Mobile robot coverage path planning method based on obstacle awareness neural field

Abstract

The invention discloses a mobile robot coverage path planning method based on an obstacle awareness neural field. A coverage path planning process is modeled as a closed-loop evolution process of a neural activation field state in discrete time; the environment space is represented as a coarse-grained grid structure, each coarse grid unit corresponds to a nerve activation state, and the nerve activation state is jointly determined by an exploration excitation component, an obstacle perception suppression component and an access suppression component; performing motion direction competition decision according to the local nerve activation field in each control period; the monotonous propulsion constraint under the Lyapunov meaning is introduced, so that the system cannot form a stable balance state in a coverage uncompleted stage, and the coverage process is ensured to be continuously propelled until the coverage process is completed. According to the method, the convergence and the full coverage of path planning are ensured, the deadlock problem is avoided in a complex environment, the robot not only avoids redundant coverage in the exploration process, but also can efficiently complete global coverage, and the overall efficiency of path planning is improved.

Description

Technical Field

[0001] This invention relates to the field of robot autonomous navigation and path planning technology, specifically to a method for mobile robot coverage path planning based on obstacle perception neural fields. Background Technology

[0002] Coverage path planning is a fundamental technology for mobile robots performing tasks such as cleaning, inspection, and environmental monitoring. Its goal is to ensure that the robot's trajectory covers all accessible areas in the environment while avoiding collisions. Currently, the coverage path planning technologies closest to this application mainly include online coverage methods based on frontier exploration and coverage planning methods based on artificial potential fields or neural networks.

[0003] (I) Coverage path planning method based on cutting-edge exploration Existing coverage path planning methods typically rely on a frontier exploration mechanism. The technical approach involves constructing a local or global map using environmental perception sensors, identifying the boundary between known and unknown areas (i.e., the frontier area), using this frontier area as a temporary navigation target, and repeatedly executing path planning and motion control until the environment is fully explored. This type of method achieves coverage by continuously updating target points, but it suffers from the following shortcomings in practical applications: 1) This method usually requires frequent global or semi-global path planning, which has high computational complexity and is not conducive to real-time operation; 2) In narrow passages or environments with dense obstacles, the distribution of the leading edge area is unstable, which can easily cause the robot to switch repeatedly between adjacent leading edges, resulting in path redundancy; 3) When the locally accessible area is not fully covered, navigation may stall or become partially deadlocked.

[0004] The aforementioned problems result in low coverage efficiency and insufficient stability of this type of method in complex environments.

[0005] (II) Coverage path planning method based on artificial potential field Another type of coverage path planning method based on artificial potential fields typically involves constructing attractive and repulsive potentials in the environmental space, or building a discrete neural activation model to guide the robot to make motion decisions based on the potential field or activation values, thereby gradually completing the coverage. This type of method usually treats obstacles as impassable areas or only applies repulsive potentials at the obstacle boundaries. Its main shortcomings are: 1) The geometric influence of the space around the obstacle was not continuously modeled, resulting in unstable robot movement in narrow areas or near concave obstacles; 2) The lack of an effective suppression mechanism for visited areas makes it easy to generate duplicate paths during the coverage process; 3) The overall stability of the coverage process is not constrained, which can easily lead to a local deadlock before the coverage is completed.

[0006] (III) Common problems existing in the prior art In summary, existing coverage path planning methods generally suffer from the following technical problems in complex environments: 1) Insufficient modeling of the spatial impact of obstacles affects the safety of movement; 2) The path repetition rate is high during the coverage process, resulting in low coverage efficiency; 3) The lack of an effective deadlock avoidance mechanism makes it difficult to ensure the stable progress of the coverage process. Summary of the Invention

[0007] This invention provides a mobile robot coverage path planning method based on obstacle perception neural field to solve the technical problems of insufficient stability of the coverage process, easy path repetition and deadlock in the existing coverage path planning method in complex environment.

