A method for adaptive exploration and task allocation of multi-robot team in unknown environment

Abstract

The application discloses a kind of unknown environment in multi-robot team's self-adapting exploration and task allocation method, belong to multi-robot system technical field.To improve the exploration efficiency and accuracy of multi-robot system in complex environment, the method of the present application is deployed before each robot frontier point detection, navigation and mapping module, in host deployment task allocation, frontier point detection and map merging module;After obtaining exploration task, each robot and host are executed frontier point detection in parallel by adaptive fast exploration random tree algorithm, host uses the evolution strategy guided by bayes to allocate target point detection task for current available robot, robot moves to target point and continues to detect, constructs local map by mapping module, host integrates global map, until completing exploration task.The present application effectively improves the exploration efficiency and accuracy of multi-robot system in unknown environment, optimizes resource allocation at the same time, reduces energy consumption, improves energy utilization efficiency.

Description

Technical Field

[0001] This invention belongs to the field of multi-robot system technology, and specifically relates to a method for adaptive exploration and task allocation of multi-robot teams in unknown environments using the Rapidly-exploring Random Tree (RRT) algorithm and Bayesian-guided evolutionary strategy. Background Technology

[0002] Following natural disasters, traditional exploration and assessment methods are not only slow and costly, but also simply unfeasible in certain environments. Meanwhile, with increasing social activity and population density, rapidly and accurately assessing the situation and deploying relief resources during emergencies has become a major challenge. For example, during large public events or sudden incidents, quickly obtaining accurate environmental information is crucial for maintaining order and timely response. However, existing exploration and monitoring technologies are often inefficient in the face of complex and changing environments, failing to meet the needs of rapid dynamic response.

[0003] Multi-robot systems, with their flexible deployment and efficient collaboration, offer new possibilities for solving these problems. Compared to single-robot systems, multi-robot systems can significantly improve the speed and accuracy of understanding unknown environments through distributed exploration and information sharing. Especially in complex environments where direct human access is impossible or where the line of sight is insufficient, multi-robot systems can demonstrate unique advantages. However, achieving effective deployment and coordinated operation of multi-robot systems requires addressing a series of technical challenges, including inter-robot communication, path planning, and task allocation. Among these challenges, efficiently detecting leading points and effectively allocating tasks is key to improving the efficiency of multi-robot systems in exploring unknown environments.

[0004] In recent years, although the application of multi-robot systems in disaster assessment, urban monitoring, and other fields has gradually increased, problems such as inaccurate frontier detection and suboptimal task allocation still exist in practical operation. These problems not only affect the exploration efficiency of multi-robot systems but also limit their application in a wider range of scenarios. Therefore, developing a new method for exploration and collaboration of multi-robot systems to improve exploration efficiency and accuracy in complex environments has become an important research direction. Summary of the Invention

[0005] This invention addresses the problems of low efficiency and insufficient accuracy of current multi-robot systems in the field of exploring unknown environments. It proposes an adaptive exploration and task allocation method for multi-robot teams in unknown environments, aiming to improve the efficiency and accuracy of multi-robot systems when exploring unknown environments. By optimizing the detection of leading edge points and task allocation, it achieves fast and accurate environment mapping and resource allocation.

[0006] The present invention provides an adaptive exploration and task allocation method for a multi-robot team in an unknown environment, comprising the following steps:

[0007] Step 1: Deploy a robot system for each robot in the multi-robot team and a host system for the server; the robot system includes a navigation module, a mapping module, and a first leading edge detection module; the host system includes a task allocation module, a map merging module, and a second leading edge detection module.

[0008] Step 2: The host obtains the area exploration task, and then executes steps 3-7 below until the area exploration task is completed;

[0009] Step 3: Each robot and the host computer call their respective leading edge detection modules to perform leading edge detection in parallel. Each robot sends the set of detected leading edge points to the host computer.

