Indoor fuzzy sound source based exploratory positioning and movement method for robots

Abstract

The application provides a kind of robot based on indoor fuzzy sound source's exploratory positioning and moving method, it is related to robot technical field.The moving method includes S1, obtaining sound source information.S2, whether sound source information is sent by target sound source is judged.Not then do not move.Is then execute subsequent step.S3, according to sound source information, the position of sound source is positioned, and whether the positioning result is accurate positioning or fuzzy positioning is judged.S4, when the positioning result is accurate positioning, the moving path is generated, and the robot is moved to the position of sound source.S5, when the positioning result is fuzzy area, through adaptive particle filtering algorithm, the position of intermediate transition station is obtained.S6, according to the position of intermediate transition point, the moving path to the position of intermediate transition point is generated, and the robot is moved to the position of intermediate transition point.S7, when the robot moves to the position of intermediate transition point, human-computer interaction can be carried out, so that the robot gradually moves close to sound source, and finally accurately positioned and moved to sound source.

Description

Technical Field

[0001] This invention relates to the field of robotics, and more specifically, to a method for exploratory localization and movement of a robot based on indoor fuzzy sound sources. Background Technology

[0002] Indoors, robots equipped with microphone arrays can identify their owner's location by sound. They then use a scene model and path planning to move closer to their owner. For example, a robotic vacuum cleaner responds to a user's call and moves to their vicinity to clean.

[0003] Currently, most commercial robots are equipped with microphone arrays that can only identify the owner's location by sound within a range of 3 to 5 meters, enabling precise positioning and navigation within that range. Beyond 5 meters, sound sources often experience multiple reflections, causing the robot to be unable to pinpoint the exact location of the sound source, only obtaining a vague area of ​​suspicion. Therefore, it cannot respond effectively.

[0004] Some robots use camera sensors to obtain the relative distance and orientation angle between the robot and its owner. This assists the microphone array sensors in determining the precise location of the sound source. However, when there are many people in the recognition scene, or when people are obstructed by other objects, the camera will be unable to help determine the sound source location. It may even lead to positioning in the wrong area.

[0005] In view of this, the applicant hereby submits this application after studying the existing technology. Summary of the Invention

[0006] The present invention provides a method for exploratory localization and movement of robots based on indoor fuzzy sound sources, in order to improve at least one of the above-mentioned technical problems.

[0007] This invention provides a method for exploratory localization and movement of a robot based on indoor fuzzy sound sources, comprising steps S1 to S7.

[0008] S1. Obtain sound source information through a microphone array.

[0009] S2. Using a voiceprint recognition algorithm, determine whether the sound source information originates from the target sound source. If not, remain in place. If so, proceed to the next step.

[0010] S3. Based on the sound source information, locate the sound source position using an array microphone sound source localization algorithm, and determine whether the localization result is precise or fuzzy. The fuzzy localization result is a fuzzy region containing the sound source position.

[0011] S4. When the positioning result is accurate positioning, a point-to-point movement path from the robot to the sound source location is generated based on the path planning algorithm, so that the robot can move to the sound source location along the movement path.

[0012] S5. When the positioning result is a fuzzy area, the location of the intermediate transition station is obtained through an adaptive particle filter algorithm based on the fuzzy area.

[0013] S6. Based on the location of the intermediate transition point, generate a movement path from the robot to the intermediate transition point using a path planning algorithm, so that the robot can move along the movement path to the intermediate transition point.

[0014] S7. When the robot moves to the intermediate transition point, it can interact with the human to repeat the previous steps, so that the robot can explore and gradually move closer to the sound source, and finally accurately locate and move to the sound source.