[0008] According to a first aspect, one embodiment provides a mobile robot coverage path planning method based on an obstacle-sensing neural field, the method comprising: Real-time acquisition of environmental perception data and construction of a two-dimensional occupancy cost map; The two-dimensional occupancy cost map is coarse-grained by merging multiple adjacent high-resolution grids into a coarse-grained grid cell. If any obstacle grid cell exists within the merged window, the corresponding coarse grid cell is determined to be an obstacle cell; otherwise, it is determined to be a passable cell. A neural activation value is constructed for each passable unit. The neural activation value is used to characterize the relative priority of the robot's movement toward the corresponding passable unit. The neural activation value is jointly determined by the exploration incentive component, the obstacle perception inhibition component, and the access inhibition component, forming a neural activation field. Within each control cycle, the robot makes competitive decisions on its motion direction based on the local neural activation field to obtain the next motion direction, and updates the state of the neural activation field in real time after the robot moves. In the process of updating the neural activation field, a coverage advancement constraint based on Lyapunov stability is introduced, which prevents the system from forming a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues to advance until the coverage is completed.

[0009] Furthermore, an obstacle perception inhibition component is constructed, specifically including: Let the coarse mesh set be The obstacle unit set is The set of passable units is For any free unit Define its distance to the nearest obstacle unit. for:

[0010] in Indicates the center coordinates of the corresponding grid cell; This refers to the center coordinates of the reference point o, which here represents the center coordinates of the nearest obstacle unit. To characterize the continuous influence of obstacles on their neighborhood space, based on distance... Define an obstacle perception inhibition value for each free unit g. Its distance Increases and monotonically decreases, obstacle perception inhibition value The functional form includes exponential decay, Gaussian decay, inverse proportional decay, or piecewise continuous function form, satisfying the condition that decays with distance. The continuity and monotonicity of change are not constrained.

[0011] Furthermore, an access inhibition component is constructed, specifically including: During the coverage process, maintain the set of visited cells. It is used to record the area that the robot has completed covering at time t; when the robot center enters a certain passable unit, or stays in the corresponding passable unit for a longer time than a preset threshold, the corresponding passable unit is determined to be an visited unit; Define the access inhibition value of free element g in the t-th control cycle. Its update rule is defined as follows:

[0012] in, This refers to the access suppression value of free unit g in the (t+1)th control cycle. The access suppression increment parameter is used; through the above update rules, access suppression always maintains its non-decreasing characteristic before the coverage is completed, thereby continuously reducing the attractiveness of the visited area in subsequent path decisions.

[0013] Furthermore, the neural activation value is jointly determined by the exploration stimulus component, the obstacle perception inhibition component, and the access inhibition component, specifically including: Define the neural activation value of free unit g at time t. for:

[0014] in: For the unvisited exploration incentive component, different function forms are defined according to different application scenarios and robot task requirements; For obstacle perception inhibition component, To access the suppressed component, The weighting parameter is used; thus, unvisited regions are given higher priority through positive incentives, while obstacle neighborhoods and visited regions are given lower priority through suppression terms.

[0015] Furthermore, within each control cycle, a competitive decision on the robot's motion direction is made based on the local neural activation field to obtain the next motion direction, and the state of the neural activation field is updated in real time after the robot moves. Specifically, this includes: Within each control cycle, a set of candidate motion directions is generated based on the robot's current pose. Each candidate direction Mapped to adjacent coarse mesh cells ; To improve motion smoothness, a steering penalty term is introduced:

[0016] in The current heading angle, For candidate heading angles, For penalty weighting; The overall score for candidate directions is defined as follows:

[0017] Choose the direction with the highest overall score as the next direction of movement:

[0018] in, This is the planned direction for the next phase of the movement.

[0019] Furthermore, a Lyapunov-based coverage advancement constraint is introduced during the neural activation field update process, preventing the system from reaching a stable equilibrium state during the incomplete coverage phase. This ensures the coverage process continues until completion, specifically including: To characterize the degree of incomplete coverage, a coverage evaluation metric is constructed. Define the coverage evaluation metric The format is:

[0020] in For the set of unvisited cells, The degree of unvisitedness is a function that depends on the position of the free unit g and the influence of obstacles, and its form includes linear, exponential, or Gaussian decay functions. During the incomplete coverage phase, the coverage evaluation is guaranteed through non-decreasing updates with access suppression, continuous incentives for unvisited regions, and non-negativity constraints with barrier suppression. satisfy: And it satisfies strict decreasing ordering when there are reachable unvisited units, where This indicates the extent of the unvisited region at the current time t. This indicates the extent of the unvisited region at the next time step t+1; thus, there is no stable equilibrium point in the system during the incomplete coverage phase, and the completed coverage state becomes the only allowed stable termination state of the system, thereby theoretically ensuring that the coverage process will not fall into local deadlock.