[0010] Step 4: The host calls the task allocation module to assign target point detection tasks to the currently available robots;

[0011] The task allocation module executes a Bayesian-guided evolutionary strategy, including: for all frontier points, first using Bayesian optimization to predict the utility value of the frontier points, and obtaining an initial target point set based on the set threshold conditions; then using a genetic algorithm to find the optimal allocation strategy for the currently available robots based on the initial target point set.

[0012] The utility value of the frontier point x is calculated as follows:

[0013] U(x)=α·IG(x)-β·C(x)-γ·RPF(x)

[0014] Where IG(x) is the information gain of the leading edge x, representing the amount of new information the robot can obtain after reaching the point, calculated based on the unexplored region around the leading edge; cost C(x) represents the effort required for the robot to reach the leading edge x from its current position, using Euclidean distance as the cost metric; RPF represents the repulsive potential field; α, β, and γ are weighting coefficients;

[0015] In the Bayesian optimization described above, a Gaussian process model is trained using historical frontier points and their corresponding utility values. The utility value and uncertainty of the exploration point are predicted by sampling the posterior distribution of the Gaussian process.

[0016] Step 5: After receiving the assigned target point, the navigation module of each robot system plans the path from the current position to the target point, guides the robot to move along the path to the target point, and obtains the environmental data collected in real time by the sensors on the robot through the mapping module, and builds a local map by combining the localization technology.

[0017] Step 6: Each robot system sends the constructed local map to the host system. The host system integrates the local maps into a global map through the map merging module, updates the area of ​​the detected region, and sends it to each robot system.

[0018] Step 7: After the robot reaches the target point, continue to step 3.

[0019] In step 3, the leading edge detection module uses the adaptive RRT algorithm to detect the leading edge. During detection, the current position of the robot is used as the initial node of the fast exploration random tree T in the current round of exploration, and then the growth step size λ of the random tree in the current round of exploration is adaptively adjusted, as follows:

[0020]

[0021] Where λ0 is the initial growth step size; S total The total area of ​​the exploration zone is obtained by the host system from the exploration mission; S covered The current explored area is represented by S. Before each round of exploration, the host system merges the explored areas and notifies each robot. In the first round of exploration, S... covered It is 0.

[0022] Compared with existing technologies, the method of the present invention has the following advantages and positive effects:

[0023] (1) The method of this invention detects leading edges by adopting an adaptive RRT tree growth strategy. While traditional RRT algorithms can explore unknown environments, their fixed growth step size limits exploration efficiency and accuracy. This invention introduces an adaptive RRT tree growth strategy, dynamically adjusting the growth step size of the RRT tree based on current environmental information and the coverage of the explored area. This allows for rapid coverage of a large area initially, followed by gradual refinement of the exploration range, thus improving exploration accuracy. This strategy effectively overcomes the limitations of traditional methods in narrow or complex environments, improving the efficiency and accuracy of leading edge detection.

[0024] (2) The method of this invention proposes to solve the global optimization problem of task allocation in multi-robot systems using a Bayesian-guided evolutionary strategy. By learning from historical exploration data through a Bayesian optimization algorithm, the potential value of different frontier points is predicted, and based on this, a genetic algorithm is used to perform globally optimal task allocation. This strategy not only considers the current exploration benefits but also takes into account long-term resource allocation and system efficiency, optimizes the accuracy of frontier point detection and task allocation, improves the quality of environment mapping, and enhances exploration precision. Intelligent task allocation avoids resource waste and task duplication, optimizes resource allocation, and improves work efficiency.