[0015] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0016] The robot in this embodiment of the invention employs an exploratory target localization scheme based on intermediate transition stations. When the location of a sound source cannot be directly and accurately pinpointed, the robot moves to an intermediate station, interacts with the sound source again, re-identifies the sound source location, and repeats the relevant steps to gradually approach the sound source, achieving exploratory target localization. This avoids the situation where the robot cannot effectively respond to distant sound sources indoors due to the initial inability to accurately locate them. Furthermore, the invention provides a movement scheme for the robot to reach ambiguous sound sources, improving the robot's intelligence and possessing significant practical value. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the move method.

[0019] Figure 2 This is the logic diagram of the move method.

[0020] Figure 3 This is a flowchart of the speech feature parameter MFCC extraction process.

[0021] Figure 4 This is a diagram of the BP neural network model structure. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1 to 4 This invention provides a method for exploratory localization and movement of a robot based on an indoor fuzzy sound source, which can be executed by a service robot such as a robotic vacuum cleaner (hereinafter referred to as "the robot"). Specifically, it is executed by one or more processors in the robot to implement steps S1 to S7.

[0024] S1. Acquire audio through a microphone array.

[0025] Specifically, a microphone array is a collection of acoustic sensors. It consists of a certain number of microphones arranged in a specific geometric space, such as in a line, cross, circular, spiral, spherical, or irregular shape. Microphone arrays can perform spatiotemporal processing on sound signals from different spatial directions, achieving functions such as noise suppression, reverberation removal, human voice interference suppression, sound source localization, sound source tracking, and array gain. They can act as the "ears" of robots to collect and process sound signals.

[0026] The main functions of a microphone array are as follows: 1. Tracking and locating sound sources: Microphone arrays can detect the location of sound sources, primarily by calculating the time difference between the arrival of signals at two microphones. This allows for the calculation of the direction and distance between the sound source and the array, achieving the goal of tracking the target sound source and enhancing the target speech. 2. Suppressing reverberation, noise, and echo: Excessive noise and echo reverberation interference can make the sound unclear, severely affecting speech recognition and significantly degrading speech quality. Microphone arrays can remove reverberation and echo. 3. Signal extraction and separation: Microphone arrays can form a beam, separating signals and collecting only the signals within the beam, thus enabling both signal extraction and signal separation.

[0027] S2. Using a voiceprint recognition algorithm, determine whether the sound source information originates from the target sound source. If not, remain in place. If so, proceed to the next step.

[0028] Preferably, the voiceprint recognition algorithm is a deep neural network, convolutional neural network, artificial neural network, BP neural network, or RBF neural network. More preferably, attention mechanisms and learning to rank are used in the voiceprint recognition algorithm to improve the training process.

[0029] Based on the above embodiments, in an optional embodiment of the present invention, the voiceprint recognition algorithm is a voiceprint recognition algorithm based on Mel-frequency cepstral coefficients and a backpropagation (BP) neural network. The voiceprint recognition algorithm based on Mel-frequency cepstral coefficients and a BP neural network includes steps A1 and A2.

[0030] A1. Based on the sound source information, perform pre-emphasis, framing, windowing, fast Fourier transform, Mel filtering, and logarithmic transformation, and then perform discrete cosine transform to obtain MFCC feature parameters.

[0031] A2. Input the MFCC feature parameters into the trained BP neural network to identify the voiceprint and obtain the identity information of the voicer.

[0032] The basic principle of Mel-scale Frequency Cepstral Coefficients (MFCC): Based on the auditory characteristics of the human cochlea, MFCC has wide applications in speech recognition and speaker identification. MFCC filters the input signal by setting bandpass filters according to critical bandwidths across the low-to-high frequency band. The signal energy output from each bandpass filter is then used as a fundamental feature of the signal. This feature, after further processing, becomes the input feature of the speech signal. The speech feature signal processed by MFCC does not depend on the properties of the signal and makes no assumptions or restrictions on the input signal. Therefore, it is more robust and has better recognition performance than other general methods for processing speech signals.

[0033] The process of extracting speech feature parameters MFCC is as follows: Figure 3 As shown. It includes steps 1 through 6.

