An autonomous decision-making method for the operational mode of an amphibious robot

Abstract

This invention relates to the field of robot control technology and discloses an autonomous decision-making method for the operational form of an amphibious robot. The method includes: reconstructing a spatiotemporal field from the robot's multi-source environmental perception data to obtain a multi-layered environmental field; analyzing the kinematic characteristics of the robot's land movement and aquatic movement to obtain the robot's reachable motion parameters; constructing a scalar field from the multi-layered environmental field and reachable motion parameters to obtain the robot's form selection potential field; injecting a task potential into the form selection potential field based on the robot's operational task objective to obtain the robot's global decision field; performing potential field optimization analysis on the global decision field to obtain the robot's dominant potential field form; and encoding the dominant potential field form to obtain the robot's optimal form control command. This invention can improve the efficiency of autonomous decision-making for the operational form of an amphibious robot.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and in particular to an autonomous decision-making method for the operational mode of an amphibious robot. Background Technology

[0002] Amphibious robots require precise environmental perception to autonomously switch between different environmental forms, adapting to complex and dynamic scenarios on land and underwater. Existing robot perception systems rely on fixed patterns and static thresholds for environmental identification, lacking the ability to deeply analyze and dynamically model multi-dimensional environmental data. This results in an inability to effectively capture dynamic changes in various environmental elements, insufficient perception accuracy and completeness, and difficulty in providing reliable data support for form switching. This technology addresses these shortcomings by fusing multi-source perception data and using dynamic modeling, thus strengthening the foundation for autonomous form decision-making.

[0003] Limited by the aforementioned perceptual deficiencies, existing amphibious robot form decision-making systems have significant shortcomings. Their environmental perception, motion parameters, and task objectives are disconnected, lacking an integrated and unified decision-making model, and failing to consider environmental conditions, hardware motion constraints, and operational task requirements. This results in insufficient adaptability and accuracy in form-switching decisions, making it difficult to balance environmental adaptation, motion safety, and operational efficiency, significantly reducing autonomous operation capabilities in complex scenarios. This technology constructs a multi-constraint collaborative decision-making mechanism, breaking down the barriers between perception, motion, and tasks, effectively improving the robot's autonomous form decision-making efficiency and scenario adaptability. Summary of the Invention

[0004] This invention provides an autonomous decision-making method for the operational mode of an amphibious robot, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an autonomous decision-making method for the operational mode of an amphibious robot, comprising: S1. Reconstruct the spatiotemporal field of the robot's multi-source environmental perception data to obtain the robot's multi-layer environmental field; S2. Analyze the kinematic characteristics of the robot's land movement and aquatic movement to obtain the robot's achievable motion parameters; S3. Construct a scalar field from the multi-layered environmental field and the reachable motion parameters to obtain the morphological selection potential field of the robot; S4. Based on the robot's task objective, inject task potential into the morphology selection potential field to obtain the robot's global decision field. S5. Perform potential field optimization analysis on the global decision field to obtain the dominant potential field form of the robot; S6. Perform morphological encoding on the dominant potential field to obtain the optimal morphological control command for the robot.

[0006] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention achieves deep and systematic coupling between environmental perception and robot motion capabilities by constructing an integrated decision-making platform that combines multi-dimensional environmental features and motion constraints. Relying on the fusion and reconstruction of multi-layered environmental fields and motion parameters, it can accurately adapt to the dynamic environmental characteristics and robot hardware motion limits of complex water and land scenarios. This significantly improves the matching accuracy between robot form-switching decisions and the actual working environment and its own hardware capabilities, making robot form control more closely aligned with real-world working conditions and fundamentally ensuring the stability and adaptability of cross-environment form-switching.

[0007] 2. This invention, relying on a task potential injection and dynamic potential field optimization mechanism, achieves a deep integration of task requirements and autonomous morphological decision-making. By dynamically adjusting the priority of morphological decisions and combining them with device actuator constraints to generate precise control commands, it can achieve multi-objective collaborative optimization of environmental adaptation, safe movement, and efficient operation. This effectively improves the intelligence level and overall operational efficiency of the robot's autonomous operation in complex scenarios, enabling stable adaptation to various autonomous operation tasks in dynamic and complex water and land scenarios, and achieving a comprehensive improvement in overall operational performance. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an autonomous decision-making method for the operational mode of an amphibious robot according to an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0009] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0010] This application provides a method for autonomous decision-making regarding the operational form of an amphibious robot. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for autonomous decision-making regarding the operational form of an amphibious robot can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0011] Reference Figure 1 The diagram shown is a flowchart illustrating an autonomous decision-making method for the operational form of an amphibious robot according to an embodiment of the present invention. In this embodiment, the autonomous decision-making method for the operational form of the amphibious robot includes: S1. Reconstruct the spatiotemporal field of the robot's multi-source environmental perception data to obtain the robot's multi-layer environmental field; The source of each data signal in the robot's multi-source environmental perception data is identified and the data type is analyzed. Based on the signal transmission channel number and the data encoding format, the mixed data is separated into several independent data streams that do not interfere with each other. Each independent data stream contains only a single type of environmental information.

