swarms is set;

initializing the RRT search tree according to the position of the current robot;

step three, random sampling is carried out in an unknown area to be detected, and the initialized RRT search tree is expanded in the unknown area to be detected through a cost map-based obstacle detection method, so that a sub-node set and branches of the expanded RRT search tree are obtained;

sampling the radiation intensity of the current robot position by adopting an omnidirectional radiation sensor, predicting the state of a radioactive source in an unknown area to be detected by combining a self-adaptive differential evolution-peak suppression particle filter algorithm, and simultaneously calculating the optimal confidence probability corresponding to all layers of particle swarm states;

step five, judging whether the confidence probability calculated in the step four is larger than or equal to the termination confidence probability threshold THR_confAnd searched for area S_expOccupying the area S of the unknown region to be measured_susRatio S of_exp/S_susWhether or not it is greater than or equal to the end-of-search area ratio THR_expIf so, completing the on-line search of the multiple radioactive sources in the unknown environment, otherwise, executing a sixth step;

wherein the termination confidence probability threshold THR_confThe range of (A) is as follows: 0.85-0.95, area ratio threshold THR_expThe range of (A) is as follows: 0.9-0.95.

Step six, according to the predicted value of the state of the radioactive source of the unknown area to be detected in the step four, calculating the radiation field information Gain of each sub-node in the sub-node set obtained in the step three_src；

Step seven, the radiation field information Gain of each sub-node is searched through the radiation Gain correction term and the repeated exploration correction term_srcCorrecting;

step eight, screening the radiation field information gains of all the sub-nodes corrected in the step seven according to a sub-node gain maximization criterion to obtain a branch with the maximum radiation field information gain of the sub-nodes, and taking a first sub-node of the branch with the maximum radiation field information gain as a target sampling point of the exploration;

step nine, navigating the robot by using a DWA algorithm, and moving the robot to the searched target sampling point in the step eight; and returning to execute the step two.

In the present embodiment, the area that has been searched for is determined based on the position of the target search point and the acquisition range of the sensor. In the present embodiment, in the second step, the fixed step length is determined according to the area of the suspected area, and is usually within a range of 0.5 m to 1.5 m.

In the first step, the method for setting the number of layers of the multilayer particle swarm comprises the following steps:

and setting the number of layers of the multilayer particle swarm to be more than or equal to the estimated value of the number of radioactive sources in the unknown area. From a priori knowledge such as: the number of missing radioactive sources or the area of the suspected area, the number of possible radioactive sources is inferred, and the number of particle groups is then made larger than this number.

Further, in the present embodiment, in the second step, a specific method for initializing the RRT search tree according to the current position of the robot is as follows:

taking the position of the robot as a center, uniformly extending q branches outwards along the circumferential direction, and expanding the RRT search tree to obtain q branches and q sub-nodes; completing initialization of the RRT search tree; wherein q is a positive integer greater than or equal to 12.

Further, in the third step of the present embodiment, the random sampling is performed in the unknown area, and the initialized RRT search tree is expanded by the obstacle detection method, so as to obtain the sub-node set of the expanded RRT search tree, specifically, the method includes:

step three, random sampling is carried out in an unknown area to be detected, a sub-node which is most adjacent to a sampling point is searched in an RRT search tree, the sampling point is connected with the nearest sub-node, and a branch path is obtained;

step two, judging whether the branch path has an obstacle or not by adopting an obstacle detection method based on a cost map, if so, giving up the sampling point, and returning to execute the step one; otherwise, executing the third step;

and thirdly, intercepting a fixed step length on the branch path, taking an intercepted point as a child node, realizing one-step expansion of the RRT search tree, judging whether the number of the current child nodes reaches a child node number threshold value, if so, obtaining a child node set, and if not, returning to execute the third step.

In this embodiment, the process of randomly sampling in the unknown region to be detected is to search for the position region to be detected through the omnidirectional radiation sensor on the robot, randomly sample in the region obtained by the search, and extend the search result in the manner of intercepting the fixed step length.