[0025] (3) The method of this invention implements a complete multi-robot collaborative exploration mechanism, ensuring that in complex and ever-changing exploration environments, each robot can efficiently share information and collaborate to complete tasks. This significantly improves the efficiency and accuracy of multi-robot systems in exploring unknown or complex environments, enhances the execution efficiency of exploration tasks, and also improves the stability and reliability of the robot operating system. The method of this invention enables rapid response to environmental changes, effectively reduces exploration time, and optimizes resource allocation. It has significant practical application value and broad application potential, and is applicable to exploration tasks in various complex environments, providing effective technical solutions for fields such as emergency rescue, environmental monitoring, and natural disaster assessment. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the robot system of the present invention;

[0027] Figure 2 This is a schematic diagram of the host system of the present invention;

[0028] Figure 3 This is a schematic diagram of a multi-machine exploration system performing location and environment exploration according to an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings.

[0030] The adaptive exploration and task allocation method for multi-robot teams in unknown environments of the present invention improves exploration efficiency by integrating an adaptive fast exploration random tree algorithm and a Bayesian-guided evolutionary strategy. By dynamically adjusting the exploration strategy and optimizing task allocation, it significantly improves the exploration efficiency and accuracy of multi-robot systems in complex environments, which has important theoretical significance and wide application value.

[0031] like Figure 1 As shown, this embodiment of the invention provides a robot system, mainly including a navigation module, a mapping module, and a first leading edge detection module. The navigation module provides a planned path for the robot's movement, guiding it to a designated target point, and includes underlying motion control and positioning functions. The mapping module provides mapping functionality for the robot, constructing a map of its surrounding environment during movement by running a corresponding mapping algorithm. The first leading edge detection module provides leading edge detection functionality for the robot, which involves running an adaptive RRT algorithm.

[0032] like Figure 2As shown, this embodiment of the invention provides a host system, mainly including a task allocation module, a map merging module, and a second frontier detection module. The task allocation module receives frontier points from the frontier detection module and uses this data to allocate tasks to multiple robot systems, assigning them frontier points to be explored. The map merging module receives sub-maps from mapping modules in different robot systems and merges these sub-maps into a main map. The second frontier detection module provides frontier detection functionality for the robots, running an adaptive RRT algorithm, similar in function to the first frontier detection module in the robot system.

[0033] like Figure 3 As shown in the figure, an adaptive exploration and task allocation method for a multi-robot team in an unknown environment according to an embodiment of the present invention deploys the following for each robot in the multi-robot team: Figure 1 The robot system shown enables each robot to autonomously explore and perform tasks, and is deployed on the host computer as follows: Figure 2 The host system shown performs exploration task allocation and global front detection. The host server can be a single computer. The implementation steps of this embodiment are described below.

[0034] Step 1: For the area exploration task acquired by the host, each robot and the host perform front edge detection in parallel, and each robot sends the set of detected front edge points to the host.

[0035] The robot and the host computer of this invention each use their respective leading-edge detection modules to run an adaptive RRT algorithm for leading-edge detection. The leading-edge detection process is illustrated using a robot as an example. The robot system calls the first leading-edge detection module to initiate leading-edge detection: initially, the initial node of the RRT tree T is automatically set to the current actual position of the robot, and the initial growth step size λ0 of the RRT tree is preset according to the size and complexity of the environment. Then, a loop exploration process is entered, which mainly includes the following steps:

[0036] Step 1.1: Random Sampling: Randomly select a location point P in the exploration environment. rand As a potential target for exploration.

[0037] The robot uses its onboard sensors to detect its surroundings in real time at its current location, acquiring information within its perception range and constructing a local map using its mapping module. The robot then randomly selects a location P in the unknown environment. rand .

[0038] Step 1.2: Find the nearest node: Find the node at distance P in the RRT tree T. rand The nearest existing node P nearest This is to ensure that the RRT tree T can continue to grow into unknown regions.

[0039] Step 1.3: Adaptively adjust the RRT tree growth step size: Dynamically adjust the growth step size λ based on the explored environmental information, as follows:

[0040]

[0041] Among them, S covered S represents the area of ​​the region currently explored. total This represents the total area to be explored. This adaptive mechanism allows the system to quickly cover a large area in the early stages of exploration and improve accuracy in the detailed exploration phase. The known areas explored by all robots in each round are merged through the host system, which then notifies each robot of the currently explored area S. covered During the first round of exploration, S covered The total exploration area S is 0. total Obtained by the host system from exploration missions.

