A robot task planning method and system based on unbounded symbol model detection

Abstract

The application discloses a robot task planning method and system based on unbounded symbol model detection, and belongs to the technical field of robot task planning and control. The method comprises the following steps: acquiring grid map data of a robot working environment, establishing an environment model and constructing an environment topology transition system; converting task instructions to be executed by the robot into a finite trajectory linear temporal logic task formula, and extracting a sub-formula set; encoding the environment topology transition system; constructing a logic state variable, combining to generate a unified state variable of the robot, and constructing a unified state transition relationship; based on the unified state variable and the unified state transition relationship, using an unbounded symbol model detection solver to perform reachability calculation; and when a target state is reachable, decoding and outputting a motion trajectory. The application realizes synchronous evolution of spatial state and logic constraint through unified modeling, effectively avoids state space explosion, and improves the efficiency and reliability of robot task planning under complex temporal constraints.

Description

Technical Field

[0001] This invention relates to the field of robot task planning and control technology, and in particular to a robot task planning method and system based on unbounded symbolic model detection. Background Technology

[0002] With the development of autonomous robots, intelligent agent planning, and complex task execution systems, traditional path planning methods that rely solely on reaching the target point are no longer sufficient to meet the task requirements in complex scenarios. Robots not only need to reach the target location, but also need to meet high-level task requirements with temporal sequence, behavioral dependencies, and safety constraints, such as sequentially visiting multiple areas, continuously avoiding dangerous areas, and executing subsequent actions after acquiring resources.

[0003] To address the aforementioned needs, existing technologies widely employ Finite Trajectory Linear Temporal Logic (LTLf) to formally express task constraints. LTLf semantics are defined on finite execution trajectories, highly consistent with the actual requirements for robot task termination. Current mainstream technical approaches typically adopt a framework of "LTLf formula transformation—deterministic finite automaton construction—product space search," that is, first converting the LTLf formula into a deterministic finite automaton, then combining the environmental state space with the automaton states to form a product automaton, and finally performing path search in the product state space.

[0004] However, existing technologies still have the following shortcomings: First, relying on DFA construction leads to state space explosion. The conversion from LTLf formulas to DFAs is highly complex, and the number of states grows exponentially with the formula size. After multiplying the environment state space with the DFA state space, the overall state size expands dramatically, resulting in a significant decrease in solution efficiency in scenarios such as large-scale mesh maps, sequential access to multiple regions, and deep nested logic.

[0005] Secondly, symbolic programming methods still rely on intermediate representations of automata. Existing symbolic programming techniques such as BDD, SAT, and SMT only symbolize the automata search process and fail to fundamentally eliminate the complexity bottleneck caused by automata architecture.

[0006] Third, bounded search relies on a preset planning length, making it inefficient for determining unsatisfactory tasks. Bounded model detection requires manually setting an upper bound on the path length; too small an upper bound may miss solutions, while too large an upper bound leads to an expansion of the constraint scale. For tasks with no feasible solutions, existing methods need to traverse a large state space to confirm failure and struggle to provide a formal proof of unreachability.

[0007] Fourth, the coupling efficiency between space and logic is low. Existing methods adopt a separate modeling framework of "physical environment + temporal automata". Spatial state evolution and logical state advancement are processed separately and then combined. The coupling cost is high and it is difficult to make full use of the structural characteristics between space and logic. Summary of the Invention

[0008] The purpose of this invention is to provide a robot task planning method and system based on unbounded symbolic model detection, aiming to solve the problems in existing LTLf constrained planning caused by reliance on deterministic finite automata construction, such as state space explosion, difficulty in determining the preset planning length, lack of efficient judgment of unsatisfactory tasks, and low coupling efficiency between spatial state and temporal logic. It aims to achieve unified symbolic planning without explicit construction of automata or preset path length.