[0034] Step 1: Pre-weighting.

[0035] During pre-emphasis processing, the speech is passed through a high-pass filter. The purpose of this operation is to flatten the signal spectrum, keeping it across the entire frequency band from low to high frequencies, so that the spectrum can be calculated using the same signal-to-noise ratio, while compensating for the high-frequency components of the speech signal that are suppressed by the vocal system.

[0036] Step 2: Frame division.

[0037] To ensure the input signal is stable, the speech needs to be divided into frames. A unit of observation consisting of N sampling points is called a frame. To avoid excessive variation between adjacent frames, an overlapping region containing M sampling points is allowed between two adjacent frames. Typically, M is 1 / 2 or 1 / 3 of N.

[0038] Step 3: Add a window.

[0039] Windowing is used to address the leakage problem caused by the non-periodic truncation of the signal, resulting in a trailing effect across the entire frequency band. Windowing makes the global signal more continuous, essentially multiplying the sampling points in each speech frame by the corresponding elements in the window function.

[0040] Step 3: Fast Fourier Transform (FFT).

[0041] The Fast Fourier Transform (FFT) first pads each frame of the time-domain signal x(n) with zeros to form a sequence of length N, then performs a Fast Fourier Transform on it to obtain the linear spectrum X(k). The conversion formula between x(n) and X(k) is as follows:

[0042]

[0043] Step 4: Mel filtering.

[0044] Since the principle analysis of MFCC is based on the characteristics of human hearing, and the pitch of a sound perceived by the human ear is not linearly proportional to its frequency, using the Mel frequency is more consistent with the characteristics of human hearing. The relationship between the actual frequency and the Mel frequency is as follows:

[0045] Mel(f) = 2595 × lg(1 + f / 700)

[0046] The linear spectrum X(k) obtained from equation (1) is passed through the Mel filter bank to obtain the corresponding Mel spectrum. The Mel frequency filter bank consists of several triangular bandpass filters set within the frequency spectrum of speech.

[0047] Step 5: Take the logarithm.

[0048] Taking the logarithm is used to calculate the logarithmic energy of the output of each filter:

[0049]

[0050] Among them, H m (k) represents the frequency response of the triangular filter.

[0051] Step 6: Discrete Cosine Transform (DTC).

[0052] The MFCC parameters are obtained through Discrete Cosine Transform (DTC):

[0053]

[0054] Therefore, MFCC is composed of: N-dimensional MFCC parameters (N / 3 MFCC coefficients + N / 3 first-order difference parameters + N / 3 second-order difference parameters) + frame energy.

[0055] The BP neural network algorithm is a supervised learning algorithm, and its model structure is as follows: Figure 4 As shown. Its main idea is: input training samples, use the backpropagation algorithm to repeatedly adjust and train the network's weights and biases, so that the output vector is as close as possible to the desired vector.

[0056] During algorithm iteration, gradient descent is used to adjust controllable parameters. Through each iteration, the BP network continuously adjusts its parameters using a specific training algorithm to reduce the training error, thereby gradually approaching the desired value. Training is complete when the sum of squared errors of the network's output layers is less than a specified error, and the network's weights and biases are saved.

[0057] Specific steps: Step 1: Initialization, randomly assigning connection weights and thresholds. Step 2: Calculating the outputs of each unit in the hidden and output layers based on the given input / output pattern. Step 3: Calculating new connection weights and thresholds; selecting the next input pattern and returning to step 2, repeating the training process until the network output error reaches the required level, then ending the training.

[0058] Backpropagation (BP) neural networks have the following advantages: nonlinear mapping capability, self-learning and adaptive capability, generalization capability, and fault tolerance.

[0059] S3. Based on the sound source information, locate the sound source position using an array microphone sound source localization algorithm, and determine whether the localization result is precise or fuzzy. The fuzzy localization result is a fuzzy region containing the sound source position.