[0042] Step 1.4: Growing new nodes: Starting from P with the adjusted step size λ nearest To P rand New node P grows in the direction new If from node P nearest To node P new Without obstacles, the new node P will be... new If the new node P is added to the RRT tree T, then... new If the new node P is located at the forefront of the robot's known environment, then... new It is placed into the set of frontier points as a frontier point.

[0043] Step 1.5: Determine if the current exploration has reached the stopping condition. If yes, stop the exploration and output the set of frontier points. Otherwise, continue to step 1.1.

[0044] In the first round of exploration, a time limit is preset, and exploration stops when the exploration time is reached.

[0045] During non-first rounds of exploration, stop exploring when the target point is reached or approached.

[0046] After the above steps, each robot system explores the leading edge points in the local map through the leading edge point detection module, and transmits the detected leading edge point set to the task allocation module in the host system for subsequent task allocation.

[0047] The host invokes the second leading edge detection module. Upon initial starting of leading edge detection, it selects unknown points for exploration based on the environment perceived by the host device, following the steps outlined above, to detect leading edge points. After each round of leading edge detection, the host acquires a merged map, obtaining the explored area. The root node of the RRT tree T generated by the host is always the host's position and can generally remain fixed. Then, the host continues to explore leading edge points in the global region.

[0048] Step 2: The host calls the task allocation module to estimate the value of the leading edge points detected by itself and each robot, and allocates the task of leading edge point detection to each robot.

[0049] The task allocation module employs a Bayesian guided evolutionary strategy (BGE) to achieve reasonable task allocation. This is a key component of the method in this invention, primarily used to optimize task allocation in multi-robot systems exploring unknown environments. This Bayesian guided evolutionary strategy combines the advantages of Bayesian optimization and genetic algorithms, achieving global optimization of task allocation through learning and prediction from historical exploration data. The implementation of the BGE strategy in this invention mainly includes two parts: the first part uses a Gaussian process to evaluate the exploration value of frontier points and select a set of target points; the second part uses a genetic algorithm to allocate tasks for the robots to explore these target points.

[0050] The method of this invention uses a utility function to quantify the value of frontier exploration. The designed utility function simultaneously considers factors such as information gain (IG), cost (C), and repulsive potential (RPF). The utility value of the frontier point x is calculated as follows:

[0051] U(x)=α·IG(x)-β·C(x)-γ·RPF(x)

[0052] Wherein, IG(x) is the information gain of the leading edge x, representing the amount of new information the robot can acquire after reaching that point, calculated based on the unexplored region surrounding the leading edge; cost C(x) represents the effort required for the robot to reach the leading edge x from its current position, and in this embodiment, Euclidean distance is used as the cost metric; the repulsive potential field RPF represents the potential field that influences the robots to avoid them heading to the same leading edge. α, β, and γ are weighting coefficients used to adjust the relative importance of each factor.

[0053] Since directly calculating the utility function, especially in a multidimensional space, is very time-consuming, the method of this invention uses a Gaussian process (GP) to model the utility function and calculate the utility function value of the frontier node, that is, to predict the potential value of the robot going to a frontier point.

[0054] First, the BGE strategy of this invention obtains an initial set of target points through Bayesian optimization. The process of performing Bayesian optimization includes:

[0055] Step 2.11: Bayesian Model Initialization: First, collect historical frontier point data for robot exploration, including the frontier point locations and corresponding utility values; use the historical frontier point data to initialize the Bayesian model, establishing a prior estimate of the exploration value of the unknown environment. The Bayesian model mainly uses Gaussian processes to simulate the exploration value function, where the value of each exploration point, i.e., the frontier point, is assumed to be a Gaussian distributed random variable.