Further, in the fourth embodiment, in step four, the specific method for predicting the states of the radiation sources in all layers of particle swarm by sampling the radiation intensity of the current robot position by using the omnidirectional radiation sensor and combining the adaptive differential evolution-peak suppression particle filter algorithm includes:

respectively updating particles of each layer of particle swarm by using the radiation intensity of the current robot position, respectively carrying out center clustering on each layer of particle swarm by using a mean shift clustering algorithm, judging whether the clustering center in each layer of particle swarm meets a central intensity limiting condition, and taking the state of the clustering center meeting the central intensity limiting condition as the predicted state of the radioactive source;

the method for obtaining the state prediction of the radioactive source of the kth layer of particle swarm specifically comprises the following steps:

fourthly, establishing a fitness function by using the radiation intensity of the current robot position according to a differential evolution algorithm, and calculating the fitness of each initial particle in the kth layer of particle swarm by using the fitness function;

performing variation treatment on each initial particle in the particle swarm of the K-th layer through a variation strategy to obtain corresponding variation particles;

step four, performing cross operation on each variation particle to obtain a corresponding test particle:

fourthly, calculating the fitness of each test particle, judging whether the fitness of each test particle is greater than the fitness of the corresponding initial particle, and if so, updating the initial particles in the layer of particle swarm into the corresponding test particles; completing one updating of the particle swarm state of the layer;

step four, judging whether the updating times of the particle swarm state reach an iteration time threshold value, if so, executing step four and step six, otherwise, taking the particle swarm updated in step four as an initial particle swarm, and returning to execute step four; the number threshold of the update iterations is typically 5-10;

fourthly, carrying out mean shift clustering on the updated particle swarm, and judging whether a clustering center meets a center strength limiting condition; if so, taking the position of the clustering center and the radiation intensity of the position as the predicted value of the radiation source state; otherwise, no radioactive source exists in the K particle swarm.

Further, in the present embodiment, in the first step, the fitness function is:

w(p_i)＝w_obs(m(S_h),p_i,C_-k)·w_ps(p_i,θ_ps)·w_dist(p_i,C_-k) Formula one

Wherein, w (p)_i) Is a particle p_iThe corresponding integrated particle weight is the initial particle p_iThe degree of fitness of (a) to (b),

w_obs(m(S_h),p_i,C_-k) Is an observation weight term of the particle, w_ps(p_i,θ_ps) For peak suppression correction terms of the particles, w_dist(p_i,C_-k) Correction of the inter-cluster distance of the particles by a term m (S)_h) For the intensity of the radiation, p, sampled at the h-th time_iIs the ith initial particle, C in the particle swarm_-kAs a cluster center not containing the k-th particle group, θ_psIs the peak suppression center;

observation weight term of particle:

wherein p (-) represents the probability function of the Poisson 'S observation model, I' (S)_h,p_i,C_-k) At the h-th sampling point S_hAt the superimposed predicted radiation intensity, from the particles p_iAnd in clusteringHeart C_-kIt is determined that,is a rounding operation.

Further, in the present embodiment, in the second step, obtaining the variant particle is calculated by the following formula:

wherein the content of the first and second substances,updating variation particles of the current-layer particle swarm in the iteration process of the g +1 th update time; k is the number of layers of the particles,is the initial particle in the current layer particle group for the g-th iteration,for the g-th iteration, the maximum fitness particles in the current layer particle swarm,random particles in the current layer of particle swarm are taken; F₁and F₂All are adaptive variation rates, alpha is the movement scale of the elite and the distance of the movement to the optimal particle,is composed ofCorresponding weight of particle, beta being the scale of random movement, moving towards random particleThe distance between the first and second electrodes,is the particle weight average.

Further, in the present embodiment, in the step four and three, the cross operation is performed on each variant particle to obtain the test particle, which is calculated by the following formula:

wherein the content of the first and second substances,test particles of the kth layer of particle swarm in the g +1 iteration process;

wherein CR is a cross rate adaptively adjusted based on particle weight, sigmaⁱWhite Gaussian noise, CR, with mean value of 0_baseFor the base value of the crossing rate, CR_scaleFor the cross-rate coefficient based on the weight of the particles,is the particle weight average.

Further, in the fourth step, the optimal confidence probability corresponding to the current particle swarm state is:

in the formula (I), the compound is shown in the specification,for the optimal confidence probability corresponding to the current particle swarm state,is the cluster center of all layer particle groups,as a cluster centerAt the sampling point S_hConfidence probability of (C), N_kIs the total number of particle layers, N_hP (-) represents the probability function of the Poisson observation model for the total number of times the robot samples the radiation,is at the same timePredicted radiation intensity of the superposition, from multi-layer clustering centerIt is determined that,for the h-th sampling point S_hThe position of the corresponding position is determined,is a rounding operation.