[0012] For each independent data stream, using the robot's current 3D coordinates as a spatial reference point, the entire spatial range within 50 meters in front, 50 meters behind, 30 meters to the left, 30 meters to the right, 20 meters above, and 20 meters below the robot is divided into several small cubes with a grid side length of 0.05 meters. Each small cube is assigned a unique 3D coordinate number. Based on the spatial coordinates carried by each data point in the independent data stream, the value of the data point is written into the small cube with the corresponding 3D coordinate number. When multiple data points are mapped to the same small cube, the arithmetic mean of all data points is taken as the final stored value of the small cube. After processing all independent data streams in the above manner, each small cube stores a multi-dimensional numerical vector. Each dimension of this vector corresponds to the environmental information value of an independent data stream at that spatial location. All small cubes carrying multi-dimensional numerical vectors are arranged in order according to their 3D coordinate numbers to form a discretized field.

[0013] In one feasible implementation, for the multidimensional numerical vectors stored in each small cube in the discretized field, features are extracted according to the type of environmental information corresponding to each dimension value.

[0014] Specifically, for terrain height information, the maximum, minimum, and average values ​​of that dimension in all small cubes are extracted as statistical indicators of terrain height features; for water depth information, the set of locations in all small cubes where the value of that dimension is greater than 0.05 meters and the distribution range of water depth values ​​are extracted as statistical indicators of water depth features; for obstacle information, the set of locations in all small cubes where the value of that dimension is greater than zero and the corresponding value are extracted as statistical indicators of obstacle features; the statistical indicators of the above environmental features are classified and organized according to physical attributes. Terrain height features are categorized into length units and their physical units are noted as meters; water depth features are categorized into length units and their physical units are noted as meters; obstacle features are categorized into dimensionless units and their physical units are noted as percentages. The name, unit category, physical unit, and statistical indicator value of each environmental feature are summarized into a structured data table, which is the environmental parameter dimension.

[0015] In one feasible implementation, semantic labels are assigned to each small cube in the discretized field according to the classification rules of various environmental features defined in the environmental parameter dimensions.

[0016] Specifically, the process reads the multidimensional numerical vector stored in a small cube, extracts the values ​​of water depth and terrain height from this vector, first determines if the water depth value is greater than 0.2 meters. If it is, the small cube is labeled "Water Area" as the primary semantic tag, and the specific water depth value is recorded as auxiliary information. If it is not greater than 0.2 meters, the process further determines the rate of change of the terrain height value in space. Using this small cube as the center, the terrain height values ​​of the twenty-six adjacent small cubes are differentially calculated. If the difference is... If the maximum rate of change is less than 5%, it is labeled as "flat land" as the main semantic label. If the maximum rate of change is between 5% and 30%, it is labeled as "gentle slope land" as the main semantic label. If the maximum rate of change is greater than 30%, it is labeled as "steep slope land" as the main semantic label. Then, the obstacle dimension value is extracted from the multidimensional numerical vector. If the obstacle dimension value is greater than 30%, "contains obstacle" is added to the small cube as an auxiliary semantic label. All small cubes in the discretized field are traversed according to the above rules. After completing the semantic labeling, the semantic label field is obtained.

[0017] Each small cube in the discretized field is matched one-to-one with the semantic label with the same three-dimensional coordinate number in the semantic label field. The multi-dimensional numerical vector stored in each small cube and the corresponding main semantic label and auxiliary semantic label are combined into a new data unit. This data unit contains both the original environmental numerical information and semantic description information. All the new data units of the small cubes are arranged in order according to the three-dimensional coordinate number to form a multi-layer environmental field.

[0018] S2. Analyze the kinematic characteristics of the robot's land movement and aquatic movement to obtain the robot's achievable motion parameters; The overall configuration of the land-based mobility morphology is structurally decomposed into a legged walking mechanism, a wheeled rolling mechanism, a fuselage support frame, and attitude adjustment joints. The component composition of each mechanism, the connection method between the components, and the relative motion relationship of each component during movement are recorded one by one. At the same time, the overall configuration of the aquatic mobility morphology is structurally decomposed into an underwater propulsion unit, a buoyancy adjustment chamber, attitude control wing surfaces, and a sealed fuselage shell. The component composition of each mechanism, the connection method between the components, and the relative motion relationship of each component during movement are recorded one by one. The morphological feature description is obtained by summarizing the structural decomposition results of the above two morphologies.

[0019] In one feasible implementation, the robot's motion performance indicators are obtained through actual measurement based on the structural characteristics and motion modes of each motion mechanism recorded in the morphological feature description.

[0020] Specifically, in a flat test area 100 meters long and 5 meters wide, the robot moved from the starting point to the end point using both legged walking and wheeled rolling methods. Each test was repeated five times, and the average speed was taken as the maximum forward speed on land for both legged walking and wheeled rolling. The larger value was selected as the maximum forward speed for land movement. At the same time, the turning radius and steering response delay time were determined by the maximum deflection angle and response time of the attitude adjustment joints. The same method was used to test each mechanism of the aquatic movement mode to obtain the maximum thrust output value of the underwater thruster, the maximum buoyancy adjustment rate of the buoyancy adjustment cabin, and the maximum deflection angle of the attitude control wing surface. All test calculation results were classified and summarized according to motion type and performance index name to obtain the kinematic properties.

[0021] Based on the specific values ​​of various kinematic performance indicators of the land mobility morphology recorded in the kinematic attributes, the kinematic limit values ​​of the land mobility morphology in each performance dimension are determined. The maximum forward speed is taken as the measured maximum value in the kinematic attributes. The vertical obstacle crossing height is measured in the test field starting from a height of 0.05 meters and increasing by 0.05 meters in step length. The highest height successfully crossed is taken as the limit value. The climbing ability is measured starting from a 5-degree inclination angle and increasing by 2.5 degrees in step length. The highest angle successfully climbed is taken as the limit value. The turning flexibility is determined by measuring the shortest path diameter required to complete a 360-degree turn in place.