[0056] Step 2.12: Gaussian Process: Train a Gaussian process model using collected historical data. A Gaussian process is defined by the mean function m(x) and the covariance function k(x,x,), used to describe the uncertainty of the value of the exploration point.

[0057] For any two exploration points x, x, the correlation of their values ​​is given by the covariance function k(x, x,). In a Gaussian process at position x... Represented as:

[0058]

[0059] Step 2.13: Calculate the fitness function: In Bayesian optimization, the fitness function is used to evaluate the value of each exploration point. It is used to predict the utility value U(x) and uncertainty of the exploration point by sampling the posterior distribution of the Gaussian process. The uncertainty is usually represented by the variance of the Gaussian process.

[0060] The task allocation module of this invention predicts the utility value U(x) of each obtained frontier point x by sampling its Gaussian posterior distribution. The higher the utility value, the higher the value. Based on preset utility value thresholds and uncertainty thresholds, frontier points that meet the threshold conditions are selected to form an initial target point set X. The initial target point set is then further optimized for task allocation using a genetic algorithm.

[0061] The genetic algorithm aims to achieve global optimization of task allocation in a multi-robot system by simulating natural selection and genetic mechanisms. It finds the optimal allocation strategy among all possible frontier point assignments, and includes the following process:

[0062] Step 2.21: Population Initialization: Based on the initial target point set obtained through Bayesian optimization, initialize a population for a task assignment scheme. The population consists of a series of individuals, each representing a possible multi-robot task assignment scheme. The population is initialized using a random generation method to ensure broad coverage of the initial search space.

[0063] When executing the genetic algorithm, the first step is to obtain a set of available robots R and an initial set of target points X. The goal of task allocation is to assign a target point to each available robot for exploration. Based on the set of available robots R and the initial set of target points X, a Gaussian process is used to predict the utility value of each robot assigned to each target point, and this value is stored in the matrix `utilities`. In this embodiment of the invention, each row of the matrix `utilities` represents a robot, and each column represents a target point. The elements of the matrix `utilities[robot]` are... i ][frontier j ] represents the i-th robot i Assigned to the j-th target frontier point j The utility value.

[0064] Suppose that the available robot set R contains n robots, and each individual is assigned a target point to each of the n robots.

[0065] Step 2.22: Fitness assessment: For each individual in the population, calculate the individual fitness value according to the task assignment scheme it represents. The higher the fitness, the better the task assignment scheme.

[0066] The fitness F(h) of individual h is calculated in the method of this invention as follows:

[0067]

[0068] Among them, h[robot i ] represents the robot in individual h. i The assigned target point; Penalty is the proximity penalty, which is the penalty when the distance between the target points assigned to two robots is less than a set threshold, to ensure that the robots do not repeatedly explore the same area.

[0069] Step 2.23: Selection Operation: The selection operation is based on the fitness of individuals. Individuals with higher fitness have a greater probability of being selected to participate in the generation of the next generation of the population. Embodiments of this invention may employ strategies such as roulette wheel selection or tournament selection to ensure the transmission of superior genes.

[0070] Step 2.24: Crossover and Mutation: In this embodiment of the invention, crossover is used to allow two parent individuals to exchange gene segments, producing new offspring individuals. The crossover rate parameter determines the proportion of individuals that undergo crossover. Mutation introduces genetic diversity by randomly altering parts of an individual's genes; the mutation rate parameter controls the frequency of mutation operations.

[0071] Step 2.25: Generation of a new generation population: By combining selection, crossover, and mutation operations, a new generation population is generated to replace the original population, simulating the natural evolutionary process and gradually optimizing the population.

[0072] Step 2.26: Iterative optimization: In this embodiment of the invention, steps 2.22-2.25 are repeated until the preset number of iterations or the fitness meets the termination condition. At this point, the optimal individual in the population represents the optimal multi-robot task allocation scheme.