[0009] To achieve the above objectives, in a first aspect, the present invention provides a robot task planning method based on unbounded symbolic model detection, comprising the following steps: S1. Obtain grid map data of the robot's working environment, establish an environment model, and merge and compress the connected components of the passable areas in the environment model to construct an environment topology transfer system. S2. Obtain the task instructions to be executed by the robot, convert the task instructions into finite trajectory linear temporal logic task formulas, perform negation normalization transformation and syntax tree recursive decomposition on the logic task formulas, extract all sub-formulas, and construct a set of sub-formulas. S3. Perform binary bit vector encoding on the abstract region in the environmental topology transfer system to obtain the location state variables; perform binary bit vector encoding on the pickable resources in the environment to obtain the resource state variables; S4. Based on the set of sub-formulas, construct logical state variables for each sub-formula. Logical state variables include current step logical variables, next step logical variables, and commitment variables. The current step logical variable indicates whether the corresponding sub-formula is true at the current moment; the next step logical variable is used to pass on unfinished logical constraints across time steps; and the commitment variable is used to mark unfinished sequential tasks. Combine the position state variables, resource state variables, and all logical state variables to construct the robot's unified state variables and set their initial values. Construct physical migration constraints based on the state transition function and dynamic constraints in the environmental topology transfer system. Construct logical evolution constraints based on the logical semantics of each sub-formula. Combine the physical migration constraints and logical evolution constraints to generate a unified state transition relationship. S5. Based on the target location in the environmental model, set all commitment variables to be in a completed state and construct a target state set; starting with the initial value of the unified state variables and using the unified state transition relationship as the jump rule, use an unbounded symbolic model detection solver to perform reachability calculation on the target state set; the unbounded symbolic model detection solver iteratively approximates the reachable state space layer by layer through inductive invariants until the state space converges to a fixed point; S6. When the target state set is reachable, decode the unified state sequence output by the solver into a two-dimensional coordinate motion trajectory that the robot can execute; when the target state set is unreachable, output proof information that the task cannot be satisfied.

[0010] Preferably, S1 specifically includes: S11. Obtain a grid map of the robot's environment, and extract the coordinate sets of obstacles, regions of interest, access control areas, pickable resources, and the unlocking mapping table between resources and access control. S12. Obtain the robot's starting position and target position in the grid map; S13. Establish an environmental model ,in For raster maps, A collection of obstacles. For a set of regions of interest, Gathering for access control areas, For the set of pickable resources, This establishes a mapping relationship between resources and access control unlocking. This is a function that maps the location of resources in a raster map. For the robot's starting position, For the target location; S14, Through abstract mapping functions Connectivity merging is performed on the passable grid regions in the environment model, and interconnected free grids are clustered into a topological node to obtain an abstract region set. ; S15. Construct an environmental topology transfer system ,in For the robot's abstract state set, It is the set of actions that a robot can perform. As the starting point, abstract state Abstract state for the target. This is a function for jumping between regions. For the set of atomic propositions, This is a region label mapping function. These are the dynamic constraints for access control resources.

[0011] Preferably, S2 specifically includes: S21. Read the task instructions to be executed by the robot and convert them into finite trajectory linear time-series logic task formulas. ; S22, Regarding the logical task formula By performing a negation paradigm shift, using De Morgan's law and the equivalence rule of temporal logic, all logical NOT operators are moved to the front of the atomic proposition, so that the transformed formula does not contain any negation symbols that act on the composite logic structure or temporal operators. S23. Perform recursive syntax tree decomposition on the transformed logical task formula, extract all sub-formulas, and construct a set of sub-formulas. .

[0012] Preferably, in S3: Position state variables Using a finite-width binary vector representation, the bit width of the position state variable is... , The number of abstract regions; the position state variable is used to uniquely identify the abstract region where the robot is currently located. Resource state variables Represented using binary vectors, each dimension of the resource state vector corresponds to a pickable resource, indicating whether the resource has been acquired by the robot.