[0022] Based on the specific values ​​of various kinematic performance indicators of aquatic movement patterns recorded in the kinematic attributes, the kinematic limit values ​​of aquatic movement patterns in each performance dimension are determined. The maximum submersible speed is taken as the measured maximum value in the kinematic attributes. The maximum diving depth is measured in a depth-controlled pool starting from one meter and increasing in one-meter increments. The maximum water depth that maintains normal movement is taken as the limit value. The surfacing speed is calculated by measuring the time it takes to rise to the surface after releasing buoyancy from the maximum diving depth. The turning radius is determined by measuring the minimum turning diameter required to complete a 180-degree turn. After classifying and summarizing all the above kinematic limit values ​​according to performance dimensions, the capability boundary characterization is obtained.

[0023] The measured values ​​of each kinematic performance index are nonlinearly proportionally transformed to the corresponding kinematic limit values ​​in the capability boundary representation. This ensures that the transformed values ​​change smoothly within the allowable range of the kinematic limit values. For dimensions where the measured values ​​are much lower than the limit values, the transformed values ​​remain close to the measured values. For dimensions where the measured values ​​are close to or reach the limit values, the transformed values ​​gradually approach the limit values ​​while retaining a safety margin. The safety margin is controlled by a safety margin adjustment factor ranging from 0.05 to 0.15. The larger the value, the larger the safety margin. The achievable kinematic parameters are obtained by summarizing the transformed values ​​of all performance dimensions according to the performance dimension name and kinematic type.

[0024] S3. Construct a scalar field from the multi-layered environmental field and the reachable motion parameters to obtain the morphological selection potential field of the robot; The multidimensional numerical vectors stored in each small cube in the multi-layered environmental field are traversed and read to extract the specific values ​​of terrain height, water depth, and obstacle dimensions. The extracted values ​​of the three dimensions are recorded according to the three-dimensional coordinate numbers of the small cube. The three-dimensional values ​​extracted from all small cubes are arranged in order according to the coordinate numbers to form environmental feature parameters.

[0025] In one feasible implementation, the passability of land movement patterns is determined point by point based on the terrain height value of each spatial location in the environmental feature parameters and the rate of change of the terrain height value between adjacent spatial locations.

[0026] Specifically, taking a certain spatial location as the center, the terrain height values ​​of 26 spatial locations adjacent to this spatial location in three-dimensional space are selected. The average of the absolute values ​​of the differences between the terrain height values ​​of these 26 spatial locations and the terrain height value of the central spatial location is calculated. This average value is used as the terrain change rate characterization value of the central spatial location. If the terrain height value of the central spatial location is greater than zero and the terrain change rate characterization value is less than 5%, then the terrain passability calibration value of this spatial location for land movement patterns is set to 1.0. If the terrain height value of the central spatial location is greater than zero and the terrain change rate characterization value is between 5% and 30%, then the terrain passability calibration value of this spatial location for land movement patterns is set to 1.0. The terrain passability calibration value of the spatial location relative to the land movement pattern is set to 0.6. If the terrain height value of the central spatial location is greater than zero and the terrain change rate characterization value is greater than 30%, then the terrain passability calibration value of the spatial location relative to the land movement pattern is set to 0.2. If the terrain height value of the central spatial location is equal to zero, then the terrain passability calibration value of the spatial location relative to the land movement pattern is set to 0.0. According to the above judgment rules, all spatial locations in the environmental characteristic parameters are calculated one by one to obtain the terrain passability calibration value corresponding to each spatial location. The terrain passability calibration values ​​of all spatial locations are arranged in order according to the three-dimensional coordinate number to form the terrain passability calibration.

[0027] In one feasible implementation, the passability of aquatic movement patterns is determined point by point based on the water depth value at each spatial location in the environmental characteristic parameters.

[0028] Specifically, the water depth value at a certain spatial location is read. If the water depth value at that spatial location is less than 0.05 meters, the hydrological transitivity calibration value for that spatial location relative to aquatic movement patterns is set to 0.0. If the water depth value at that spatial location is greater than or equal to 0.05 meters and less than 0.3 meters, the hydrological transitivity calibration value for that spatial location relative to aquatic movement patterns is set to 0.4. If the water depth value at that spatial location is greater than or equal to 0.3 meters and less than 1.0 meters, the hydrological transitivity calibration value for that spatial location relative to aquatic movement patterns is set to 0.7. If the water depth value at that spatial location is greater than or equal to 1.0 meters, the hydrological transitivity calibration value for that spatial location relative to aquatic movement patterns is set to 1.0. The above judgment rules are applied to all spatial locations in the environmental characteristic parameters one by one to obtain the hydrological transitivity calibration value corresponding to each spatial location. The hydrological transitivity calibration values ​​of all spatial locations are arranged in order according to the three-dimensional coordinate number to form the hydrological transitivity calibration.

[0029] The topographic accessibility calibration value of each spatial location in the topographic accessibility calibration is paired with the hydrological accessibility calibration value of the same three-dimensional coordinate number in the hydrological accessibility calibration. After each spatial location is paired, a set of two-dimensional vectors containing the topographic accessibility calibration value and the hydrological accessibility calibration value is obtained. The two-dimensional vectors of all spatial locations are arranged in order according to the three-dimensional coordinate number to form a morphological adaptive distribution.