[0060] In this embodiment, a fuzzy sound source refers to a sound source that is far away from the robot and has been reflected multiple times indoors, so the robot array microphone sound source localization technology cannot identify the precise location, but can only obtain a sound source area that includes the location of the sound source.

[0061] Specifically, locating sound sources using array microphones is existing technology, and will not be elaborated upon here.

[0062] S4. When the positioning result is accurate positioning, a point-to-point movement path from the robot to the sound source location is generated based on the path planning algorithm, so that the robot can move to the sound source location along the movement path.

[0063] Preferably, in this embodiment, the path planning algorithm adopts an algorithm that fuses the A* algorithm and the dynamic window algorithm to perform path planning and obstacle avoidance.

[0064] The A*(A-Star) algorithm is the most effective direct search method for solving the shortest path in a static road network. It is also a commonly used heuristic algorithm for many other problems, and it is one of the most effective direct search algorithms. The formula is expressed as: f(n) = g(n)+h(n). Wherein, f(n) is the minimum estimated cost from the initial state to the target state via state n; g(n) is the minimum estimated cost from the initial state to state n in the state space; h(n) is the minimum estimated cost of the path from state n to the target state. For the path search problem, states are nodes and costs are distances;

[0065] Selection of true h(n): The key to ensuring that the shortest path (optimal solution) is found lies in the selection of the evaluation function f(n), or the selection of h(n). Let d(n) represent the estimated distance from state n to the target state, then the selection of h(n) roughly has the following three cases. The first case: if h(n) < d(n), in this case, the number of searched points is large, the search range is large, and the efficiency is low, but the optimal solution can be obtained; the second case: if h(n)=d(n), the search efficiency is the highest at this time; the third case: if h(n)>d(n), the number of searched points is small, the search range is small, and the efficiency is high, but the optimal solution cannot be guaranteed to be obtained.

[0066] The evaluation function formula of the dynamic window algorithm is:

[0067] G(v,w) = σ(α·heading(v,w)+β·dist(v,w)+γ·v(v,w)) In the formula, G(v,w) is the trajectory evaluation value of the robot at the speed (v,w); heading(v,w) is used to evaluate the angle difference between the orientation and the target when the robot reaches the end of the simulated trajectory at the currently set sampling speed; dist(v,w) is the distance between the robot and the nearest obstacle on the current trajectory; if there is no obstacle on this trajectory, it is set to a constant; v(v,w) is the current speed value of the robot; α is the orientation weight parameter of the robot, β is the safe distance weight parameter, γ is the speed weight parameter, and σ is the smoothing function.

[0068] The dynamic window algorithm performs calculation according to the above evaluation function to generate the trajectory evaluation value of the corresponding movement path.

[0069] S5: When the positioning result is a fuzzy area, the position of the intermediate transition station is obtained through an adaptive particle filter algorithm according to the positioned fuzzy area.

[0070] This invention utilizes an adaptive particle filtering algorithm to select intermediate transition stations, thereby determining the target point for the robot. This avoids situations where the robot cannot respond to user commands when the sound source location cannot be accurately identified, representing a significant improvement.

[0071] Based on the above embodiments, in an optional embodiment of the present invention, step S4 specifically includes steps S51 to S53.

[0072] S51 When the positioning result is a fuzzy area, based on the robot's previous mapping, and based on the robot's first position in the grid map, the fuzzy area of ​​the fuzzy positioning is mapped to the pre-built grid map to obtain the mapped area.

[0073] Specifically, the blurred areas identified by the microphone array are mapped onto the established grid map. The grid map is constructed by the robot based on visual SLAM or laser SLAM, or a pre-constructed map can be directly imported; this invention does not impose specific limitations on this.

[0074] The robot can determine the position and extent of an ambiguous region relative to itself using array microphone sound source localization technology. Based on the robot's position in the grid map, the ambiguous region can be directly mapped onto the grid map, thus obtaining the specific location of the ambiguous region in the grid map.