[0013] Preferably, in S4: Current step logical variable Used to represent the sub-formula corresponding to the current time. Is it valid? Next step logical variable Used to represent the corresponding sub-formula at the next moment The logical value is used to pass the logical constraints that are not completed at the current time to the next time. Commitment variables Used to mark the unfinished sequential tasks corresponding to the final operator, the start operator, and the until operator in the formula of a linear sequential logic task with a finite trajectory. The commitment variable takes the value of true when the corresponding sequential task has not been completed and takes the value of false when the corresponding sequential task has been completed. All sub-formulas satisfy the logical consistency constraint: ,in Sub-formula The logical semantics at the current moment satisfy the condition.

[0014] Preferably, in S4: Physical state is represented as ; The unified state variable is represented as ,in This is the set of current-step logical variables, commitment variables, and next-step logical variables corresponding to all sub-formulas; The initial values ​​of the unified state variables are set as follows: position state variables Retrieve the abstract region code corresponding to the robot's starting position and the resource state variables. All dimensions are set to false, the current step logical variable corresponding to the top-level formula is set to true, and all other logical state variables are set to false.

[0015] Preferably, the physical migration constraints and logical evolution constraints in S4 specifically include: S41. Based on the inter-region jump function Dynamic constraints on access control resources Construct physical migration constraints Physical migration constraints include movement constraints, resource acquisition constraints, and restricted area access constraints. The movement constraint is: the robot is allowed to perform a movement operation when the target abstract region and the current abstract region are adjacent. Resource acquisition constraint is: when there is a pickable resource at the robot's current location, the corresponding resource state variable is determined according to... Updated to the "acquired" status and maintained thereafter. For the next moment The status of a resource For the first The location of each resource; The restricted area access constraint is: the robot is allowed to enter the restricted area only if the resource state variable corresponding to the restricted area is in the acquired state; S42. Based on the set of sub-formulas The temporal logical semantics of each sub-formula are used to construct logical evolution constraints; the logical evolution constraints include the semantic encoding of the next step operator, the final operator, the always operator, and the until operator; For the next operator, the semantic constraints are: ,in For the next step operator and sub-formula The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, For sub-formula The next logical variable, This is a flag variable indicating whether the current state is the end point of the robot's trajectory; For the final operator, the semantic constraints are: ,in, For the final operator and sub-formula The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, For sub-formula The next logical variable; For the time operator, the semantic constraints are: ,in, For the sake of the operators and sub-formulas The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, For sub-formula The next logical variable; For the until operator, the semantic constraints are: ,in, For sub-formula , sub formula The composite formula composed of operators up to the end, Representing the formula for the compound The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, For sub-formula The next logical variable; S43. Perform a logical AND operation between the physical migration constraints and all logical evolution constraints to generate a unified state transition relationship. ,in Sub-formula The logical value to be taken at the next moment. Sub-formula The logical semantics at the current moment satisfy the conditions; the unified state transition relationship forces the robot's physical displacement and logical task progress to be completed synchronously within a single state update cycle.

[0016] Preferably, in S5: Target state set The construction conditions are: and The robot's current position has reached the target abstract state. And the complete set of commitment variables All commitment variables are in a completed state; The unbounded symbolic model detection solver constructs a sequence of reachable state frames layer by layer using inductive invariants. ,in , The initial values ​​for the unified state variables, and , That is, from The state reached after performing a state transition step in any state is included in middle; when When the state space search reaches a fixed point, it is determined that the search has been completed. If the target state set If the target state set is [value], then it is determined to be reachable; if the target state set is [value], then it is determined to be reachable. If so, it is determined to be unreachable.