[0021] Furthermore, the method also includes: The robot's movement status is monitored in real time to detect deadlock. If a deadlock is detected, a local search algorithm is initiated or the path is replanned to prevent the robot from stalling and to resume the task.

[0022] According to a second aspect, one embodiment provides a mobile robot coverage path planning system based on an obstacle-sensing neural field, the system comprising: The environmental map building module is used to acquire environmental perception data in real time and build a two-dimensional occupancy cost map; The coarse-graining processing module is used to coarse-grain the two-dimensional occupancy cost map, merging multiple adjacent high-resolution grids into a coarse-grained grid unit. If any obstacle grid exists within the merged window, the corresponding coarse grid unit is determined to be an obstacle unit; otherwise, it is determined to be a passable unit. The neural activation field fusion modeling module is used to construct neural activation values ​​for each passable unit. The neural activation values ​​are used to characterize the relative priority of the robot's movement toward the corresponding passable unit. The neural activation values ​​are jointly determined by the exploration incentive component, the obstacle perception inhibition component, and the access inhibition component, forming a neural activation field. The direction competition decision module is used to make competitive decisions on the robot's motion direction based on the local neural activation field in each control cycle, obtain the next motion direction, and update the neural activation field state in real time after the robot moves. The coverage advancement constraint module is used to introduce Lyapunov stability-based coverage advancement constraints during the neural activation field update process, so that the system cannot form a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues to advance until the coverage is completed.

[0023] Furthermore, the system also includes: The deadlock detection and recovery module is used to monitor the robot's motion status in real time, detect whether a deadlock has occurred, and if a deadlock is detected, it will start using a local search algorithm or replan the path to prevent the robot from stopping and resume the covered task.

[0024] This invention provides a mobile robot coverage path planning method based on obstacle-sensing neural fields. By combining Lyapunov stability analysis, the inhibition potential of obstacle perception, and the exploration mechanism of information theory, it has the following beneficial effects: 1) Solving the deadlock problem: By introducing Lyapunov stability theory, this invention ensures the convergence and full coverage of path planning, and avoids the deadlock problem in complex environments.

[0025] 2) Improve obstacle avoidance capability: The obstacle suppression potential has a continuous impact on the entire environment through spatial decay, avoiding the oscillation phenomenon that is prone to occur in narrow or complex environments by traditional methods, thus improving the safety and efficiency of navigation.

[0026] 3) Improve exploration efficiency: The exploration mechanism of information theory enables robots to not only avoid redundant coverage during the exploration process, but also to efficiently complete global coverage, thereby improving the overall efficiency of path planning. Attached Figure Description

[0027] Figure 1 A flowchart illustrating a mobile robot coverage path planning method based on an obstacle-sensing neural field, as provided in one embodiment of the present invention; Figure 2 This is a schematic diagram of the hardware architecture in a mobile robot coverage path planning method based on an obstacle-sensing neural field according to an embodiment of the present invention. Figure 3 A diagram illustrating the working mechanism of a neural field in a mobile robot coverage path planning method based on an obstacle perception neural field, provided as an embodiment of the present invention. Figure 4 A flowchart of path planning in a mobile robot coverage path planning method based on obstacle perception neural field provided in one embodiment of the present invention; Figure 5 This is a diagram showing the full-coverage path planning result in a mobile robot coverage path planning method based on an obstacle-sensing neural field, provided as an embodiment of the present invention. Detailed Implementation

[0028] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0029] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0030] The first embodiment of this invention provides a mobile robot coverage path planning method based on obstacle perception neural fields, which is particularly suitable for real-time coverage path planning tasks of mobile robots such as quadruped robots and wheeled robots in unknown or partially known complex environments. The coverage path planning process is modeled as a closed-loop evolution process of the neural activation field state in discrete time. In this closed-loop process, the environmental space is represented as a coarse-grained grid structure, with each coarse grid cell corresponding to a neural activation state, which is jointly determined by exploration excitation components, obstacle perception inhibition components, and access inhibition components. The mobile robot makes directional competition decisions based on the local neural activation field in each control cycle and updates the neural activation field state in real time after robot movement. By introducing a Lyapunov-style monotonic propagation constraint during the neural activation field update process, the system cannot form a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues until completion. The following is a related analysis... Figure 1 Please provide a detailed explanation.