[0073] Through the detailed implementation methods described above, the genetic algorithm in the Bayesian-guided evolutionary strategy can efficiently solve the task allocation problem of multi-robot systems in unknown environments, significantly improving the efficiency and quality of task completion.

[0074] After the task allocation module completes its execution, it will obtain a set of optimal multi-robot task allocation schemes, which will assign a frontier target point for exploration to different robot systems.

[0075] Step three: After the robot system's navigation module receives the assigned target point, it plans an effective path from the current position to the target point and guides the robot along that path. The navigation module mainly consists of two parts: path planning and dynamic obstacle avoidance. Path planning uses the A* algorithm, and dynamic obstacle avoidance uses the Dynamic Window Approach (DWA) to ensure the robot's safe and efficient movement in unknown environments. The following is the implementation process of the navigation module planning a path to the target point in this embodiment of the invention.

[0076] Step 3.1: Environment Model Construction: First, the robot's surrounding environment needs to be represented as a model that can be used for path planning. Common environment models include grid maps and topological maps. In a grid map, the environment is divided into many small grids, each representing a location in the environment, which can be either passable or impassable. This embodiment of the invention uses a commonly used grid map, which can effectively reduce memory usage and improve system operating efficiency.

[0077] Step 3.2: Cost Map Generation: Based on the environment model, a cost map is generated, where the cost value for each location represents the cost of moving from the current location to that location. The cost value can be determined based on factors such as obstacle distance and terrain difficulty. The cost map can be expressed as:

[0078] C(a,b)=w obs C obs (a,b)+w terr C terr (a,b)

[0079] Where C(a,b) is the total cost of location (a,b), C obs (a,b) are costs based on obstacles, C terr (a,b) are the costs based on terrain, w obs and w terr These are the corresponding weighting coefficients.

[0080] Step 3.3: Path Search: Using the cost map, the path search algorithm A* is employed to search for the lowest-cost path from the robot's current position to the target point. Each step in the A* algorithm considers the known minimum cost from the starting point to the current point v and the estimated cost from the current point v to the destination. The core expression of the A* algorithm is:

[0081] f(v) = g(v) + q(v)

[0082] Where f(v) is the total estimated cost of node v, g(v) is the actual cost from the starting point to node v, and q(s) is the estimated cost from node v to the destination (heuristic function).

[0083] Step 3.4: Dynamic Obstacle Avoidance: Dynamic obstacle avoidance also needs to be considered during path planning. When the robot encounters unknown or dynamic obstacles during movement, the navigation module needs to adjust the path in real time to avoid collisions. Dynamic obstacle avoidance is achieved through the local path planning algorithm—the Dynamic Window Approach (DWA).

[0084] Step 3.5: Path Tracking: After path planning is completed, the robot moves along the planned path via the path tracking controller. The path tracking controller needs to handle the robot's speed and direction control to ensure that the robot reaches the target point smoothly and accurately.

[0085] Through the above process, the navigation module effectively guides the robot to move autonomously in unknown environments, realizing path planning and tracking from the current position to the target point at the forefront.

[0086] Step four: During the robot's operation, a map of the surrounding environment is built in real time using the mapping module, and then the constructed environmental map is transmitted to the host system's map merging module for merging. In multi-robot exploration tasks in unknown environments, mapping and map merging are key steps to achieve effective environmental understanding.

[0087] The mapping process primarily relies on the sensors and localization technology carried by the robot. Sensors include LiDAR and visual cameras, while localization technology includes Simultaneous Localization and Mapping (SLAM). Each robot collects environmental data through its sensors and uses SLAM technology to build a local map in real time during exploration. The specific steps are as follows:

[0088] Step 4.1: Data Acquisition: The robot collects information about the surrounding environment through sensors, such as the location of obstacles and terrain features.

[0089] Step 4.2: Feature Extraction: Extract useful feature information from the collected data. These features will be used for subsequent map building and location estimation.

[0090] Step 4.3: Position estimation: Using the extracted features and known map information, the robot estimates its position in the environment using the SLAM algorithm.