Further, in this embodiment, in step six, the radiation field information Gain of each sub-node in the sub-node set_srcComprises the following steps:

wherein the content of the first and second substances,single point radiation gain function for the jth radiation source for the child node, N_jThe total number of radiation sources in the prediction of radiation source status for the unknown region,an inter-source correction factor of the jth radioactive source is given to the child node, and is used for relieving the influence caused by multi-source benefit superposition;

wherein the content of the first and second substances,searching the mth child node n in the tree for the RRT_mTo the jth radiation source A_jNormalized distance of (d);

wherein the content of the first and second substances,searching for child node n in tree for RRT_mThe location information of (a) is stored in the storage unit,to predict the radiation source A_jThe location information of (a);normalizing the distance scale factor for radiation gain, H_srcNormalizing the distance offset for the radiation gain, wherein the offset is usually a tiny offset to prevent the situation of infinite numerical value caused by over-small distance;

gain superposition effect correction factor

Wherein the content of the first and second substances,the maximum distance ratio from the mth child node to the nth and jth radioactive sources is defined, and the minimum value is 1; lambda [ alpha ]²Is a scale factor for adjusting the multi-source benefit superposition effect;

and the normalized distance from the mth child node to the nth radiation source in the RRT search tree.

Furthermore, in the present embodiment, in step seven, the radiation gain correction term C is used_radAnd repeatedly exploring the correction term C_odomRadiation field information Gain for each sub-node_srcThe correction method comprises the following steps:

computing a repeat search correction term C_odom:

Wherein N is_odomTotal number of sampling points for the robot history, n_mSearching the mth child node in the tree for the RRT,is the m-th sub-node n_mNormalized distance to the ith sample point;

wherein n is_mThe mth node in the tree is searched for the RRT,is the position of the ith sample point, ξ_odomCorrection of the scale factor of the normalized distance for repeated exploration, H_odomNormalized distance offset corrected for repeated exploration;

calculating a radiation gain correction term C_rad；

Wherein the content of the first and second substances,is equal to n_mThe radiation sample value at the point closest to the sample point,for ambient background radiation, theta_radFor horizontal offset of radiation gain correction curve, b_radThe radiation gain correction scale parameter is used for controlling the change speed of the correction effect, eta is the background value after the radiation gain correction, and the lower limit value after the correction is controlled;

modified radiation information gainComprises the following steps:

firstly, the omnidirectional ionizing radiation sensor cannot obtain gradient direction information of a multi-source superposed radiation field, secondly, a simple unknown region exploration strategy is not applicable to the problem of online prediction and sampling of multiple radioactive sources, and finally, due to the fact that the number of the radioactive sources is unknown, an exploration process can be iteratively carried out only in a prediction-sampling mode, and if the exploration criterion is not ended, the robot carries out redundant sampling. Therefore, the invention provides an integrated exploration framework for online multi-radiation-source prediction and efficient sampling in an unknown environment by only utilizing the accumulated radiation data of sparse omnidirectional gamma rays.

The specific embodiment is as follows:

the characteristics of the radioactive source search scene corresponding to the experiment are as follows: 1. global suspicious region exploration area 21m × 21m, 2 radiation dose rate range: 0-1500 nGy/h, 3. expanding the tree node number: 50, 4. number of particle group: 5, 5. number of particles in single particle group: 150, 6 true number of radioactive sources: 4, 7, initial exploration pose: (-3.0m,1.8m,57 °). The four-point source scene constructs a multi-peak radiation field with two radiation sources as a group, as shown in fig. 3. The flow chart of the method of the invention is shown in figure 1, and the implementation process is explained in detail as follows:

step one, initializing an expansion tree (RRT search tree) according to the current pose state of the mobile platform, wherein the configuration space of the expansion tree comprises the plane position and the corner posture of the robot. In the process, in addition to adding the optimal path in the last planning to the current expansion tree, vector nodes uniformly extending from the circumferential direction of the current position need to be added to the expansion tree;

and secondly, randomly sampling in an obtained barrier-free space explored by the robot in an unknown area, searching nearest neighbor points of sampling points in a search tree, then determining a configuration relation between nodes, selecting proper state points through barrier detection and a maximum distance threshold value, and constructing an expansion tree on the basis of an initialization state.