[0017] Secondly, the present invention provides a robot task planning system based on unbounded symbolic model detection, comprising: The environment modeling module is used to build an environment model based on the grid map data of the robot's environment, and to merge and compress the connected domains of the passable areas in the environment model to construct an environment topology transfer system. The task logic parsing module is used to convert the task instructions to be executed by the robot into finite trajectory linear temporal logic task formulas, perform negation normal form transformation and syntax tree recursive decomposition on the logic task formulas, extract all sub-formulas, and construct a set of sub-formulas. The state encoding module is used to encode the abstract region in the environmental topology transfer system into a position state variable by binary bit vector encoding, to encode the pickable resources in the environment into a resource state variable by binary bit vector encoding, to construct a logical state variable for each sub-formula according to the sub-formula set, and to combine the position state variable, resource state variable and all logical state variables to construct a unified state variable for the robot, and to set the initial value of the unified state variable. The state transition coupling module is used to construct physical transition constraints based on the state jump functions and dynamic constraints in the environmental topology transition system, construct logical evolution constraints based on the logical semantics of each sub-formula, and combine the physical transition constraints and logical evolution constraints to generate a unified state transition relationship. The unbounded model detection and solution module is used to calculate the reachability of the target state set starting from the initial value of the unified state variables and using the unified state transition relationship as the jump rule. It iteratively approximates the reachable state space layer by layer through inductive invariants until the state space converges to a fixed point and outputs the search results. The target state set is set according to the target location in the environment model, and all committed variables are completed states. The trajectory decoding and execution module is used to decode the unified state sequence output by the solver into a two-dimensional coordinate motion trajectory that the robot can execute when the target state set is reachable; when the target state set is unreachable, it outputs proof information that the task cannot be satisfied.

[0018] Therefore, the robot task planning method and system based on unbounded symbolic model detection described above, as used in this invention, have the following beneficial effects: (1) There is no need to explicitly construct deterministic finite automata, which fundamentally avoids the problem that the state size of the automaton increases exponentially with the complexity of the formula, and effectively alleviates the state space explosion. (2) Spatial state and temporal logic are modeled synchronously and simultaneously. During the same state update process, environmental state transition and LTLf logical state evolution are completed at the same time, which reduces redundant calculations and invalid searches and improves the efficiency of planning and solving. (3) No need to preset the planning path length. The trajectory length that meets the task requirements is automatically determined through inductive invariants and fixed-point reasoning, avoiding the omission of solutions or the expansion of constraint scale due to unreasonable step size settings. (4) It has the ability to formally prove unsatisfiable tasks. When there is no feasible solution to the task, it can generate a formal proof of unsatisfiability based on inductive invariants, which improves the reliability and interpretability of the planning system. (5) It supports complex tasks such as resource dependency and can effectively handle complex task scenarios such as key-door unlocking, permission access, and multi-stage condition triggering, which involve spatial movement, resource acquisition and temporal sequence coupling.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a robot task planning method based on unbounded symbolic model detection according to the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In the description of this invention, the term "robot" refers to an autonomously moving intelligent agent, including but not limited to warehousing and logistics robots, inspection robots, and service robots; "environment model" refers to a digital description of the robot's operating environment; "environment topology transfer system" refers to a state transfer system obtained after topological abstraction of the environment model; "abstract region" refers to a topological node obtained after merging connected components of traversable grids in the environment model; "position state variable" refers to a binary encoded variable used to uniquely identify the abstract region where the robot is currently located; "resource state variable" refers to a binary vector used to indicate whether each pickable resource has been acquired by the robot; "finite trajectory linear temporal logic task formula" refers to a task constraint formula expressed using the LTLf language; "sub-formula set" refers to the set of all sub-formulas obtained after recursively decomposing the LTLf task formula into a syntax tree; "logic state variable" refers to a variable used to represent the temporal logic state of the sub-formula. The variables in a state include current-step logical variables, next-step logical variables, and commitment variables; "unified state variables" refer to the comprehensive state variables obtained by uniformly encoding position state, resource state, and logical state; "physical migration constraints" refer to the constraint rules followed by the robot's physical motion and interaction with the environment; "logical evolution constraints" refer to the temporal logical semantic constraints of each sub-formula in the LTLf task formula; "unified state transition relations" refer to the state transition rules obtained by combining physical migration constraints and logical evolution constraints; "target state set" refers to the set of all unified states that satisfy the task completion conditions; "unbounded symbolic model detection solver" refers to a formal verification tool for state reachability analysis based on inductive invariant reasoning; "inductive invariant" refers to the reasoning method that approximates the reachable state space by constructing a reachable state framework layer by layer; "fixed point" refers to the termination condition when the state space search converges and no new states are generated.