[0091] Step 4.4: Map Update: Update the current map based on the location estimation results and newly collected environmental data. The map update process can be represented by the following formula:

[0092] M t+1 =g(M t O t X t )

[0093] Where M t+1 It is the map at time t+1, M t It is the map at time t, O t X is the environmental observation data collected by the robot at time t. t is the robot's estimated position at time t, and g is the map update function.

[0094] Step five: In a multi-robot system, each robot independently explores and builds its own local map. The map merging module is responsible for integrating these local maps into a globally consistent map. The map merging process includes:

[0095] Step 5.1: Map Alignment: First, it is necessary to determine the relative positions of the various local maps, which is usually achieved by finding common features between the maps. Map alignment can be represented as solving for an optimal transformation that maximizes the overlap between the local maps under this transformation.

[0096] Step 5.2: Map Fusion: After determining the relative positions of the local maps, the next step is to merge them into a global map. The map fusion process takes into account potential conflicts and inconsistencies between maps, employing an optimization method to smooth these inconsistencies and ultimately generate a globally consistent map.

[0097] Through the mapping and map merging processes in steps four and five above, the multi-robot system can effectively explore unknown environments and construct accurate environmental maps, providing support for subsequent navigation and task execution.

[0098] Step six: Determine whether the current area exploration task is completed. If not, repeat steps one to five above to continue frontier point detection, multi-robot task allocation, map building and merging until the exploration task is completed.

[0099] Through the above embodiments, this invention successfully addresses the problem of exploration and task allocation in multi-robot systems in unknown environments by proposing and implementing an efficient method combining an adaptive fast exploration random tree algorithm and a Bayesian-guided evolutionary strategy. This method effectively improves the accuracy of frontier detection and the optimization level of task allocation, ensuring that multi-robot systems can efficiently and accurately complete exploration tasks in complex environments. Simultaneously, through intelligent path planning and dynamic adjustment mechanisms, it significantly reduces the energy consumption of robots during exploration, improving the overall energy efficiency of the system. Furthermore, the modular design and scalability of this method allow it to be customized and expanded according to different application scenarios and needs, demonstrating broad application potential.

[0100] Except for the technical features described in the specification, all other technologies are known to those skilled in the art. Descriptions of well-known components and technologies are omitted in this invention to avoid redundancy and unnecessary limitation. The embodiments described above do not represent all embodiments consistent with this application. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this invention are still within the protection scope of this invention.