[0031] like Figure 1 As shown, in step S100, environmental perception data is acquired in real time, and a two-dimensional occupancy cost map is constructed.

[0032] In this embodiment, the mobile robot acquires environmental perception data through LiDAR and depth camera, and constructs a two-dimensional cost map to represent the accessibility status of each location in the environment.

[0033] like Figure 1 As shown, in step S200, the two-dimensional occupancy cost map is coarsened by merging multiple adjacent high-resolution grids into a coarse-grained grid cell. If any obstacle grid cell exists within the merged window, the corresponding coarse grid cell is determined to be an obstacle cell; otherwise, it is determined to be a passable cell.

[0034] In this embodiment, the cost map is coarsened based on the robot's maximum external dimensions, merging multiple adjacent high-resolution grids into a single coarse-grained grid cell. If any obstacle grid cell exists within the merged window, the corresponding coarse grid cell is determined to be an obstacle cell; otherwise, it is determined to be a passable cell, thereby ensuring geometric feasibility for the robot to move within a single coarse grid cell.

[0035] like Figure 1 As shown, in step S300, a neural activation value is constructed for each passable unit. The neural activation value is used to characterize the relative priority of the robot's movement toward the corresponding passable unit. The neural activation value is jointly determined by the exploration incentive component, the obstacle perception inhibition component, and the access inhibition component, forming a neural activation field.

[0036] The above steps specifically include: S310, Construction of the obstacle perception inhibition field: Let the coarse mesh set be The obstacle unit set is The set of passable units is For any free unit Define its distance to the nearest obstacle unit. for:

[0037] in Indicates the center coordinates of the corresponding grid cell; This refers to the center coordinates of the reference point o, which here represents the center coordinates of the nearest obstacle unit. To characterize the continuous influence of obstacles on its neighborhood space, an obstacle perception inhibition value is defined for each free unit g. Its distance As it increases and monotonically decreases, the obstacle perception inhibition value exhibits an exponential decay.

[0038] in To suppress the amplitude parameter, Let be the decay rate parameter. The above form indicates that when a free element is close to an obstacle, its suppression value is relatively large; as the distance from the obstacle gradually increases, the suppression value continuously and smoothly decays and approaches zero. The suppression function can also be a Gaussian decay, inverse proportional decay, or a piecewise continuous function, as long as it satisfies the constraints of continuity and monotonically non-increasing with distance.

[0039] S320, access suppression mechanism: During the coverage process, maintain the set of visited cells. It is used to record the area that the robot has completed covering at time t; when the robot center enters a certain passable unit, or stays in the corresponding passable unit for a longer time than a preset threshold, the corresponding passable unit is determined to be an visited unit; Define the access inhibition value of free element g in the t-th control cycle. Its update rule is defined as follows:

[0040] in, This refers to the access suppression value of free unit g in the (t+1)th control cycle. To suppress incremental parameters for access; update rules are based on whether the free element g belongs to the access region. If it belongs to the category, the access suppression value is increased by an increment. This reflects that the cell has been visited, thus reducing its attractiveness in subsequent path decisions. Through these update rules, access suppression maintains its non-decreasing characteristic until coverage is complete, thereby continuously reducing the attractiveness of visited areas in subsequent path decisions.

[0041] S330, neural activation field fusion modeling: Define the neural activation value of free unit g at time t. for:

[0042] in: For the unvisited exploration incentive component, different function forms are defined according to different application scenarios and robot task requirements; For obstacle perception inhibition component, To access the suppressed component, The above fusion form represents the following: unvisited regions are given higher priority through positive incentives, while obstacle neighborhoods and visited regions are given lower priority through suppression terms.

[0043] In this embodiment, Primarily used to represent the exploratory stimuli of a robot on a free unit g, the form of the stimuli function is usually determined based on the distance from the free unit g to the target area or other environmental characteristics. For example, the stimuli function can be a function related to the Euclidean distance from g to the target point, or it can be a function based on environmental complexity. Therefore, the specific form of R(g,t) can be designed according to specific circumstances to ensure that it can effectively guide the robot to prioritize the exploration of uncovered areas in a coverage task, while avoiding obstacles affecting the choice of motion path.

