A Behavior Decision-Making Method for Underwater Robots Based on Behavior Trees

Abstract

This invention relates to the field of autonomous decision-making technology for robots, and discloses a behavior decision-making method for underwater robots based on behavior trees. This method describes and controls the specific behaviors of underwater robots using behavior trees, organizing and managing complex behavioral logic in a tree structure. By analyzing specific tasks, it decomposes them into behavior subtrees, and uses control nodes and action nodes in these subtrees to complete specific logic switching functions. Compared to traditional finite state machines, this method has the advantages of modularity, scalability, and high readability. The modules within the subtrees have low coupling, and the method can be ported to multiple locations within the behavior tree, effectively solving the problem of high workload associated with reusing states in finite state machines, and has significant practical application value.

Description

Technical Field

[0001] This invention relates to the field of autonomous decision-making technology for robots, and in particular to a behavior decision-making method for underwater robots based on behavior trees. Background Technology

[0002] In recent years, unmanned intelligent technology has developed rapidly, and underwater robots, as an important component, play a significant role in underwater engineering and marine scientific research. Currently, underwater robots typically employ the Finite State Machine (FSM) method for task management and switching, used for simple management of behaviors such as path planning, obstacle avoidance, and multi-agent navigation. This method is practical and easy to implement, but for more complex systems, the excessive number of state transitions makes the finite state machine difficult to maintain, and the increase in the number of states leads to an exponential increase in complexity.

[0003] Meanwhile, since the underwater robot operates in a communication-denied environment, the task sequence must meet constraints of rationality and economy. For example, the total path length after combining different sub-modules must meet the endurance performance requirements, and the switching between different modules must be reasonable and executable. Manually set task sequences need to be validated and optimized into executable sequences.

[0004] Behavior Tree (BT) is a modular decision-making method for describing and controlling robot behavior. It organizes and manages complex behavioral logic using a tree structure, achieving robot decision-making and behavior control through different node combinations. It can express functions equivalent to finite state machines and exhibits advantages in modularity, scalability, and readability. Each subtree in a behavior tree is a separate module, with low coupling between different modules, facilitating the modification of behavior sequences and effectively solving the problem of high workload associated with reusing states in finite state machines.

[0005] For underwater robots, the design of an autonomous decision-making system is key to improving their autonomy in the context of the slow development of underwater communication and sensing technologies. By designing a decision-making system in a reasonable way, the constraints of other technologies on the autonomy of underwater robots can be reduced to a certain extent. Summary of the Invention

[0006] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a behavior decision-making method for underwater robots based on behavior trees.

[0007] A behavior decision-making method for underwater robots based on behavior trees.

[0008] The underwater robot's task is decomposed into multiple independent behavior nodes, and the input parameters, output results, and behavior switching rule tables for each behavior node are defined.

[0009] Connect the behavior nodes according to the task requirements to construct at least one task event subtree and at least one exception handling subtree.

[0010] The task event subtree is used to process preset task sequences and generate navigation commands required by the controller based on input task parameters and environmental information.

[0011] The exception handling subtree is used to handle emergencies and to generate emergency control commands based on the input robot operating status information and environmental information.

[0012] The task event subtree and the exception handling subtree are integrated into the main behavior tree. The behavior switching rule table is used to verify the rationality of the behavior nodes in the main behavior tree, and unreasonable nodes are optimized. After the judgment of all nodes is completed, the main behavior tree and the description file of the main behavior tree are generated.

[0013] Load the description file of the main behavior tree, execute the main behavior tree and monitor its running status in real time, and output the behavior decision results of the underwater robot.

[0014] Furthermore, the behavior switching rule table includes parameter restrictions and connection restrictions for behavior nodes.

[0015] Furthermore, the execution priority of the exception handling subtree is higher than that of the task event subtree.

[0016] Furthermore, the behavioral nodes include leaf nodes and composite nodes;

[0017] The combined node is used to control the traversal method of its child nodes, and is selected from at least one of sequence nodes, selection nodes, or parallel nodes.

[0018] The leaf nodes are used to perform specific functions and are selected from at least one of condition nodes or action nodes.

[0019] Furthermore, the task event subtree includes condition nodes and action nodes. The condition nodes of the task event subtree are used to parse the input task parameters and select the corresponding action nodes to execute. The action nodes output cruise, stationary, or depth adjustment commands based on the input navigation parameter information and external environment information.