[0030] In one feasible implementation, the degree of land mobility morphology adaptation at each spatial location in the morphological adaptation distribution is numerically compared with the limit value of the land mobility morphology at that spatial location recorded in the reachable motion parameters.

[0031] Specifically, in the morphological adaptation distribution, the land movement morphological adaptation value at a certain spatial location is directly taken from the terrain passability calibration value in the two-dimensional vector of that location. The movement capability limit value of the land movement morphology recorded in the reachable motion parameters at that spatial location is comprehensively determined by looking up the corresponding maximum forward speed and maximum obstacle clearance height from the reachable motion parameters based on the terrain slope and obstacle density at that location. If the land movement morphological adaptation value at that location is less than or equal to the land movement capability limit value, the adaptation value at that location is kept unchanged as the land adaptation value after constraint injection. If the land movement morphological adaptation value at that location is greater than the land movement capability limit value, the adaptation value at that location is adjusted to the product of the land movement capability limit value multiplied by 0.9 as the land adaptation value after constraint injection. In the morphological adaptation distribution, the aquatic movement morphological adaptation value at that spatial location is directly taken from the hydrological passability calibration value in the two-dimensional vector of that location. The limit value of the aquatic movement morphology recorded in the reachable movement parameters for a given spatial location is determined by finding the corresponding maximum submersible speed from the reachable movement parameters based on the water depth and current velocity at that location. If the aquatic movement morphology adaptation value at that location is less than or equal to the limit value of aquatic movement capability, the adaptation value at that location is kept unchanged as the aquatic adaptation value after constraint injection. If the aquatic movement morphology adaptation value at that location is greater than the limit value of aquatic movement capability, the adaptation value at that location is adjusted to the product of the limit value of aquatic movement capability multiplied by 0.85 as the aquatic adaptation value after constraint injection. All spatial locations in the morphological adaptation distribution are processed one by one according to the above comparison and adjustment method. After constraint injection, each spatial location obtains a set of two-dimensional vectors containing the adjusted land adaptation value and the adjusted aquatic adaptation value. The two-dimensional vectors of all spatial locations are arranged in order according to the three-dimensional coordinate number to form a constraint adaptation field.

[0032] In one feasible implementation, the land adaptation value and aquatic adaptation value of each spatial location in the constrained adaptation field are mapped to the same numerical range.

[0033] Specifically, the land adaptation values ​​of all spatial locations in the constrained adaptation field are traversed, and the maximum and minimum values ​​are found. For each spatial location's land adaptation value, the minimum value is subtracted from the value, and then divided by the difference between the maximum and minimum values. The resulting quotient is used as the normalized land adaptation value for that location. The same calculation method is used to normalize the aquatic adaptation values ​​of all spatial locations in the constrained adaptation field, and the maximum and minimum values ​​of all aquatic adaptation values ​​are found. For each spatial location's aquatic adaptation value, the minimum aquatic value is subtracted from the value, and then divided by the difference between the maximum and minimum aquatic values. The resulting quotient is used as the normalized aquatic adaptation value for that location. After normalization, the land and aquatic adaptation values ​​of all spatial locations are distributed within the range of zero to one. Each spatial location, after normalization, yields a set of two-dimensional vectors containing normalized land and aquatic adaptation values. The two-dimensional vectors of all spatial locations are arranged in order according to their three-dimensional coordinate numbers to form a unified scalar field.

[0034] In one feasible implementation, the normalized fitness value of each spatial location in the unified scalar field is converted into a decision potential value.

[0035] Specifically, the normalized land fitness value of a spatial location in the unified scalar field is read, and a nonlinear enhancement transformation is performed on the normalized land fitness value. When the normalized land fitness value is less than 0.5, the transformed decision potential energy value increases slowly with the increase of the fitness value at a low ratio. When the normalized land fitness value is greater than 0.5, the transformed decision potential energy value increases rapidly with the increase of the fitness value at a high ratio. Finally, the transformed decision potential energy value is distributed in the range of 0.1 to 0.95. The normalized aquatic fitness value of each spatial location in the unified scalar field is processed in the same nonlinear enhancement transformation method. After transformation, each spatial location obtains a set of two-dimensional vectors containing land decision potential energy value and aquatic decision potential energy value. The two-dimensional vectors of all spatial locations are arranged in order according to the three-dimensional coordinate number to form a morphological selection potential field.

[0036] S4. Based on the robot's task objective, inject task potential into the morphology selection potential field to obtain the robot's global decision field. The robot's task objectives are broken down layer by layer. From the task objectives, the target position coordinates of the task execution, the sequence of actions that need to be completed in sequence during the task execution, the execution space range corresponding to each action, and the termination conditions for task completion are extracted. The extracted items are arranged according to the order of action execution and the specific meaning of each item is recorded. All the recorded items are summarized to form the task elements.

[0037] In one feasible implementation, attribute analysis is performed on each element item in the task elements to extract the inherent characteristics of that element item in the spatial and temporal dimensions.