[0024] Example 1 This embodiment provides a robot task planning method and system based on unbounded symbolic model detection.

[0025] like Figure 1 As shown, a robot task planning method based on unbounded symbolic model detection includes the following steps: S1. Obtain grid map data of the robot's working environment, establish an environment model, and merge and compress the connected components of the passable areas in the environment model to construct an environment topology transfer system. Specifically, it includes: S11. Obtain a grid map of the robot's environment, and extract the coordinate sets of obstacles, regions of interest, access control areas, pickable resources, and the unlocking mapping table between resources and access control. S12. Obtain the robot's starting position and target position in the grid map; S13. Establish an environmental model ,in A grid map (a two-dimensional data structure that discretizes the robot's working environment into several uniform grid cells, each grid cell recording the terrain attributes at that location) consists of several grid nodes, each node corresponding to a discrete location in the environment. This is a set of obstacles, where the grid cells within the obstacle area are impassable, used to limit the robot's range of motion. A set of regions of interest (ROIs) is used to define atomic propositions in the task logic, such as target regions, inspection regions, and danger regions. Each ROI consists of several connected grid cells. This refers to a set of access control areas, including doors, access checkpoints, or resource-restricted passageways. This is a set of pickable resources, which may include keys, permission tokens, tools, etc., used to remove restrictions on corresponding restricted areas. This represents the mapping relationship between resources and access control, indicating the dependency mapping relationship between resources and restricted areas; for example, there is a one-to-one or many-to-many mapping relationship between keys and doors. This is a function that maps the location of resources in a raster map, used to determine the position of each resource within the grid. For the robot's starting position, For the target location; The process of establishing the aforementioned environmental model involves unifying the spatial information, semantic information, and interaction relationships in the real physical environment into a discrete state space that can be processed by a computer. Here, an atomic proposition is the smallest logical unit in LTLf logic, used to describe whether the robot is in a specific region or satisfies a specific condition. Regions of interest, as the carriers of atomic propositions, associate abstract temporal logical tasks with specific spatial locations. For example, the statement "the robot is in region A" can correspond to an atomic proposition, the truth value of which is determined by whether the robot's current position belongs to region A.

[0026] S14, Through abstract mapping functions Connectivity merging is performed on the passable grid regions in the environment model, and interconnected free grids are clustered into a topological node to obtain an abstract region set. ; The purpose of topological abstraction is to transform the original high-dimensional mesh environment into a more compact state-space structure. Since directly planning the original mesh space can easily lead to an excessively large state size, this invention merges interconnected mesh cells with consistent topological properties within passable grids into the same abstract state. For example, a corridor region composed of several grids can be merged into a single abstract node, thereby significantly reducing the number of states required for subsequent planning. However, it should be noted that topological abstraction is not a mandatory step; if the original mesh size is small, subsequent planning steps can also be performed directly on the original mesh space.