[0020] The anomaly handling subtree includes pipeline sequence nodes, hazard monitoring nodes, collision avoidance nodes, return-to-home nodes, and emergency ascent nodes. The pipeline sequence nodes continuously traverse the preceding nodes during the subtree's operation. The hazard monitoring nodes determine whether to trigger emergency behavior based on the input robot operating status information and environmental information. The output of the anomaly handling subtree includes collision avoidance commands, return-to-home commands, or emergency ascent commands.

[0021] The output of the exception handling subtree takes precedence over the output of the task event subtree.

[0022] Furthermore, the steps to optimize unreasonable nodes include:

[0023] Read the main behavior tree and the behavior switching rule table, and make a reasonable judgment on adjacent behavior nodes in the main behavior tree;

[0024] If the behavior node is deemed unreasonable, it is removed from its position in the main behavior tree. Then, according to the behavior switching rule table, a new reasonable position is searched in the main behavior tree for insertion. If no reasonable position exists, the unreasonable node is deleted.

[0025] Furthermore, when the main behavior tree is executed, it first traverses the exception handling subtree. If there is no emergency, it executes the behavior node in the task event subtree and returns to the initial position after completing the task.

[0026] Furthermore, the behavior switching rule table includes one of the following restrictions: a single depth adjustment does not exceed a preset range, state-maintaining behavior cannot be connected to collision avoidance behavior, and collision avoidance behavior cannot be connected to state-maintaining behavior.

[0027] Furthermore, the nodes of the task event subtree and the exception handling subtree include at least one of sequence nodes, selection nodes, parallel nodes, condition nodes, action nodes, waiting nodes, and loop nodes, and the nodes return a status of successful execution, failed execution, or execution in progress based on the input parameters.

[0028] Furthermore, the description file is in XML text format.

[0029] The description file is parsed by a visualization tool to display the structure of the main behavior tree and show the running status of each node in real time. The running status includes execution success, execution failure, or execution in progress.

[0030] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0031] This invention addresses the problem of task management and behavior decision-making during underwater robot navigation. The proposed behavior tree-based underwater robot behavior decision-making method ensures the efficiency of switching between various behaviors; visualization and XML description improve development and debugging efficiency and facilitate cross-platform deployment.

[0032] While retaining the functionality of traditional finite state machines, this method has advantages in modularity, scalability, and readability. Furthermore, the low coupling between different modules allows for reuse, effectively solving the problem of the large workload involved in adding or deleting states in finite state machines. The behavior tree structured modeling solves the problems of state expansion and difficulty in maintenance in FSMs. The modularization of behavior nodes reduces module coupling and improves behavior reusability.

[0033] The feasibility assessment of the established behavior tree, modification of unreasonable node positions or deletion of unreasonable nodes, has good practical application value; automatic verification of behavior rules improves the rationality and security of the system's behavioral logic; and priority control of abnormal events ensures the robot's autonomy and emergency response capabilities.

[0034] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0036] Figure 1 This is a flowchart of the algorithm in this invention;

[0037] Figure 2 This is a schematic diagram of the combined node and leaf node in this invention;

[0038] Figure 3 This is a schematic diagram of the underwater robot anomaly handling subtree in this invention;

[0039] Figure 4 This is a schematic diagram of the underwater robot mission event subtree in this invention;

[0040] Figure 5 This is a schematic diagram of the main behavior tree of the underwater robot in this invention;

[0041] Figure 6 This is the flow chart of the node reassignment algorithm in this invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but should not be used to limit the scope of this invention.

[0043] The following is combined with Figures 1 to 6 The technical solution of the present invention is described.

[0044] Figure 1 The main flow of the decision-making method of the present invention is shown.

[0045] This invention discloses a behavior decision-making method for underwater robots based on behavior trees. The method decomposes a complete navigation task into sub-behavior trees and independent behavior nodes. According to task requirements, the behavior tree is divided into a task event subtree and an exception handling subtree. The task event subtree processes the current task sequence, while the exception handling subtree handles unexpected events during navigation and makes corresponding decision outputs. The overall behavior tree framework is described using XML text format, and the running status of the behavior tree is monitored in real time using software such as Groot. The specific execution steps include:

[0046] S1. Decompose the general tasks of underwater robots into independent behavior nodes or independent subtrees such as trajectory tracking, collision avoidance, target tracking, and area search. Set the feasible domain of each node and establish a behavior switching judgment rule table based on the range of the feasible domain.

