Unmanned vehicle action modeling and control method based on behavior tree

Abstract

The invention discloses an unmanned vehicle action modeling and control method based on a behavior tree. An unmanned aerial vehicle action model is constructed according to a visual behavior tree; when the unmanned vehicle action model is used for simulation, when a target enters a sensor detection range, a target node is found to transmit brief information; the target identification node identifies the model and the load of the target to obtain intelligence information; the target threat analysis node performs threat degree analysis on the target according to an identification result of the target identification node and gives a processing suggestion; the target tracking node is used for realizing mobile tracking or non-mobile tracking according to environmental conditions, hiding capability and target detection capability, and when mobile tracking is implemented, the target tracking node controls the moving speed and the moving direction of the unmanned vehicle to realize target tracking; and the cluster cooperative control node is used for realizing cooperative control among the unmanned vehicle entities in the unmanned vehicle cluster. According to the invention, the behavior tree is applied to the unmanned vehicle simulation system.

Description

Technical Field

[0001] The present invention relates to the technical field of driverless vehicles, and particularly relates to a method for modeling and controlling the actions of a driverless vehicle based on a behavior tree. Background Art

[0002] How to generate realistic and credible behaviors is the core issue in the research of CGF (Computer Generated Forces). Currently, the main behavior modeling techniques include: behavior modeling methods based on finite state machines (FSM), CGF behavior modeling methods based on Agent and multi-Agent, etc.

[0003] Based on finite state machines (FSM) is the most widely used behavior modeling method in simulations. The disadvantage of FSM is that it restricts the autonomy of behaviors and is more suitable for the description and modeling of low-level physical behaviors, such as behaviors like movement and weapon firing, but it is difficult to model high-level intelligent behaviors or complex behaviors containing parallel behaviors.

[0004] There are applications of CGF behavior modeling methods based on Agent and multi-Agent, such as the Agent modeling environment SOAR developed by the University of Michigan, which is based on symbolic representation and rule reasoning. SOAR is applied in multiple CGF systems. The disadvantage of the CGF behavior modeling method based on Agent is that there is no unified understanding of the concept and definition of Agent, and no clear theoretical system has been formed yet.

[0005] Behavior modeling has always been a difficult problem in simulation systems. The use of visual behavior modeling technology can lower the programming threshold for modelers, give full play to the ability of operators to input and edit behavior models by themselves, allow trainees to write and visually debug operation behavior rules by themselves, which can accelerate the model development process and the operation rules are more applicable.

[0006] A behavior tree is an improvement of a finite state machine. Due to its advantages of modularity, reusability, and easy extensibility, it has been widely applied in complex control systems of NPCs (Non-Player-Controlled Characters) in games, unmanned aerial vehicles, and robots. A behavior tree can support the combination of simple primitive behaviors into complex and robust operation programs.

[0007] A behavior tree has advantages such as modularity, reusability, and good readability. The modularity of the behavior tree supports model extension, making the modeling process more flexible; the reusability of the behavior tree reduces the workload of model development. Each behavior subtree can run independently and can be reused as a subtree of other behavior trees. The characteristics of modularity and reusability enable it to support planning problems with multi-node large-scale states, and at the same time, due to its provision of multiple logical structures, it can still maintain good readability.

[0008] The modeling elements of the behavior tree include three aspects: nodes, return status, and tree traversal.

[0009] (1) Nodes. The behavior tree nodes include a root node, a set of action nodes, a set of condition nodes, and a set of logic nodes. Among them, the root node sends an enabling signal (Tick) to its child nodes at a certain frequency, which is passed layer by layer to the action nodes or condition nodes through the logic nodes, and receives the return results of the nodes. The action nodes (Action) change the system state. The condition nodes (Condition) query the current state. The logic nodes control the execution logic of the entire behavior tree, responsible for passing the enabling signal (Tick) downward and reporting the execution status upward. Commonly used logic nodes include decorator nodes, sequence nodes, selector nodes, and parallel nodes. The descriptions of each node are as follows: (a) Action nodes: Nodes used to complete a certain action or task, generally changing the system state, and usually the leaf nodes of the behavior tree; (b) Condition nodes: Check whether the given conditions are met. If met, return Success; otherwise, return Failed; (c) Decorator nodes: Used to specify additional conditions for the execution of this node, such as time intervals, execution times, and frequencies, etc.; (d) Sequence nodes: For vertical behavior trees, execute the next-level nodes or behavior subtrees in the order from left to right, and for horizontal behavior trees, execute the next-level nodes or behavior subtrees in the order from top to bottom. For horizontal ones, execute the next-level nodes or behavior subtrees in the order from top to bottom. If Failed is returned midway, end this sequence node; (e) Selector nodes: For vertical behavior trees, select a child node with a true execution condition in the order from left to right, and for horizontal behavior trees, select a child node with a true execution condition in the order from top to bottom, and immediately return the result of the current child node and end the selector node when encountering the first successfully executed (Success) or currently executing (Running) child node. If all child nodes return Failed, the currently executed node is also Failed. Commonly used are improved selector nodes with priorities or probability values, that is, preferentially select child nodes with higher priorities to execute or execute child nodes or subtrees by probability sampling; (f) Parallel nodes: Execute all child nodes or subtrees in parallel.