[0038] Specifically, the target spatial coordinates corresponding to the execution of a task element are extracted from the task element. These coordinates are compared with the three-dimensional coordinate numbers of each spatial region in the morphological selection potential field to determine the spatial region number expected to be executed by the element. The sequence number of the element in the execution time sequence is extracted from the task element. The execution order relationship between the element and the preceding and following elements is determined based on the sequence number. The task type category to which the element belongs is extracted from the task element. This category includes positioning and arrival actions and state maintenance actions. If the element belongs to the positioning and arrival action, its spatial attribute is marked as a point target. If the element belongs to the state maintenance action, its spatial attribute is marked as a region target. The extracted target spatial region number, execution sequence number, execution order relationship, and spatial attribute marking are recorded according to the name of the element. The attribute records of all elements are summarized to form the task topology feature.

[0039] In one feasible implementation, spatial relationship analysis is performed on the various spatial regions that have been divided in the morphological selection potential field to determine the positional adjacency and inclusion relationships between the regions.

[0040] Specifically, taking a spatial region in the morphological selection potential field as the center, the boundary coordinate range of that region in three-dimensional space is read. The boundary coordinate range of that region is extended by 0.5 meters along the positive and negative directions of the three-dimensional coordinate axes to form the extended boundary range of that region. All other spatial regions in the morphological selection potential field are traversed. If the boundary coordinate range of another spatial region overlaps with the extended boundary range, it is determined that the other spatial region is adjacent to the central region, and the number of the other spatial region is recorded in the adjacent region list of the central region. If the boundary coordinate range of another spatial region is completely contained within the boundary coordinate range of the central region, it is determined that the other spatial region is a sub-region of the central region, and the number of the other spatial region is recorded in the sub-region list of the central region. The adjacency and inclusion relationships of each spatial region in the morphological selection potential field are determined one by one in the above manner. The adjacent region list and sub-region list of all spatial regions are summarized according to the region number to form a spatial topology.

[0041] The target spatial region number corresponding to each task element item in the task topology feature is bound to the spatial region with the same region number in the spatial topology structure. The adjacent region list and sub-region list corresponding to the region number are read from the spatial topology structure to determine the spatial connection path between the task element item and other task element items. If the target spatial region numbers corresponding to two task element items are adjacent in the spatial topology structure, it is determined that there is a direct spatial reachability between the two task element items and the reachability is marked as direct connection. If the target spatial region numbers corresponding to two task element items are not directly adjacent in the spatial topology structure, the intermediate region sequence between the two regions is read from the spatial topology structure to form a spatial transition path and the transition path is marked as indirect connection. The spatial connection relationship between all the above task element items is recorded according to the task element name. All connection relationship records are summarized to form a spatial association relationship.

[0042] Based on the importance of each spatial region in the task execution recorded in the spatial relationship, a task weight value is assigned to each spatial region in the morphological selection potential field.

[0043] Specifically, the process iterates through all connection records in the spatial relationships, counts the number of times each spatial region appears as a direct connection endpoint and the number of times it appears as an indirect connection intermediate node, and adds the product of the number of direct connection endpoint occurrences multiplied by two and the product of the number of indirect connection intermediate node occurrences multiplied by one. The sum is used as the initial importance score of the spatial region. After calculating the initial importance score for all spatial regions in the morphological selection potential field, the maximum and minimum values ​​among all initial importance scores are found. For each spatial region, the minimum value is subtracted from its initial importance score, and then divided by the difference between the maximum and minimum values. The quotient is used as the task weight value of the spatial region. After calculating the task weight value for each spatial region in the above manner, each spatial region obtains a task weight value distributed between zero and one. The task weight values ​​of all spatial regions are arranged in order according to the three-dimensional coordinate number to form a task weight field.

[0044] Using the task weight value corresponding to each spatial region in the task weight field, the land decision potential energy value and aquatic decision potential energy value at the same three-dimensional coordinate number position in the morphological selection potential field are modulated respectively. The land decision potential energy value at a certain spatial position is multiplied by the task weight value at that position and then added to one-fifth of the task weight value at that position. The resulting value is used as the modulated land decision potential energy value at that position. Similarly, the aquatic decision potential energy value at a certain spatial position is multiplied by the task weight value at that position and then added to one-fifth of the task weight value at that position. The resulting value is used as the modulated aquatic decision potential energy value at that position. The above modulation method is used to calculate all spatial positions in the morphological selection potential field one by one. After modulation, each spatial position obtains a set of two-dimensional vectors containing the modulated land decision potential energy value and the modulated aquatic decision potential energy value. The two-dimensional vectors of all spatial positions are arranged in order according to the three-dimensional coordinate number to form the task modulation potential field.

[0045] Iterate through the modulated land decision potential energy values ​​at all spatial locations in the task modulation potential field, finding the maximum and minimum values. For each spatial location, subtract the minimum value from the modulated land decision potential energy value, then divide by the difference between the maximum and minimum values. The quotient is used as the regularized land decision potential energy value for that location. Apply the same calculation method to the modulated aquatic decision potential energy values ​​at all spatial locations in the task modulation potential field, finding the maximum and minimum values ​​among all modulated aquatic decision potential energy values. For each spatial location, the aquatic decision potential energy value is then... The decision potential value is obtained by subtracting the minimum aquatic value from the decision potential value and dividing it by the difference between the maximum and minimum aquatic values. The quotient is used as the regularized aquatic decision potential value for that location. After regularization, the land and aquatic decision potential values ​​for all spatial locations are distributed in the range of zero to one. Each spatial location is given a set of two-dimensional vectors containing the regularized land and aquatic decision potential values ​​after regularization. The two-dimensional vectors of all spatial locations are arranged in order according to the three-dimensional coordinate number to form the global decision field.