[0075] S52. Number the grid cells within the mapped area.

[0076] Specifically, the number of grid cells in the blurred region is counted as k, and the center point coordinates of each grid cell in the blurred region are obtained. Then, the grid cells in the blurred region are mapped one-to-one and numbered as k integers in the range [1, k].

[0077] S53. Randomly obtain A numbers according to the number, obtain a first number set, and according to the grid corresponding to each number in the first number set, obtain the grid closest to the first position as an intermediate transition station, or obtain the grid at the center of the fuzzy area as an intermediate transition station.

[0078] Specifically, the number of numbers A is adaptively set. By randomly seeding A particles into K grids in the fuzzy region, A grids with random numbers are selected to obtain a first set of numbers. Based on the first set of numbers (A numbers corresponding to the selected grid numbers), the distance (Euclidean distance or Manhattan distance) from the robot's current position to the center point of the A selected grids is calculated. The grid closest to the robot's position among the grids containing the A particles is selected as an intermediate transition station, or the grid at the center of the fuzzy region is selected as an intermediate transition station.

[0079] Based on the above embodiments, in an optional embodiment of the present invention, when the grid closest to the first position is selected as an intermediate transition station, step S53 specifically includes steps S53A1 to S53A6.

[0080] S53A1. Randomly obtain A numbers according to the given numbering to obtain a first set of numbers. Preferably, A is a value less than k. A can be determined based on k, such as half, one-third, one-fifth, etc. of the value of k (i.e., A changes proportionally to the value of k). This invention does not specifically limit the specific value of A.

[0081] S53A2. Calculate the distance (Euclidean distance or Manhattan distance) from each grid cell corresponding to each number in the first set of numbers to the first position, and select the center of the grid cell with the smallest distance as the center of the circle.

[0082] S53A3. Repeat the following steps B times, using the last updated center as the intermediate transition station.

[0083] S53A4. Based on the center of the circle (i.e., the center point of the selected grid), with a radius of three times the side length of the grid, re-sample randomly, and randomly distribute A / 2 particles to obtain the numbers of the selected grids near the center, that is, randomly obtain... Calculate the raster number of the region and obtain the second set of numbers.

[0084] S53A5, Based on the grid number within the mapped area, randomly obtain another... The third set of numbers is obtained by determining the grid number of each particle.

[0085] S53A6. Calculate the distance from the grid corresponding to the second number set and the third number set to the first position, and select the grid with the smallest distance to update the center of the circle.

[0086] In this embodiment, a suitable point is selected based on the fuzzy region, making it closest to the robot's position. Specifically, an appropriate number of particles are randomly scattered based on the determined size of the fuzzy region. Then, the distances between all particles and the robot's current position are calculated, and the position of the particle with the shortest distance is selected as the initially determined intermediate station position. Next, using this distance as the initial assignment for the intermediate station, particle resampling is performed. Half of the particles are randomly scattered in the vicinity of the initially determined shortest-distance particle, while the other half are randomly scattered throughout the entire suspicious area. The distances between all particles and the robot are calculated again, and if any distance is smaller than the initially determined value, the intermediate station position is updated. This process is repeated three times to determine the position of the particle with the final minimum distance as the selected intermediate station.

[0087] The specific steps include: 1. Adaptively selecting the number of particles A based on the total number of grid cells k. Let the weight of each particle be... Each particle represents a unique identifier. 2. Particle filtering randomly distributes particles within the blurred region. This results in the random output of A integers in the range [1, k]. 3. The distances d1, d2...d1 between the robot's current position and the center points of A grid cells are calculated. A (Euclidean distance or Manhattan distance. Let the robot's current position be (x0, y0), and the coordinates of the center positions of each grid be (x1, y1), (x2, y2)...(x...) A ,y A ), calculate Euclidean distance Or Manhattan distance |x i -x0|+|y i -y0| discards other particles and selects the grid corresponding to the particle with the smallest distance (i.e., obtains the number of the nearest grid). 4. Using half of the particles, with the center of the grid containing the particle with the smallest distance as the center and a radius three times the side length of the grid, resample randomly. The other half of the particles are randomly sampled within the original range. Calculate the position of each particle and the robot. Repeat three times, retaining the coordinates of the center of the grid containing the particle with the shortest distance.