[0010] The design of behavior tree nodes is simple and easy to improve. With the wide application of behavior trees, many improved node types have emerged, such as sequence nodes and selector nodes with time slices, inverter nodes, etc.

[0011] (2) Return result status. When a node executes, it will get a return result. The status of the return result includes Success, Failed, Running, and Error, where: (a) Success status: The condition of the conditional node is satisfied, the behavior of the action node is successfully executed, or the logic node meets the logic of successful execution, and return Success; (b) Failed status: The condition of the conditional node is not satisfied, the behavior of the action node fails to be successfully executed, or the logic node does not meet the logic of successful execution, and return Failed; (c) Running status: Generally used for action nodes, indicating that the behavior is in progress but not yet completed, and return Running; (d) Error status: An error occurs during the node execution, and return Error.

[0012] (3) Traversal of the tree. In the basic implementation of the behavior tree, the system traverses from the root of the tree every frame, and checks whether each node is activated in the order from top to bottom and from left to right until the currently activated node refreshes the behavior tree. Commonly used are improved sequential nodes with priorities or weights, that is, the more important or influential child nodes are run first.

[0013] The logical structure of the behavior tree is as Figure 1 shown.

[0014] The behavior tree (BT) is a behavior description technology developed in recent years in the field of game artificial intelligence (Game AI). Compared with FSM, BT occupies an increasingly high share in the commercial game market due to its high modularity, "two-step control", hierarchical description structure, etc. However, there is no relevant report on its application in the unmanned vehicle simulation system. Summary of the Invention

[0015] Aiming at the deficiencies of the existing technology, the present invention aims to provide a method for unmanned vehicle action modeling and control based on a behavior tree.

[0016] To achieve the above object, the present invention adopts the following technical solutions: A method for unmanned vehicle action modeling and control based on a behavior tree, comprising the following steps: S1. Construct an unmanned vehicle action model according to the visual behavior tree; the unmanned vehicle action model includes a target discovery node, a target recognition node, a target threat analysis node, a target tracking node, and a cluster cooperative control node; S2. When performing simulation using the unmanned vehicle action model established in step S1, when the target enters the detection range of the sensor, the target node is discovered and brief intelligence is transmitted; the target recognition node identifies the model and payload of the target to obtain intelligence information; the target threat analysis node analyzes the threat level of the target based on the recognition result of the target recognition node and gives handling suggestions; the target tracking node is used to achieve mobile tracking or non-mobile tracking according to environmental conditions, concealment capabilities, and target detection capabilities. When implementing mobile tracking, the target tracking node controls the moving speed and direction of the unmanned vehicle to achieve target tracking; the cluster cooperative control node is used to achieve cooperative control among the unmanned vehicle entities in the unmanned vehicle cluster.

[0017] Further, the threat level analysis result is that the target has no threat, the target has a general threat, or the target has a serious threat.

[0018] The present invention also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented.

[0019] The present invention also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented.

[0020] The beneficial effect of the present invention is that the present invention realizes the application of the behavior tree in the unmanned vehicle simulation system and gives full play to the advantages of the behavior tree in the action modeling and control of unmanned vehicles. Description of the Drawings

[0021] Figure 1 It is a schematic diagram of the logical structure of the behavior tree; Figure 2 It is a schematic diagram of the execution of the selection node; Figure 3 It is a schematic diagram of the execution of the sequence node; Figure 4 It is a schematic diagram of the execution of the parallel node; Figure 5 It is a schematic diagram of the execution of the decoration node; Figure 6 It is a view of the detection behavior rules of the unmanned vehicle action model established according to the visual behavior tree in the embodiment of the present invention. Detailed Embodiment

[0022] The following will further describe the present invention with reference to the drawings. It should be noted that this embodiment is based on the technical solution of the present application, and detailed implementation manners and specific operation processes are given, but the protection scope of the present invention is not limited to this embodiment.

[0023] This embodiment provides a method for unmanned vehicle action modeling and control based on a behavior tree, including the following steps: S1. Establish an unmanned vehicle action model according to the visual behavior tree; The visual behavior tree is a directed tree composed of nodes and directed edges. The execution of the behavior tree in a simulation cycle is a process of traversing the control flow from the root node to the leaf node. The process of traversing the control flow from the parent node to the child node is called a Tick. After the child node obtains the control flow and executes, it must report the execution result to its parent node. The execution results include three items: success (SUCCESS), failure (FAILURE), and running (RUNNING). The parent node determines the direction of the control flow by checking the execution result.

[0024] Except for the root node, each node has exactly one parent node. The behavior tree has the following four major types of nodes: Composite node (CompositeNode); Decorator node (DecoratorNode); Condition node (ConditionNode); Action node (ActionNode).

[0025] 1) The composite node is a control node, which can be used as a parent node and allows multiple child nodes. According to its composite nature, it can be further divided into three types: Selection node (SelectionNode); Sequence node (SequenceNode); Parallel node (ParallelNode).