Claims

1. An adaptive exploration and task allocation method for a multi-robot team in an unknown environment, characterized in that, The method includes the following steps: Step 1: Deploy a robot system for each robot in the multi-robot team. The robot system includes a navigation module, a mapping module, and a first leading edge detection module. Deploy a host system for the host robot. The host system includes a task allocation module, a map merging module, and a second leading edge detection module. Step 2: The host obtains the area exploration task, and then executes steps 3-7 below until the area exploration task is completed; Step 3: Each robot and the host computer call their respective leading edge detection modules to perform leading edge detection in parallel. Each robot sends the set of detected leading edge points to the host computer. Step 4: The host calls the task allocation module to assign target point detection tasks to the currently available robots; The task allocation module executes a Bayesian-guided evolutionary strategy, including: for all frontier points, first using Bayesian optimization to predict the utility value of the frontier points, and obtaining an initial target point set based on the set threshold conditions; then using a genetic algorithm to find the optimal allocation strategy for the currently available robots based on the initial target point set. The utility value of the frontier point x is calculated as follows: in, The information gain at the leading edge x represents the amount of new information the robot can acquire after reaching that point, calculated based on the unexplored area surrounding the leading edge; cost The effort required for the robot to reach the front edge x from its current position is represented by Euclidean distance as a measure of cost; RPF represents the repulsive potential field. , and These are weighting coefficients; In the Bayesian optimization described above, a Gaussian process model is trained using historical frontier points and their corresponding utility values. The utility value and uncertainty of the exploration point are predicted by sampling the posterior distribution of the Gaussian process. Step 4 uses a genetic algorithm to find the optimal allocation strategy for the currently available robots, including: Step 4.1) Obtain the available robot set R and the initial target point set X. Predict the utility value of each robot in R assigned to each target point in X using a Gaussian process and store it in the matrix utilities. Initialize the population, with each individual representing a task allocation scheme that assigns one target point to each robot in R. Step 4.2) Evaluate the suitability of each individual; the higher the suitability, the better the corresponding task allocation scheme. Calculate the fitness of individual h as follows: Where n is the number of robots in R; h[robot i ] represents the i-th robot in individual h. i The assigned target point; For robots i Assigned to target point h[robot] i The utility value; Penalty is the proximity penalty, which represents the penalty if the distance between the target points assigned to two robots is less than a set threshold; Step 4.3) Select individuals based on their fitness, perform crossover and mutation operations, and then generate a new generation of population; Step 4.4) Repeat steps 4.2 and 4.3 until the preset number of iterations is reached or the fitness meets the termination condition. At this point, the best individual in the population represents the optimal multi-robot task allocation scheme. Step 5: After receiving the assigned target point, the navigation module of each robot system plans the path from the current position to the target point, guides the robot to move along the path to the target point, and obtains the environmental data collected in real time by the sensors on the robot through the mapping module, and builds a local map by combining the localization technology. Step 6: Each robot system sends the constructed local map to the host system. The host system integrates the local maps into a global map through the map merging module, updates the area of ​​the detected region, and sends it to each robot system. Step 7: After the robot reaches the target point, continue to step 3.

2. The method according to claim 1, characterized in that, In step 3, the robot invokes the first front detection module and uses the adaptive fast exploration random tree algorithm to detect front points, including: Use the robot's current position as the initial node for the rapid exploration of the random tree T in the current round of exploration; first, adaptively adjust the growth step size of the random tree in the current round of exploration. ,as follows: in, This is the initial growth step size; The total area of ​​the exploration zone is obtained by the host system from the exploration mission; The current explored area is the total area. Before each round of exploration, the host system merges the explored areas and notifies each robot. In the first round of exploration... =0; Then perform the following steps: Step 3.1) Randomly sample a location point in the robot's exploration environment as a potential exploration target. Find the distance in T. The latest existing nodes , with step size λ from Towards New nodes growing in the direction If from the node To the node Without obstacles, the nodes Add to tree T if node Located at the forefront of the robot's known environment, the nodes Add to the frontier set; Step 3.2) Determine if the stopping condition has been met. If yes, stop the exploration and output the set of leading edges. Otherwise, proceed to step 3.

1. The stopping condition refers to the following: during the first round of exploration, a time length is preset, and exploration stops when the exploration reaches the preset time length; during non-first rounds of exploration, exploration stops when the target point is reached or approached.

3. The method according to claim 1 or 2, characterized in that, In step 3, the host calls the second frontier detection module and uses the adaptive fast exploration random tree algorithm to detect frontier points. The root node of the fast exploration random tree generated by the host is always the host position. In the first round of detection, frontier points are explored based on the environment that the host device can perceive. Starting from the second round of detection, frontier points are detected based on the currently explored global map.

4. The method according to claim 1 or 2, characterized in that, In step 4, during Bayesian optimization, frontier points that meet the threshold conditions are selected to form the initial target point set X based on preset utility value thresholds and uncertainty thresholds.

5. The method according to claim 1 or 2, characterized in that, In step 5, the navigation module uses the A* algorithm for path planning and the dynamic window method for dynamic obstacle avoidance.

6. The method according to claim 1 or 2, characterized in that, In step 6, the map merging module first aligns the local maps, determines the relative positional relationship between the local maps by finding common features and optimal transformations, and then performs map fusion.

Images