[0088] Based on the above embodiments, in an optional embodiment of the present invention, when the grid at the center of the fuzzy region is selected as an intermediate transition station, step S53 specifically includes steps S53B1 to S53B3.

[0089] S53B1. Repeat the following steps C times to obtain C fuzzy region center points, then calculate the mean of the C fuzzy region center points to obtain the intermediate transition station.

[0090] S53B2. Randomly obtain A numbers according to the numbering to obtain the first number set.

[0091] S53B3. Calculate the average coordinates of the grid corresponding to each number in the first set of numbers to obtain the center point of the fuzzy region.

[0092] In this embodiment, a suitable point is selected based on the ambiguous region, such that this point is approximately at the center of the suspicious region. Specifically, an appropriate number of particles are randomly scattered according to the determined size of the ambiguous local area. The average value of the grid coordinates mapped by all particles is calculated and recorded as the initial value of the intermediate station. This operation is repeated 20 times, and the average value of these 20 iterations is recorded as the determined position of the robot's intermediate transition station.

[0093] The specific steps include: 1. Adaptively selecting the number of particles A based on the total number of grid cells k, and assuming each particle has a weight of . 2. Particle filtering randomly seeds particles in the blurred region. This involves randomly outputting A integers in the range [1, k]. 5. Calculate the average coordinates of the grid center points corresponding to these A integers. Repeat this process 20 times, using the average coordinates from these 20 iterations as the final coordinates of the intermediate transition stations.

[0094] S6. Based on the location of the intermediate transition point, generate a movement path from the robot to the intermediate transition point using a path planning algorithm, so that the robot can move along the movement path to the intermediate transition point.

[0095] S7. When the robot moves to the intermediate transition point, it can interact with the human to repeat the previous steps, so that the robot can explore and gradually move closer to the sound source, and finally accurately locate and move to the sound source.

[0096] The moving method of this invention mainly includes the following steps:

[0097] 1. Preliminary Steps: The robot needs to complete indoor mapping. The robot can generate an indoor grid map based on visual SLAM or laser SLAM. The generation of a grid map of the robot's activity range is existing technology and will not be elaborated upon in this invention. Of course, in other embodiments, existing maps can also be directly imported, and this invention does not specifically limit this.

[0098] 2. Voiceprint Recognition. This identifies whether the voice source is the owner's. Voiceprint recognition algorithms can be deep learning algorithms; the initial stage involves collecting voiceprint information from the person who needs to respond for training.

[0099] 3. Based on the first step's identification of the owner's call, determine if the sound source can be precisely located. If it can be precisely located, directly use a path planning algorithm to reach the target location. If the sound source is located in a vague area, it cannot be precisely located, and proceed to step 4.

[0100] 4. Selection of intermediate transition sites: The adaptive particle filter algorithm is used to select intermediate transition sites in the identified fuzzy areas.

[0101] 5. The robot reaches the intermediate station through path planning. It then uses human-computer interaction to re-identify the location of the owner's voice. If the owner's location is accurately identified, the path planning proceeds directly to that location. If the identified sound source is still unclear, proceed to step three.

[0102] The robot movement method of this invention employs an exploratory target localization scheme based on the selection of intermediate transition stations. This scheme allows the robot to gradually approach the sound source through exploratory localization when the sound source cannot be directly and accurately located. Ultimately, it accurately locates the sound source target, avoiding the situation where the robot cannot effectively respond indoors due to the distance of the sound source and the inability to identify the sound source location using existing hardware and microphone array localization algorithms. This improves the robot's intelligence and has significant practical implications.