[0044] like Figure 1 As shown, in step S400, the robot motion direction competition decision is made based on the local neural activation field in each control cycle to obtain the next motion direction, and the neural activation field state is updated in real time after the robot moves.

[0045] In this embodiment, within each control cycle, a set of candidate motion directions is generated based on the robot's current pose. Each candidate direction Mapped to adjacent coarse mesh cells ; To improve motion smoothness, a steering penalty term is introduced:

[0046] in The current heading angle, For candidate heading angles, For penalty weighting; The overall score for candidate directions is defined as follows:

[0047] Choose the direction with the highest overall score as the next direction of movement:

[0048] like Figure 1 As shown, in step S500, a coverage advancement constraint based on Lyapunov stability is introduced during the neural activation field update process, so that the system cannot form a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues to advance until the coverage is completed.

[0049] In this embodiment, the coverage propulsion constraint based on Lyapunov stability is as follows: To characterize the degree of incomplete coverage, a coverage evaluation metric is constructed. Define the coverage evaluation metric The format is:

[0050] in For the set of unvisited cells; The degree of unvisitedness is a function that depends on the position of the free unit g and the influence of obstacles. Its form can be linear, exponential, or Gaussian decaying function, and the specific form will vary depending on different environments and task requirements. During the incomplete coverage phase, the coverage evaluation is guaranteed through non-decreasing updates with access suppression, continuous incentives for unvisited regions, and non-negativity constraints with barrier suppression. satisfy: And it satisfies strict decreasing ordering when there are reachable unvisited units, where This indicates the extent of the unvisited region at the current time t. This indicates the extent of the unvisited region at the next time step t+1; thus, there is no stable equilibrium point in the system during the incomplete coverage phase, and the completed coverage state becomes the only allowed stable termination state of the system, thereby theoretically ensuring that the coverage process will not fall into local deadlock.

[0051] In this embodiment, the method further includes: engineering-level deadlock detection and recovery. In engineering implementation, when a robot displacement change, access set, or activation distribution remains unchanged over multiple control cycles, a deadlock risk is determined. At this time, the robot is guided to the nearest reachable unvisited unit to continue the neural field coverage process. This recovery mechanism, as an engineering supplement, does not violate the monotonic propulsion constraint based on Lyapunov stability.

[0052] To implement the obstacle-sensing neural field-based mobile robot coverage path planning method of this embodiment, the hardware architecture, main functional modules, and workflow are as follows: 1. Hardware architecture, such as Figure 2 As shown 1) Sensor module: LiDAR: Used to acquire depth information of the surrounding environment and generate an environmental occupancy map in real time. LiDAR is one of the key sensors in this invention, providing high-precision obstacle detection.

[0053] Depth cameras: provide supplemental perception in indoor environments, enhance the capture of environmental details, and help generate higher quality environmental data.

[0054] Inertial Measurement Unit (IMU): Used to provide motion state information of the robot, including data such as acceleration and angular velocity, to help improve the accuracy of path planning.

[0055] 2) Calculation unit: Computing platform: Equipped with a main control processor, it performs tasks such as path planning, neural field computation, and deadlock detection. The computing platform has real-time processing capabilities, ensuring the robot can adjust its path in real time in dynamic environments.

[0056] 3) Actuator module: Motor drive system: The main component controlling the robot's movement, including the motion drive system for wheeled, tracked, or quadruped robots. The motor receives commands from the control output module to achieve precise control of the robot.

[0057] Servo control system: For quadruped robots and the like, the servo control system provides joint drive to ensure the robot's flexible movement and stability.

[0058] 2. Functional Modules The workflow of the method in this embodiment can be implemented through the following functional modules: Map Acquisition Module: This module is responsible for acquiring environmental cost maps in real-time or offline via sensors and converting them into an occupancy grid map for the robot. It primarily processes environmental perception information, providing the foundational data for subsequent path planning and obstacle avoidance.

[0059] The coarse-grained modeling module coarsely processes the acquired environmental cost map, merging multiple high-resolution grids into a single coarse-grained mesh cell based on the robot's dimensions. This reduces computational complexity while ensuring the robot's path planning executes correctly in the real-world environment.