And step three, predicting the radioactive source state of the suspected area by combining an ADE-PSPF algorithm according to the existing radiation sampling set. The ADE-PSPF algorithm is an improved algorithm based on the PSPF algorithm. The main difference is that in the resampling link, the ADE-PSPF algorithm takes the particles in each group as the initial population of the differential evolution.

(1) The fitness function is shown in equation (1). Wherein, w_obsIs the observed weight of the particle, w_psFor peak suppression correction terms of the particles, w_distThe term is corrected for the inter-cluster distance of the particles.

w(p_i)＝w_obs(m(S_h),p_i,C_-k)·w_ps(p_i,θ_ps)·w_dist(p_i,C_-k) (1)

Wherein, w (p)_i) Is a particle p_iThe corresponding integrated particle weight is the initial particle p_iThe degree of fitness of (a) to (b),

w_obs(m(S_h),p_i,C_-k) Is an observation weight term of the particle, w_ps(p_i,θ_ps) For peak suppression correction terms of the particles, w_dist(p_i,C_-k) Correction of the inter-cluster distance of the particles by a term m (S)_h) For the intensity of the radiation, p, sampled at the h-th time_iIs the ith initial particle, C in the particle swarm_-kAs a cluster center not containing the k-th particle group, θ_psIs the peak suppression center;

(2) and (3) carrying out mutation treatment on the particles in each group by using the mutation strategy of the formula (2) to obtain a mutation individual. Wherein the content of the first and second substances,as the initial particles, the particles are,for the maximum fitness particle in the current population,are random particles.Andfor adaptive variation rate, alpha is the movement scale of the elite, the distance of movement to the optimal particle is controlled,is composed ofCorresponding particle weight, beta is a random movement scale, the distance of movement to the random particles is controlled,is the particle weight average.

(3) The test individuals generated by the crossover operation were obtained by the formula (3).

Wherein the content of the first and second substances,to adaptively adjust the crossover rate based on the particle weight, the impact of the mutation operation increases as the particle weight increases. To avoid particle depletion, a Gaussian noise σ with mean 0 is superimposed on a target-individual basisⁱ。

(4) According to a formula (4), comparing the fitness of the initial particles with the fitness of the corresponding test individual, and selecting the individual with higher fitness as the particles in the particle swarm;

(5) and (5) recalculating and normalizing the particle weight according to the formula (5) to obtain a particle weight mean value.

Pn is the number of particles in a single-layer particle swarm;

and fourthly, identifying the particle centers in the aggregation state in each layer of particle swarm through a Mean-Shift clustering (Mean Shift clustering) algorithm and a judgment criterion, and realizing parameter estimation of current state distribution.

Selecting the maximum proportion of the clustering states for screening according to the clustering centers of all layers of particle swarms and the corresponding particle proportions, designing a screening criterion to judge the prediction states of all layers of particle swarms, wherein the screening criterion can be expressed as:

in the above formula, the first and second carbon atoms are,to be the degree of clustering,for the corresponding particle clustering ratio, THR_strAnd THR_prRespectively representing a screening intensity threshold value and a particle clustering proportion threshold value;

step two, calculating the integral confidence probability of the current radiation field by utilizing the estimated parameters of the radioactive source in the step one, and calculating the Poisson observation mean value of the current prediction state relative to the actual measurement setConfidence probability as exploration areaThe calculation formula is as follows:

wherein p (-) represents Poisson Observation model, I'_cum(. represents a multipoint predicted stateIn thatThe evaluation model also uses the maximum weight of each measurement positionAs denominators, the normalization problem of the weight range of the Poisson evaluation model is solved, and based on the confidence probability, the prediction framework judges whether to store the optimal prediction configuration;

step five, judging step fourCalculated confidence probabilityWhether or not greater than the historical optimum confidence probabilityIf yes, updating the historical optimal confidence probabilityAnd storing the corresponding particle swarm state and the state of the clustering center; otherwise, the states of each particle swarm and the cluster center are restored to the configuration with the optimal history.

Step six, combining the current radiation source prediction state with the sampling point set obtained in the step two, and calculating the radiation field Gain corresponding to each point by using a formula (6)_src。

Wherein beta is a radiation benefit proportion coefficient used for coordinating the proportion of geometric benefits.Is a single point benefit function, is calculated by using the formula (7),is an inter-source correction factor, calculated by equation (9).