[0047] By breaking down complex tasks into modular action nodes, it is easier to build and manage behavior trees. By defining feasible domains and switching rules, the logical rationality between behavior nodes is ensured, reducing invalid or unexecutable behavior combinations. The modular node design enhances the scalability and reusability of the system, making it easier to add or modify behaviors in the future.

[0048] S2. Based on task requirements, establish a task event subtree for executing specific task sequences. Establish a trajectory tracking subtree, whose input data includes navigation parameter information and external environment information, and whose output is the route information or collision avoidance information required by the trajectory tracking controller.

[0049] The task event subtree modularizes the specific task logic, clearly expressing the task execution flow; through the combination of condition nodes and action nodes, it enables flexible task selection and execution to adapt to different task requirements; the modular design of the subtree reduces the coupling between modules, making it easier to modify and expand the task sequence.

[0050] S3. Based on mission requirements, establish an anomaly handling subtree for real-time monitoring of the unmanned aerial vehicle's (UAV) navigation status and handling of emergencies. This includes nodes such as emergency ascent and emergency return. The input data for the anomaly subtree is the UAV's operational status information, and the output is the controller's desired commands.

[0051] The exception handling subtree monitors the robot's status in real time, ensuring a rapid response to emergencies and improving system robustness. Through priority design, it ensures that emergency events are handled first, guaranteeing robot safety. The modular exception handling logic facilitates the expansion of new emergency behaviors.

[0052] S4. Integrate the subtrees established in S2 and S3 into the main behavior tree, and use XML text format to describe the overall behavior tree framework. Read the behavior tree sequence, extract adjacent behavior nodes, judge the rationality according to the behavior switching rule table. If there are unreasonable nodes, traverse the behavior tree. If a reasonable position is found, insert it. If it does not exist, delete the unreasonable node and generate the final XML text.

[0053] The main behavior tree integrates task execution and exception handling logic to form a unified decision-making framework. Through rationality verification and optimization, it eliminates unexecutable behavior sequences and ensures that behavior switching conforms to robot hardware and logical constraints. The standardized description in XML format facilitates the storage, modification, and cross-platform portability of the behavior tree.

[0054] S5. Use visualization tools such as Groot to view the behavior tree described in the XML file, and run the behavior tree to view the current running status in real time.

[0055] Visualization tools intuitively present the structure and operational status of behavior trees, facilitating debugging and optimization by developers. Real-time monitoring ensures the correctness of behavior trees in actual operation, promptly identifies potential problems, provides operational status feedback, and offers data support for subsequent behavior tree optimization.

[0056] The detailed process of S1 is as follows:

[0057] Based on typical underwater robot tasks, they are broken down into specific functional modules, which serve as action nodes in the behavior tree. Action nodes can respond to and control actions based on input parameters. Some basic behavior names and definitions are listed in Table 1.

[0058] The purpose of this step is to solve the state explosion problem in traditional FSM and provide a standardized unit for behavior tree scheduling.

[0059] Table 1. Basic Behavior Names and Definitions

[0060]

[0061] There are many types of underwater robots, and depending on the characteristics of the robots, there are various restrictions on the connection of their behavioral modules. Taking the two restrictions as examples, "maximum single depth adjustment of 5m, maximum diving depth of 10m, and state-maintaining behavior cannot be connected to collision avoidance behavior", the former is a parameter restriction of the underwater robot, and the latter is a behavior tree logic restriction. The behavior switching rule table is established as follows.

[0062] Table 2 Behavior Switching Rules

[0063]

[0064] The processes and methods for constructing subtrees in steps S2 and S3 are similar, and will be introduced together here.

[0065] Table 2 shows the definitions and functions of several basic nodes in a behavior tree. Structurally, they can be divided into leaf nodes and composite nodes.

[0066] Leaf nodes include conditions, actions, waits, loops, etc.

[0067] Condition nodes are used to detect whether a specified condition is met.

[0068] Action nodes are used to perform actual actions.

[0069] Composite nodes are used to control the traversal of the tree. A node can be connected to other child nodes, meaning that the out-degree of the node is greater than 0.

[0070] The execution characteristics of behavior trees are:

[0071] 1) The return value of the execution of any subtree root node depends only on the execution result of its child nodes and is independent of the rest of the tree;

[0072] 2) The execution order of the behavior tree is fixed. The entire tree is traversed each time. Even if some subtrees need to be skipped due to logic, the execution order between nodes will not be changed.