[0060] Obstacle Perception Module: This module calculates the obstacle perception suppression value for each grid cell using an obstacle perception algorithm and generates the corresponding obstacle perception field. Based on the position, shape, and distance of obstacles, the obstacle perception module determines the potential impact of obstacles on the robot's movement, ensuring that the robot avoids obstacle areas.

[0061] Access Suppression Module: This module maintains a set of visited cells to ensure that the robot does not perform duplicate path calculations within already visited areas. After each visit, the system updates the access suppression value, reducing the activation value of the visited area.

[0062] Neural Activation Field Module: This module generates a neural activation field based on the current obstacle perception inhibition value, unvisited stimuli, and visited inhibition, such as... Figure 3 As shown, the neural activation field calculates the activation value of each grid cell based on the weights and activation functions, thus determining the robot's optimal motion direction at each moment.

[0063] Path planning module: This module generates the optimal path based on the activation values ​​output by the neural activation field and the target information. The path planning module is responsible for providing the robot with the optimal path selection, ensuring that the robot can effectively cover the target area and avoid collisions and path redundancy.

[0064] Control Output Module: The path planning results output by the path planning module are transmitted to the control output module, which is responsible for converting the planned path into control commands that the robot can execute. The control commands drive actuators (such as motors and servos) to perform actual physical movements.

[0065] Deadlock Detection and Recovery Module: This module monitors the robot's motion status in real time and detects whether a deadlock has occurred. If a deadlock is detected, the recovery module will be activated, using a local search algorithm or replanning the path to prevent the robot from stalling and to resume its assigned tasks.

[0066] 3. Work Process The overall workflow is as follows Figure 4 As shown, the path planning task is completed through the collaboration of various modules: 1) Environmental map acquisition: Collect surrounding environmental data in real time through sensors (LiDAR, depth camera) to build an occupancy map.

[0067] 2) Coarse-grained processing: Convert high-resolution maps into coarse-grained meshes to reduce computational complexity and meet the actual motion requirements of robots.

[0068] 3) Obstacle perception and inhibition: Calculate the perception inhibition value of obstacles and generate an obstacle perception field.

[0069] 4) Activation Field Calculation and Path Generation: Based on the neural activation field model, the activation value of each grid cell is calculated, and the optimal path is generated. The full-coverage path planning result is shown in the figure below. Figure 5 As shown.

[0070] 5) Motion control command output: Convert the generated path plan into execution commands to drive the robot to move.

[0071] 6) Real-time deadlock detection and recovery: During the path planning process, deadlock is monitored. Once a deadlock occurs, a recovery mechanism is triggered to replan the path.

[0072] Specific operation example: Taking a mobile robot equipped with a two-dimensional LiDAR performing coverage work in an indoor environment as an example, the specific implementation process of the method of the present invention is described. The robot acquires environmental perception data in real time through the two-dimensional LiDAR and constructs a two-dimensional occupancy cost map. The resolution of the cost map is set to 0.05m, and the cost map is coarsened according to the maximum external dimensions of the robot, with the side length of the coarse-grained grid cell set to 0.3m. During the construction of the obstacle perception suppression field, for each free coarse grid cell g, the distance d(g) to the nearest obstacle cell is calculated, and the obstacle perception suppression value is calculated using the following exponential decay form:

[0073] in During the coverage process, when the robot first enters a coarse grid cell, the cell is marked as visited, and the access inhibition value is updated according to the following rules:

[0074] in During the neural activation field fusion phase, the activation value is calculated using the following method:

[0075] in The robot generates eight candidate motion directions in each control cycle and selects the direction with the largest activation value as the next motion direction, incorporating a turning penalty term. Experimental results show that, without global path planning, the robot can stably complete a high-coverage traversal of the indoor environment, and no local deadlock occurs during the incomplete coverage phase.