Wherein the content of the first and second substances,is a nodeTo each radioactive sourceIs calculated by equation (8).

WhereinPlanning a node n for a sport_kIncluding x and y axis coordinates (m);to predict global position information for the radiation source, including x and y axis coordinates (m);is a distance scale factor and is determined according to the exploration area range and the positioning precision of the radioactive source; h_srcThe small offset prevents infinite values due to too small distance.

Benefit additive effect correction factorThe method is used for relieving the influence caused by multi-source benefit superposition.

Wherein the content of the first and second substances,

the maximum distance ratio from the mth child node to the nth and jth radioactive sources is defined, and the minimum value is 1; lambda [ alpha ]²Is a scale factor for adjusting the multi-source benefit superposition effect;

step seven, the Gain of the radiation field exists at each point_srcRespectively by a radiation gain correction term C_radAnd repeatedly exploring the correction term C_odomAnd correcting the original gain, and further improving the sampling diversity of the radiation field on the premise of ensuring the exploration efficiency. The repetitive search correction term C is calculated by equation (11)_odomAnd the obtaining process needs to depend on the normalized distance Calculated by equation (12). Calculating a radiation gain correction term C by equation (13)_rad，。

Wherein, P_posFor planning the position of the target point, x and y axis coordinates (m) are included;the position of the sampled point, including x and y axis coordinates (m);the distance scale factor is determined according to the limit range of the repeated exploration area; h_odomThe small offset prevents infinite values due to too small distance.

Wherein the content of the first and second substances,to plan a target point P_posThe radiation sample values in the nearest neighbourhood,for ambient background radiation, theta_radFor horizontal offset of radiation gain correction curve, b_radAnd eta is a base value after radiation gain correction and a lower limit value after correction.

And step eight, performing state screening on the generated expansion tree according to a gain maximization criterion, and selecting a first branch with the maximum leaf node gain as a target point of an exploration strategy. And local navigation is performed by combining with a DWA algorithm, and real-time obstacle avoidance driving of the robot from the current pose to the expected pose is realized.

Step nine, repeating the iteration steps from the first step to the eighth step until the optimal confidence probabilityAnd a termination confidence probability threshold THR_confSatisfy the requirement ofAnd searched area S_expOccupying the suspicious region area S_susRatio S of_exp/S_sus≥THR_expWhen the search is completed, the search is ended.

The environment exploration and state prediction process in the four-point source scene is respectively shown in fig. 1 to fig. 3. Fig. 1 shows the distribution state of each layer of particle swarm corresponding to different stages in the exploration process of the robot and the positions of the robot sampling the radiation intensity. Fig. 2 shows the positions and trajectories of the robot corresponding to different stages in the robot exploration process and the map model constructed for the obstacles in the scene. Fig. 3 shows the sampled position, the true source position and the position predicted by the method at the end of the exploration by the robot. The algorithm can effectively identify the state and the number of radioactive sources in an unknown multi-source radiation environment, the prediction precision can be gradually improved along with the increase of the number of sampling points, and the method for exploring corresponding 15 times of iterative estimation in each step can still meet the online prediction requirement.

The statistical results of the predictions in the four-point radioactive source scene are shown in fig. 4-6, and fig. 4 shows the predicted position errors and the optimal confidence probabilities of the radioactive sources corresponding to different exploration stages. FIG. 5 illustrates the predicted intensity error and the optimal confidence probability for each radiation source for different exploration phases. Fig. 6 shows the run-time distribution of the different modules of the method for different exploration phases. Meanwhile, the feasibility of an unknown environment source searching strategy based on a path planner of multi-source radiation gain and a peak suppression particle filter algorithm is verified. The strategy can independently search a suspected radiation area, confirm the state in a surrounding sampling mode and effectively solve the problems of non-parametric estimation of the number of sources and pseudo source identification; meanwhile, the mode of fusing the multi-mode sensor data gives consideration to the construction of the environmental geometric barrier and the effective sampling exploration of the radiation field.

Although the invention herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present invention. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that features described in different dependent claims and herein may be combined in ways different from those described in the original claims. It is also to be understood that features described in connection with individual embodiments may be used in other described embodiments.