[0073] A behavior tree can contain multiple child nodes or behavior subtrees. Executing a subtree or child node will result in different return statuses: Success means execution was successful, Failure means execution failed, and Running means execution is in progress.

[0074] Table 3. Definition of Behavior Tree Nodes

[0075]

[0076] Composite nodes and leaf nodes can form subtrees, such as Figure 2 As shown, the subtree consists of sequence nodes, condition nodes, and action nodes.

[0077] According to the behavior tree traversal method, the traversal proceeds from left to right. The sequence node traverses its child nodes from left to right. First, the condition node is run. If the condition is met, it returns success and the action node is executed. If the action node returns success, the entire subtree returns success. If the condition is not met, subsequent nodes are not run and the subtree directly returns failure.

[0078] Based on the behavior tree execution logic, an underwater robot task event subtree and an exception handling subtree can be established.

[0079] First, establish an exception handling subtree, such as Figure 3 As shown, this subtree makes decisions based on input external environmental information and its own navigation status information. It outputs and executes different action nodes according to different decision results, thus implementing an exception handling function. The detailed functions of each node are as follows:

[0080] The PipelineSequence node, based on the sequence node, continues to traverse the previous nodes of the running subtree during the subtree operation until all nodes return success;

[0081] The ConditionCheck node, through external environment and its own status information, determines whether to take emergency action and returns an emergency status flag.

[0082] CollisionAction node: If the emergency state returns to collision avoidance, this node will be executed, outputting collision avoidance commands and controlling the aircraft to perform collision avoidance behavior;

[0083] The Reversal Action is executed if the emergency system detects insufficient remaining battery power, and the system returns to the original route.

[0084] The Surfacing Action is executed if an emergency status is detected as dangerous, causing the vehicle to surface.

[0085] The purpose of constructing an anomaly event subtree for detecting and responding to emergencies is to improve system robustness and address underwater communication denial and emergencies.

[0086] Similarly, a task event subtree can be built, such as Figure 4 As shown, this subtree selects different task behaviors based on the different input tasks. Task nodes can be added or deleted according to the task type. The detailed functions of each node are as follows:

[0087] The task node (MissionCheck) is used to parse the task parameter information input by the user and select different behaviors according to different tasks;

[0088] PatrolAction is used to perform area patrol missions;

[0089] PositionAction: The vehicle remains stationary at its current position until a preset duration is reached.

[0090] The Depth Adjustment Node (DepthadjustAction) adjusts the vehicle's sailing depth based on the set mission area.

[0091] Constructing a task event subtree to handle preset task flows aims to execute the main task and maintain the robot's autonomous navigation capability.

[0092] The detailed process of step four is as follows:

[0093] Based on the behavior subtrees established in S2 and S3, and combined with the task rules, the various subtrees and control nodes are connected to construct the main behavior tree, such as... Figure 5 As shown, PathTrackingAction is a path tracking node that outputs execution commands based on the input path tracking information to track the specified route.

[0094] In the main behavior tree, the exception handling subtree has the highest priority. During each traversal, the exception event subtree is traversed first. If there is no emergency state, the path tracking node is executed. After reaching the specified coordinates, the corresponding task is executed. After the task is completed, the path tracking node is executed again to return to the initial position.

[0095] Reread the behavior tree sequence, traverse each node, extract its adjacent behavior nodes, and determine its rationality based on the behavior switching rule table.

[0096] If its adjacent node is not in the rule table, it is an unreasonable node. Traverse the behavior tree again. If a node that matches the unreasonable node in the behavior switching rule table is found, insert the unreasonable node at that position.

[0097] If no matching node is found in the behavior switching rule table after traversal, the unreasonable node is deleted.

[0098] After each node is evaluated, the final XML text is generated. The algorithm flow is as follows: Figure 6 As shown.

[0099] This step ensures the rationality and execution stability of the behavior switching logic and automatically handles defects in the behavior tree structure.