[0076] Experimental data: Real-world experiments show that, compared to traditional methods such as ZCC, FBE, and HPF, the method of this invention achieves a faster 99% coverage completion time, significantly reduces path redundancy, and avoids deadlock. Specific data are shown in the table below:

[0077] Corresponding to the aforementioned method for mobile robot coverage path planning based on obstacle-sensing neural fields, this invention also discloses a mobile robot coverage path planning system based on obstacle-sensing neural fields, which specifically includes: The environmental map building module is used to acquire environmental perception data in real time and build a two-dimensional occupancy cost map; The coarse-graining processing module is used to coarse-grain the two-dimensional occupancy cost map, merging multiple adjacent high-resolution grids into a coarse-grained grid unit. If any obstacle grid exists within the merged window, the corresponding coarse grid unit is determined to be an obstacle unit; otherwise, it is determined to be a passable unit. The neural activation field fusion modeling module is used to construct neural activation values ​​for each passable unit. The neural activation values ​​are used to characterize the relative priority of the robot's movement toward the corresponding passable unit. The neural activation values ​​are jointly determined by the exploration incentive component, the obstacle perception inhibition component, and the access inhibition component, forming a neural activation field. The direction competition decision module is used to make competitive decisions on the robot's motion direction based on the local neural activation field in each control cycle, obtain the next motion direction, and update the neural activation field state in real time after the robot moves. The coverage advancement constraint module is used to introduce Lyapunov stability-based coverage advancement constraints during the neural activation field update process, so that the system cannot form a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues to advance until the coverage is completed.

[0078] Furthermore, the system also includes: The deadlock detection and recovery module is used to monitor the robot's motion status in real time, detect whether a deadlock has occurred, and if a deadlock is detected, it will start using a local search algorithm or replan the path to prevent the robot from stopping and resume the covered task.

[0079] It should be noted that for a detailed description of the mobile robot coverage path planning system based on obstacle perception neural field provided in the embodiments of the present invention, please refer to the relevant description of the mobile robot coverage path planning method based on obstacle perception neural field provided in the embodiments of the present invention, which will not be repeated here.

[0080] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for traversing a mobile robot based on an obstacle-sensing neural field, characterized in that, The method includes: Real-time acquisition of environmental perception data and construction of a two-dimensional occupancy cost map; The two-dimensional occupancy cost map is coarse-grained by merging multiple adjacent high-resolution grids into a coarse-grained grid cell. If any obstacle grid cell exists within the merged window, the corresponding coarse grid cell is determined to be an obstacle cell; otherwise, it is determined to be a passable cell. A neural activation value is constructed for each passable unit. The neural activation value is used to characterize the relative priority of the robot's movement toward the corresponding passable unit. The neural activation value is jointly determined by the exploration incentive component, the obstacle perception inhibition component, and the access inhibition component, forming a neural activation field. Within each control cycle, the robot makes competitive decisions on its motion direction based on the local neural activation field to obtain the next motion direction, and updates the state of the neural activation field in real time after the robot moves. In the process of updating the neural activation field, a coverage advancement constraint based on Lyapunov stability is introduced, which prevents the system from forming a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues to advance until the coverage is completed.

2. The method for mobile robot coverage path planning based on obstacle perception neural field as described in claim 1, characterized in that, Constructing obstacle perception inhibition components, specifically including: Let the coarse mesh set be The obstacle unit set is The set of passable units is For any free unit Define its distance to the nearest obstacle unit. for: in Indicates the center coordinates of the corresponding grid cell; This refers to the center coordinates of the reference point o, which here represents the center coordinates of the nearest obstacle unit. To characterize the continuous influence of obstacles on their neighborhood space, based on distance... Define an obstacle perception inhibition value for each free unit g. Its distance Increases and monotonically decreases, obstacle perception inhibition value The functional form includes exponential decay, Gaussian decay, inverse proportional decay, or piecewise continuous function form, satisfying the condition that decays with distance. The continuity and monotonicity of change are not constrained.

3. The method for mobile robot coverage path planning based on obstacle perception neural field as described in claim 2, characterized in that, Constructing access suppression components specifically includes: During the coverage process, maintain the set of visited cells. It is used to record the area that the robot has completed covering at time t; when the robot center enters a certain passable unit, or stays in the corresponding passable unit for a longer time than a preset threshold, the corresponding passable unit is determined to be an visited unit; Define the access inhibition value of free element g in the t-th control cycle. Its update rule is defined as follows: in, This refers to the access suppression value of free unit g in the (t+1)th control cycle. The access suppression increment parameter is used; through the above update rules, access suppression always maintains its non-decreasing characteristic before the coverage is completed, thereby continuously reducing the attractiveness of the visited area in subsequent path decisions.