[0026] Among them, the execution logic of the selection node is to successfully run one of the multiple child nodes. Its traversal strategy is to traverse until the first child node that can be successfully run is encountered. When it receives the successful execution result returned by the child node, it will immediately return this state to the parent node; when the execution result returned by the traversed child node is a failure, the selection node will continue to traverse the next child node; when the execution result returned by the traversed child node is a running result, the behavior tree will record the running child node and directly start traversing from this child node in the next simulation cycle. The selection node is logically equivalent to a logical OR operation. The execution schematic diagram of the selection node is as Figure 2 shown.

[0027] The serial node is a control node used to sequentially execute child nodes. Its traversal strategy is to traverse until the first child node that fails to run is encountered. When it receives a failed execution result returned by a child node, it immediately returns this status to the parent node; when the execution result returned by the traversed child node is successful, the serial node will continue to traverse the next child node; when the execution result returned by the traversed child node is a running result, the behavior tree will record the running child node and directly start traversing from this child node in the next simulation cycle. The serial node is logically equivalent to a logical AND operation. The schematic diagram of the serial node execution is as Figure 3 shown.

[0028] The traversal strategy of the parallel node is different from that of the selection node and the control node. Whenever, in a simulation cycle, the parallel node always Ticks all its child nodes. When any child node returns a running execution result, the parallel node will return a running status to the parent node after traversing all child nodes. When no child node returns a running execution result and the number of successfully returned child nodes exceeds a given value N, the parallel node returns a successful execution result to the parent node; otherwise, it returns a failed execution result. The schematic diagram of the parallel node execution is as Figure 4 shown.

[0029] 2) The decorator node is used to change the return value strategy of the control node, control the traversal times of the control node, and set execution conditions. It performs additional processing on the execution result returned by the child node after the child node execution ends and then returns it to its parent node. The decorator node can be used as a parent node, but it can only have one child node. The schematic diagram of the decorator node execution is as Figure 5 shown.

[0030] 3) The function of the condition node is very clear, that is, it returns True when the pre-defined conditions of the node are met.

[0031] 4) The behavior action node is the leaf node that actually serves as the task execution module in the behavior tree and is the realization of the behavior ability of the unmanned vehicle. Each time the behavior action node obtains the control flow, it will process the actions of the unmanned vehicle in this simulation cycle, including perceiving environmental information, calculating action operations, judging the behavior execution status and returning, etc. When the unmanned vehicle behavior needs to be carried out step by step among nodes, technologies such as the blackboard can be introduced for simple data interaction.

[0032] In this embodiment, the unmanned vehicle action model includes a target discovery node, a target recognition node, a target threat analysis node, a target tracking node, and a cluster cooperative control node; S2. When conducting simulation using the unmanned vehicle operation model established in step S1, when the target enters the sensor detection range, the target node is discovered and brief intelligence is transmitted; the target recognition node uses different sensor detection methods, such as visible light, infrared, thermal imaging and other sensor detection methods, to identify intelligence information such as the model and payload of the target; the target threat analysis node analyzes the threat level of the target based on the recognition result of the target recognition node, and the analysis result is that the target has no threat, the target has a general threat or the target has a serious threat. For the analysis result of the target having a serious threat, the target threat analysis node can give suggestions for self-processing (processing according to the payload of the platform) or only tracking (without the ability of self-processing) according to requirements; the target tracking node is used to achieve mobile tracking or non-mobile tracking (i.e., perspective tracking) according to environmental conditions, concealment capabilities, and target detection capabilities. When implementing mobile tracking, the target tracking node controls the moving speed and moving direction of the unmanned vehicle to achieve target tracking; the cluster cooperative control node is used to achieve cooperative control among the unmanned vehicle entities in the unmanned vehicle cluster.

[0033] For those skilled in the art, various corresponding changes and deformations can be given based on the above technical solutions and concepts, and all such changes and deformations should be included within the protection scope of the claims of the present invention.

Claims

1. A behavior tree-based unmanned vehicle action modeling and control method, characterized in that: The steps include: S1. Constructing a UAV action model according to a visualized behavior tree; the UAV action model includes a target discovery node, a target identification node, a target threat analysis node, a target tracking node, and a cluster collaborative control node; S2. When the unmanned vehicle action model established in step S1 is used for simulation, when the target enters the detection range of the sensor, the target node is found to transmit brief intelligence; The target recognition node identifies the target model and payload and obtains intelligence information; The target threat analysis node analyzes the threat level of the target based on the identification results of the target identification node and gives processing suggestions; The target tracking node is used to implement mobile tracking or non-mobile tracking according to environmental conditions, concealment capabilities, and target detection capabilities. When implementing mobile tracking, the target tracking node controls the moving speed and direction of the unmanned vehicle to achieve target tracking; The cluster collaborative control node is used to realize the collaborative control between each unmanned vehicle entity in the unmanned vehicle cluster.

2. The method according to claim 1, characterized in that The threat level analysis results are: the target is not a threat, the target is a moderate threat, or the target is a serious threat.

3. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 2 is implemented.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 2 is implemented.

Images