[0100] The detailed process of step five is as follows:

[0101] Tools such as Groot can be used to visualize XML text and read and call it in navigation control software, enabling underwater robots to make autonomous decisions.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A behavior decision method for underwater robots based on behavior tree, characterized in that, a task of an underwater robot is decomposed into a plurality of independent behavior nodes, and input parameters, output results and behavior switching rule tables of the behavior nodes are defined, the behavior nodes are connected according to a task requirement, at least one task event sub-tree and at least one exception handling sub-tree are constructed, the task event sub-tree is used for processing a preset task sequence and generating a required navigation instruction of a controller according to input task parameters and environmental information, the exception handling sub-tree is used for processing an emergency event and generating an emergency control instruction according to input robot running state information and environmental information, the task event sub-tree and the exception handling sub-tree are integrated into a main behavior tree, the behavior switching rule tables are used for checking rationality of the behavior nodes in the main behavior tree, and an unreasonable node is optimized, after judging all the nodes, a final main behavior tree and a description file of the final main behavior tree are generated, the description file of the main behavior tree is loaded, the main behavior tree is executed and a running state thereof is monitored in real time, and a behavior decision result of the underwater robot is output. 2.The behavior decision method for underwater robots based on behavior tree according to claim 1, characterized in that, the behavior switching rule tables include parameter limits and connection limits of the behavior nodes. 3.The behavior decision method for underwater robots based on behavior tree according to claim 1, characterized in that, an execution priority of the exception handling sub-tree is higher than that of the task event sub-tree. 4.The behavior decision method for underwater robots based on behavior tree according to claim 1, characterized in that, the behavior nodes include leaf nodes and combination nodes, the combination nodes are used for controlling an iteration mode of the child nodes and are selected from at least one of a sequence node, a selection node or a parallel node, the leaf nodes are used for executing specific functions and are selected from at least one of a condition node or an action node. 5.The behavior decision method for underwater robots based on behavior tree according to claim 1, characterized in that, the task event sub-tree includes condition nodes and action nodes, the condition nodes of the task event sub-tree are used for analyzing input task parameters and selecting corresponding action nodes to execute, and the action nodes output a cruising instruction, a static instruction or a depth adjustment instruction according to input navigation parameter information and external environmental information, the exception handling sub-tree includes a pipeline sequence node, a dangerous state monitoring node, an avoidance node, a homing node and an emergency floating node, the pipeline sequence node continuously iterates a previous node in a sub-tree running process, the dangerous state monitoring node judges whether an emergency behavior is triggered according to input robot running state information and environmental information, and an output of the exception handling sub-tree includes an avoidance instruction, a homing instruction or an emergency floating instruction, the output of the exception handling sub-tree is executed in priority to the output of the task event sub-tree. 6.The behavior decision method for underwater robots based on behavior tree according to claim 1, characterized in that, the step of optimizing the unreasonable node includes, reading the main behavior tree and the behavior switching rule tables, and performing rationality judgment on adjacent behavior nodes in the main behavior tree. If the behavior node is determined to be unreasonable, the unreasonable behavior node is removed from its position in the main behavior tree and is inserted into a new reasonable position in the main behavior tree according to the behavior switching rule table, and if there is no reasonable position, the unreasonable node is deleted.

7. The behavior tree-based behavior decision method for an underwater robot according to claim 1, wherein the behavior switching rule table comprises at least one of the following conditions: a single depth adjustment is not more than a preset range, a state maintaining behavior cannot be connected to an obstacle avoidance behavior, and the obstacle avoidance behavior cannot be connected to the state maintaining behavior.

8. The behavior tree-based behavior decision method for an underwater robot according to claim 1, wherein the main behavior tree, when executed, first traverses the exception handling sub-tree, and if there is no emergency state, executes the behavior nodes in the task event sub-tree, and returns to the initial position after completing the task.

8. The behavior tree-based behavior decision method for an underwater robot according to claim 1, wherein the behavior switching rule table comprises at least one of the following conditions: a single depth adjustment is not more than a preset range, a state maintaining behavior cannot be connected to an obstacle avoidance behavior, and the obstacle avoidance behavior cannot be connected to the state maintaining behavior.

9. The behavior tree-based behavior decision method for an underwater robot according to claim 1, wherein the nodes of the task event sub-tree and the exception handling sub-tree comprise at least one of a sequence node, a selection node, a parallel node, a condition node, an action node, a wait node, and a loop node, and the nodes return a state of execution success, execution failure, or execution in progress according to an input parameter.

10. The behavior tree-based behavior decision method for an underwater robot according to claim 1, wherein the description file is in an XML text format.

10. The behavior tree-based behavior decision method for an underwater robot according to claim 1, wherein the description file is parsed by a visual tool to show the structure of the main behavior tree and to display a running state of each node in real time, and the running state comprises execution success, execution failure, or execution in progress. ​ ​ ​

Images