[0046] S5. Perform potential field optimization analysis on the global decision field to obtain the dominant potential field form of the robot; The system iterates through the regularized land and aquatic decision potential energy values ​​of all spatial locations in the global decision field. For each spatial location, it selects all adjacent locations within two grid units in three-dimensional space, centered on that location. It reads the difference in decision potential energy values ​​between the location and each adjacent location, and uses the ratio of this difference to the spatial distance between the two locations as the decision potential energy gradient value in that direction. When the decision potential energy gradient value in a certain direction of the location is greater than 0.1, the direction is marked as the direction of rising decision potential energy. The number of directions marked as the direction of rising decision potential energy in all directions around the location is counted. If the number of marked directions is greater than four, the location is determined to be the center of the steep rise region of decision potential energy. The coordinate numbers of all locations determined to be the center of the steep rise region and the motion morphology type corresponding to the larger of the land and aquatic decision potential energy values ​​at that location are recorded. The recorded information of all the centers of the steep rise region is summarized to form the dominant decision primitive.

[0047] In one feasible implementation, a local search area for each dominant decision element is defined in the global decision field, centered on the coordinate number of each dominant decision element and including the central location and all adjacent locations within five grid units in three-dimensional space.

[0048] Specifically, within the local search area, the land decision potential energy value and aquatic decision potential energy value are read one by one at each spatial location. For the land decision potential energy value, the value at that location is compared with the average land decision potential energy value of all locations within the local search area. If the land decision potential energy value at that location is greater than the average and the land decision potential energy value of at least six of the eight adjacent locations is less than the land decision potential energy value at that location, then that location is marked as a land potential energy maximum point. For the aquatic decision potential energy value, the value at that location is compared with the average aquatic decision potential energy value of all locations within the local search area. If the land decision potential energy value at that location is greater than the average and the land decision potential energy value of at least six of the eight adjacent locations is less than the land decision potential energy value at that location, then that location is marked as a land potential energy maximum point. If the aquatic decision potential energy value is greater than the average value and the aquatic decision potential energy values ​​of at least six of the eight adjacent locations are less than the aquatic decision potential energy value of this location, then this location is marked as a hydroelectric potential energy maximum point. All marked terrestrial potential energy maximum points and hydroelectric potential energy maximum points are classified according to their spatial coordinates. Maximum points with a spatial distance of less than 0.3 meters are grouped into the same extreme value cluster. The point with the largest decision potential energy value in each extreme value cluster is taken as the representative point of the extreme value cluster. The coordinate numbers of all representative points of extreme value clusters, their corresponding decision potential energy values, and the motion morphology type to which the representative point belongs are recorded and summarized to form the potential field dominant structure.

[0049] Based on the coordinate number and motion morphology type of each extreme value cluster representative point recorded in the dominant potential field structure, the land decision potential energy value and aquatic decision potential energy value at that coordinate number location are read from the global decision field. If the representative point belongs to the land motion morphology type, the location is marked as a land decision advantage point; if the representative point belongs to the aquatic motion morphology type, the location is marked as an aquatic decision advantage point. Taking each land decision advantage point as the center, a range is delineated in the global decision field that includes the center location and all adjacent locations within three grid units in three-dimensional space as the local clustering region of the land decision advantage point. Within this local clustering region, all points marked as land decision advantage points are statistically analyzed. If the number of locations is greater than five, the arithmetic mean of the spatial coordinates of all land decision-making advantage points within the local cluster area is calculated. The spatial location corresponding to this arithmetic mean is taken as the land decision-making convergence center of the local cluster area. Taking each aquatic decision-making advantage point as the center, the local cluster area is delineated in the same way, and the number of locations marked as aquatic decision-making advantage points within the area is counted. If the number is greater than five, the arithmetic mean of the spatial coordinates of all aquatic decision-making advantage points within the local cluster area is calculated as the aquatic decision-making convergence center of the local cluster area. All the calculated land decision-making convergence centers and aquatic decision-making convergence centers are recorded and summarized according to their spatial location coordinate numbers to form the morphological decision-making hub.

[0050] The distribution density of land-based morphological decision-making hubs is represented by the ratio of the number of land-based hubs to the sum of the number of land-based and aquatic hubs. Similarly, the distribution density of aquatic morphological hubs is represented by the ratio of the number of land-based hubs to the sum of the number of land-based and aquatic hubs. The arithmetic mean of the land-based decision-making potential energy values ​​at all land-based decision-making hubs is used as the potential energy intensity representation of land-based morphology. The arithmetic mean of the aquatic decision-making potential energy values ​​at all aquatic decision-making hubs is used as the potential energy intensity representation of aquatic morphology. Finally, the distribution density representation of land-based morphology is multiplied by... The product obtained by multiplying the landform's potential energy intensity characterization value by 60% is added together to obtain the comprehensive adjudication score for the landform. The product obtained by multiplying the aquatic form's distribution density characterization value by 40% is added together to obtain the product obtained by multiplying the aquatic form's potential energy intensity characterization value by 60% to obtain the comprehensive adjudication score for the aquatic form. The comprehensive adjudication scores for the landform and aquatic forms are compared. If the comprehensive adjudication score for the landform is greater than that for the aquatic form, the landform movement form is determined to be the dominant potential field form. If the comprehensive adjudication score for the aquatic form is greater than that for the landform, the aquatic movement form is determined to be the dominant potential field form.

[0051] S6. Perform morphological encoding on the dominant potential field to obtain the optimal morphological control command for the robot. In one feasible implementation, when the dominant potential field pattern is determined to be a land movement pattern, control parameters are extracted for the overall configuration of the land movement pattern.