[0103] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for exploratory localization and movement of a robot based on indoor fuzzy sound sources, characterized in that, Include: Acquire sound source information using a microphone array; The voiceprint recognition algorithm is used to determine whether the sound source information is emitted by the target sound source; if not, the process remains unchanged; if so, the subsequent steps are executed. Based on the sound source information, the sound source location is located using an array microphone sound source localization algorithm, and the localization result is determined to be either precise localization or fuzzy localization; wherein, the localization result of fuzzy localization is a fuzzy region containing the sound source location. When the positioning result is accurate positioning, a point-to-point movement path from the robot to the sound source is generated based on the path planning algorithm, so that the robot can move to the sound source along the movement path. When the localization result is a fuzzy region, an adaptive particle filter algorithm is used to obtain the intermediate transition point position based on the fuzzy region. Specifically, when the localization result is a fuzzy region, based on the robot's initial mapping and its first position on the grid map, the fuzzy region is mapped onto the pre-built grid map to obtain the mapped region. The grids within the mapped region are numbered; based on the total number of grids within the mapped region... Adaptive selection of particle number Then randomly obtain Given a set of numbers, obtain a first set of numbers; calculate the distance from each grid cell corresponding to each number in the first set of numbers to the first position, and select the center of the grid cell with the smallest distance as the center of the circle; repeat the following steps B times, using the last updated center of the circle as the intermediate transition point position: based on the center of the circle, with a radius of three times the grid side length, obtain the numbers of the grid cells near the center of the circle, and then randomly obtain half of the numbers to obtain a second set of numbers; based on the numbers of the grid cells in the mapped area, randomly obtain the other half of the numbers to obtain a third set of numbers; calculate the distance from the grid cells corresponding to the second set of numbers and the third set of numbers to the first position, and select the grid cell with the smallest distance to update the center of the circle; Based on the location of the intermediate transition point, a path planning algorithm is used to generate a movement path from the robot to the intermediate transition point, so that the robot can move along the movement path to the intermediate transition point. When the robot moves to the intermediate transition point, it can interact with humans to repeat the previous steps, allowing the robot to explore and gradually move closer to the sound source, eventually accurately locating and moving to the sound source.

2. The exploratory localization and movement method for a robot based on indoor fuzzy sound sources according to claim 1, characterized in that, Randomly obtained according to the number. Given a set of numbers, obtain a first set of numbers, and based on the grid corresponding to each number in the first set of numbers, obtain the grid at the center of the blurred region as an intermediate transition point, specifically including: Repeat the following steps C times to obtain C center points of fuzzy regions, then calculate the mean of the C center points of fuzzy regions to obtain the intermediate transition point; Randomly obtained according to the number. Given a set of numbers, retrieve the first set of numbers; Calculate the average coordinates of the grid corresponding to each number in the first set of numbers to obtain the center point of the fuzzy region.

3. A method for exploratory localization and movement of a robot based on indoor fuzzy sound sources according to claim 1 or 2, characterized in that, The voiceprint recognition algorithm is a deep neural network, a convolutional neural network, an artificial neural network, a BP neural network, or an RBF neural network. The path planning algorithm is A. A fusion algorithm of algorithm and dynamic window algorithm.

4. The exploratory localization and movement method for a robot based on indoor fuzzy sound sources according to claim 3, characterized in that, The voiceprint recognition algorithm is a voiceprint recognition algorithm based on Mel-frequency cepstral coefficients and a BP neural network. The voiceprint recognition algorithm based on Mel-frequency cepstral coefficients and BP neural network includes: Based on the sound source information, pre-emphasis, framing, windowing, fast Fourier transform, Mel filtering, and logarithmic transformation are performed, followed by discrete cosine transform to obtain MFCC feature parameters. The MFCC feature parameters are input into a trained BP neural network to identify voiceprints and obtain the identity information of the voicer.

Images