4. The method for mobile robot coverage path planning based on obstacle perception neural field as described in claim 3, characterized in that, The neural activation values ​​are jointly determined by the exploration stimulus component, the obstacle perception inhibition component, and the access inhibition component, specifically including: Define the neural activation value of free unit g at time t. for: in: For the unvisited exploration incentive component, different function forms are defined according to different application scenarios and robot task requirements; For obstacle perception inhibition component; To access the suppressed component; The weighting parameter is used; thus, unvisited regions are given higher priority through positive incentives, while obstacle neighborhoods and visited regions are given lower priority through suppression terms.

5. The method for mobile robot coverage path planning based on obstacle perception neural field as described in claim 4, characterized in that, Within each control cycle, the robot makes competitive decisions about its motion direction based on the local neural activation field to obtain the next motion direction. The neural activation field state is then updated in real-time after the robot moves. Specifically, this includes: Within each control cycle, a set of candidate motion directions is generated based on the robot's current pose. Each candidate direction Mapped to adjacent coarse mesh cells ; To improve motion smoothness, a steering penalty term is introduced. : in The current heading angle, For candidate heading angles, For penalty weighting; Overall score of candidate directions Defined as: Choose the direction with the highest overall score as the next direction of movement: in, This is the planned direction for the next phase of the movement.

6. The method for mobile robot coverage path planning based on obstacle perception neural field as described in claim 1, characterized in that, In the neural activation field update process, a Lyapunov-based coverage advancement constraint is introduced to prevent the system from reaching a stable equilibrium state during the incomplete coverage phase, thereby ensuring the coverage process continues until completion. Specifically, this includes: To characterize the degree of incomplete coverage, a coverage evaluation metric is constructed. Define the coverage evaluation metric The format is: in For the set of unvisited cells; The degree of unvisitedness of free cell g at time t depends on the position of free cell g and the influence of obstacles, and its form includes linear, exponential or Gaussian decay functions; During the incomplete coverage phase, the coverage evaluation is guaranteed through non-decreasing updates with access suppression, continuous incentives for unvisited regions, and non-negativity constraints with barrier suppression. satisfy: And it satisfies strict decreasing ordering when there are reachable unvisited units, where This indicates the extent of the unvisited region at the current time t. This indicates the extent of the unvisited region at the next time step t+1; thus, there is no stable equilibrium point in the system during the incomplete coverage phase, and the completed coverage state becomes the only allowed stable termination state of the system, thereby theoretically ensuring that the coverage process will not fall into local deadlock.

7. The method for mobile robot coverage path planning based on obstacle perception neural field as described in claim 1, characterized in that, The method further includes: The robot's movement status is monitored in real time to detect deadlock. If a deadlock is detected, a local search algorithm is initiated or the path is replanned to prevent the robot from stalling and to resume the task.

8. A mobile robot coverage path planning system based on obstacle perception neural field, characterized in that, The system includes: The environmental map building module is used to acquire environmental perception data in real time and build a two-dimensional occupancy cost map; The coarse-graining processing module is used to coarse-grain the two-dimensional occupancy cost map, merging multiple adjacent high-resolution grids into a coarse-grained grid unit. If any obstacle grid exists within the merged window, the corresponding coarse grid unit is determined to be an obstacle unit; otherwise, it is determined to be a passable unit. The neural activation field fusion modeling module is used to construct neural activation values ​​for each passable unit. The neural activation values ​​are used to characterize the relative priority of the robot's movement toward the corresponding passable unit. The neural activation values ​​are jointly determined by the exploration incentive component, the obstacle perception inhibition component, and the access inhibition component, forming a neural activation field. The direction competition decision module is used to make competitive decisions on the robot's motion direction based on the local neural activation field in each control cycle, obtain the next motion direction, and update the neural activation field state in real time after the robot moves. The coverage advancement constraint module is used to introduce Lyapunov stability-based coverage advancement constraints during the neural activation field update process, so that the system cannot form a stable equilibrium state in the incomplete coverage stage, thereby ensuring that the coverage process continues to advance until the coverage is completed.

9. A mobile robot coverage path planning system based on obstacle perception neural field according to claim 8, characterized in that, The system also includes: The deadlock detection and recovery module is used to monitor the robot's motion status in real time, detect whether a deadlock has occurred, and if a deadlock is detected, it will start using a local search algorithm or replan the path to prevent the robot from stopping and resume the covered task.

Images