[0052] Specifically, the allowable rotation angle ranges of the hip, knee, and ankle joints of each mechanical leg are extracted from the legged walking mechanism of the land-mobile form. These ranges are determined based on the mechanical limiting structure of the legged walking mechanism. The allowable rotation angle range of the hip joint is -30 degrees to +60 degrees, the allowable rotation angle range of the knee joint is -10 degrees to +120 degrees, and the allowable rotation angle range of the ankle joint is -20 degrees to +30 degrees. The maximum output speed range of the drive wheel axle and the maximum deflection angle range of the steering servo are extracted from the wheeled rolling mechanism of the land-mobile form. The maximum output speed range of the drive wheel axle is 0 to 120 revolutions per minute, and the maximum deflection angle range of the steering servo is -45 degrees to +45 degrees. The lifting stroke range of the fuselage attitude adjustment mechanism is extracted from the fuselage support frame of the land-mobile form. The lifting stroke range is -5 centimeters to +15 centimeters. All the extracted control parameters are classified and recorded according to the name of the mechanism and the name of the control parameter, and then summarized to form the morphological decision parameters.

[0053] In one feasible implementation, when the dominant potential field pattern is determined to be an aquatic motion pattern, control parameters are extracted for the overall configuration of the aquatic motion pattern.

[0054] Specifically, the speed adjustment range of the propulsion motor and the vector deflection angle range of the propulsion unit are extracted from the underwater propulsion unit in the aquatic motion form. The speed adjustment range of the propulsion motor is 0 to 150 revolutions per minute, and the vector deflection angle range of the propulsion unit is -30 degrees to +30 degrees. The displacement adjustment range and the upper limit of the adjustment rate of the buoyancy chamber are extracted from the buoyancy adjustment chamber in the aquatic motion form. The displacement adjustment range of the buoyancy chamber is 0 ml to 500 ml, and the upper limit of the adjustment rate is 50 ml per second. The deflection angle range and the upper limit of the deflection response time of each wing surface are extracted from the attitude control wing surface in the aquatic motion form. The deflection angle range of each wing surface is -20 degrees to +20 degrees, and the upper limit of the deflection response time is 0.2 seconds. The upper limit of the hull pressure bearing depth is extracted from the fuselage sealing shell in the aquatic motion form. The upper limit of the hull pressure bearing depth is 30 meters underwater. All the above-extracted control parameters are classified and recorded according to the name of the organization to which they belong and the name of the control parameter. After being summarized, they form the morphological decision parameters.

[0055] For each control parameter in the morphological decision parameters, a corresponding control target value is set. The control target value is determined based on the ideal motion state corresponding to the decision potential energy value of the dominant potential field at the current execution position. For the maximum output speed parameter of the drive wheel axle, the control target value is set to 60 revolutions per minute. For the maximum deflection angle parameter of the steering servo, the control target value is set to the steering angle value of the current path planning. For the lifting stroke parameter of the fuselage attitude adjustment mechanism, the control target value is set to the difference between the current terrain height and the fuselage reference height. All the above control target values ​​are recorded according to their corresponding parameter names. The execution sequence of each control parameter is determined according to the spatial movement sequence of the robot when performing the task. First, the lifting action of the body attitude adjustment mechanism is executed to make the body reach the target height. Then, the speed adjustment action of the drive wheel axle or propulsion motor is executed to make the robot move at the target speed. At the same time as the movement action is executed, the steering servo or thruster vector deflection action is executed to make the robot turn according to the planned path. During the movement action, the current lifting position of the body attitude adjustment mechanism is continuously monitored and compared with the target value to implement closed-loop adjustment. The control target value, execution sequence determination and closed-loop adjustment rules are summarized to form the control feature architecture.

[0056] Read the physical specifications manuals for all robot actuators and record the physical limit values ​​for each actuator one by one. The maximum safe speed of the drive motor is 100 revolutions per minute, the maximum safe deflection angle of the steering servo is ±45 degrees, the maximum safe travel of the lifting push rod is ±15 centimeters, the maximum safe speed of the propulsion motor is 130 revolutions per minute, the maximum safe deflection angle of the vector deflection servo is ±30 degrees, the maximum safe drainage volume of the buoyancy regulating pump is 450 milliliters, and the maximum safe deflection angle of the wing deflection servo is ±20 degrees. Record the control target values ​​for each control parameter in the control feature architecture. The control target value is compared with the physical limit value of the corresponding actuator. If the control target value is less than or equal to the physical limit value, the control target value remains unchanged. If the control target value is greater than the physical limit value, the control target value is adjusted to 95% of the physical limit value. After completing the comparison and adjustment of all control target values, the name of each control parameter, the adjusted control target value, the execution sequence of the control parameter, and the closed-loop adjustment rule corresponding to the control parameter are combined into a complete control instruction entry. The control instruction entries of all control parameters are arranged in the execution sequence to form the optimal form control instruction.

[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0058] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for autonomous decision of operation mode of a body-attached amphibious robot, characterized in that, The method includes: S1. Reconstruct the spatiotemporal field of the robot's multi-source environmental perception data to obtain the robot's multi-layer environmental field; S2. Analyze the kinematic characteristics of the robot's land movement and aquatic movement to obtain the robot's achievable motion parameters; S3. Construct a scalar field from the multi-layered environmental field and the reachable motion parameters to obtain the morphological selection potential field of the robot; S4. Based on the robot's task objective, inject task potential into the morphology selection potential field to obtain the robot's global decision field. S5. Perform potential field optimization analysis on the global decision field to obtain the dominant potential field form of the robot; S6. Perform morphological encoding on the dominant potential field to obtain the optimal morphological control command for the robot.

2. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 1, characterized in that, The spatiotemporal field reconstruction of the robot's multi-source environmental perception data to obtain the robot's multi-layer environmental field includes: Based on the data type of the multi-source environmental perception data, the robot's multi-source environmental perception data is decoupled into multiple channels to obtain an independent data stream of the multi-source environmental perception data; Based on the independent data stream, a three-dimensional grid mapping is performed on the environment in which the robot is located to obtain the discretized field of the robot; The discretized field is subjected to feature parameter extraction to obtain the dimensions of the robot's environmental parameters. Based on the physical properties represented by the dimensions of the environmental parameters, hierarchical semantic injection is performed on the discretized field to obtain the semantic labeling field of the robot; The discretized field and the semantically labeled field are then fused to obtain the multi-layered environmental field of the robot.

3. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 1, characterized in that, The kinematic analysis of the robot's land and aquatic movement patterns yields the robot's achievable motion parameters, including: The morphological characteristics of the robot are obtained by decomposing its land movement and aquatic movement patterns. Based on the aforementioned morphological features, the motion performance of the robot on land and underwater is characterized to obtain the robot's kinematic properties. Based on the kinematic properties of the robot, the reachability boundary of the robot's movement limit is defined to obtain the capability boundary characterization of the robot; The achievable motion parameters of the robot are obtained by performing a generalized solution on the kinematic properties and the capability boundary representation.

4. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 3, characterized in that, The specific formula for calculating the reachable motion parameters is as follows: ； In the formula, The first of the reachable motion parameters Item component, The first in the capability boundary representation Item component, The kinematic property of the first Item component, As a safety margin adjustment factor, It is the hyperbolic tangent function.

5. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 3, characterized in that, The step of constructing a scalar field from the multi-layered environmental field and the reachable motion parameters to obtain the robot's shape selection potential field includes: Based on the multi-layered environmental field of the robot, an environmental adaptability analysis is performed on the robot to obtain the morphological adaptability distribution of the robot; Based on the reachable motion parameters of the robot, motion capability constraints are injected into the morphological adaptive distribution to obtain the constraint adaptive field of the robot. The constraint adaptation field is normalized to obtain the unified scalar field of the robot; By applying decision potential energy to the unified scalar field, the morphological selection potential field of the robot is obtained.

6. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 5, characterized in that, The environmental adaptability assessment of the robot based on its multi-layered environmental field, to obtain the robot's morphological adaptability distribution, includes: Feature parameters are extracted from the multi-layered environmental field to obtain the environmental feature parameters of the robot; Based on the environmental characteristic parameters, the land movement pattern and the aquatic movement pattern are analyzed separately to obtain the terrain passability calibration and hydrological passability calibration of the robot; The terrain passability calibration and hydrological passability calibration are comprehensively characterized and mapped to obtain the morphological adaptability distribution of the robot.

7. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 1, characterized in that, Based on the robot's task objective, a task potential is injected into the morphology selection potential field to obtain the robot's global decision field, including: The semantic elements of the robot's task objectives are analyzed to obtain the robot's task elements; Based on the spatial region corresponding to the potential field selected by the morphology, the spatial topology of the task elements is deconstructed to obtain the spatial relationship of the robot. Based on the spatial correlation, the spatial region of the morphology selection potential field is weighted to obtain the task weight field of the robot. Based on the task weight field, the local potential energy distribution of the morphology selection potential field is modulated by the field strength gradient to obtain the task modulation potential field of the robot. The task modulation potential field is regularized and integrated to obtain the robot's global decision field.

8. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 7, characterized in that, Based on the spatial region corresponding to the morphological selection potential field, the spatial topology of the task elements is deconstructed to obtain the spatial relationships of the robot, including: Task attributes are extracted from the task elements to obtain the task topology features of the robot; The spatial topology of the robot is obtained by performing spatial topology extraction on the spatial region. The task topology features are associated with the spatial topology to construct a topology, and the spatial association relationship of the robot is finally obtained.

9. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 3, characterized in that, A potential field optimization analysis is performed on the global decision field to obtain the dominant potential field pattern of the robot, including: Gradient feature analysis is performed on the global decision field to obtain the dominant decision primitives of the robot; Based on the dominant decision-making unit, key configuration identification is performed on the potential energy extreme value region in the global decision field to obtain the potential field dominant structure of the robot. Based on the dominant potential field structure, the decision point convergence of the decision advantage regions of the land movement mode and the aquatic movement mode is performed to obtain the morphological decision hub of the robot. By comprehensively assessing the spatial distribution and potential energy intensity of the morphological decision hub, the dominant potential field morphology of the robot is obtained.

10. The autonomous decision-making method for the operational mode of an amphibious robot as described in claim 1, characterized in that, The step of encoding the dominant potential field shape to obtain the optimal shape control command for the robot includes: The morphological parameters of the robot are obtained by extracting morphological parameters from the dominant potential field morphology. The control logic reduction of the morphological decision parameters yields the control feature architecture of the robot. The physical constraints of the robot's actuators are injected into the control feature architecture to obtain the optimal shape control instructions for the robot.

Images