[0027] S15. Construct an environmental topology transfer system ,in This is a set of abstract robot states, including general free region states, interest region states, gate states, and restricted region states. Define the action set as the set of actions that the robot can perform. It should be noted that this invention does not limit the types of actions; the set of actions can be expanded or adjusted according to the actual application scenario. For example, actions such as "loading" and "unloading" can be added in a warehousing scenario, and actions such as "detection" and "taking photos" can be added in an inspection scenario. Expanding the types of actions does not affect the core method of this invention. This is the starting abstract state, corresponding to the robot's starting position. , The target abstract state corresponds to the robot's target position. , For inter-region jump functions: This describes the state changes of a robot after it performs an action. This indicates that the robot is in a certain state. Next action Entering the state afterward , This is a set of atomic propositions, each representing a logically meaningful state attribute, such as "the robot is located in region A" or "the robot has acquired key K1". An atomic proposition is the smallest indivisible logical unit in LTLf logic, and its truth value is determined by both the current environment state and the robot's state. In this invention, each region of interest, access control area, and pickable resource can correspond to one or more atomic propositions. This is a region label mapping function used to define which atomic propositions each abstract region possesses. For example, the label set for region A might contain the atomic proposition "located in region A". These are dynamic constraints on access control resources, used to describe resource dependencies. For example, if a door... Key required Unlocking allows state transitions only if the key state meets the conditions.

[0028] Through the above topological abstraction process, the original high-dimensional grid environment can be transformed into a more compact state space structure, thereby effectively reducing the complexity of subsequent symbolic encoding.

[0029] S2. Obtain the task instructions to be executed by the robot, convert the task instructions into finite trajectory linear temporal logic task formulas, perform negation normalization transformation and syntax tree recursive decomposition on the logic task formulas, extract all sub-formulas, and construct a set of sub-formulas. Specifically, it includes: S21. Read the task instructions to be executed by the robot and convert them into finite trajectory linear time-series logic task formulas. ; Task instructions are typically given in natural or structured language, such as "The robot must first reach area A, then area B, and is prohibited from entering area C throughout the process." This step converts this task instruction into an LTLf formula. LTLf (Linear Temporal Logic on Finite Traces) is a formal logic language whose semantics are defined on finite execution trajectories. It closely aligns with the actual requirement for robot termination after task completion and is therefore widely used in finite-step task planning.

[0030] The core syntactic structures of LTLf include three categories: atomic propositions, logical connectors, and temporal operators. Atomic propositions include... This indicates "the robot has arrived at region A"; logical connectors include ∧ (AND), ∨ (OR), ¬ (NOT), → (implication), etc.; timing operators include X (next step), F (finally), G (always), U (until), R (release), etc. The following examples illustrate the meaning of each operator: This indicates "finally reaching region A", meaning a certain moment in the trajectory. True; This means "never entering region C", that is, at every moment of the trajectory. All are true; This indicates "next step to reach region B", that is, the next moment. True; This means "maintaining the existence of region A until reaching region B," that is, starting from the current moment. It remains true until Until it becomes true for the first time.

[0031] For example, the task "The robot first arrives at area A, then at area B, and is prohibited from entering area C throughout the entire process" can be represented as: ; in, This indicates that the final destination is area A, and after arriving at area A, the final destination is area B. This means that the system must not enter area C throughout the entire process. The formula above means that the system ultimately needs to reach area A, and after satisfying requirements in area A, it must eventually reach area B. Simultaneously, entry into area C is prohibited at every moment of the execution trajectory, thus fulfilling the task requirements with a sequential order and safety constraints. For tasks with resource dependencies, such as "the robot must obtain an electronic key before it can enter a special warehouse," further resource dependency constraints can be established. This constraint is described as follows: If the robot has already obtained the first... Individual access resources ( If so, entry into the corresponding restricted access area is permitted. This constraint manifests as a conditional judgment in the state transition relation, i.e., it applies only if... At that time, from the current area to the access control area Only redirects are allowed.

[0032] S22, Regarding the logical task formula By performing a negation paradigm shift, using De Morgan's law and the equivalence rule of temporal logic, all logical NOT operators are moved to the front of the atomic proposition, so that the transformed formula does not contain any negation symbols that act on the composite logic structure or temporal operators. The purpose of Negation Normal Form (NNF) transformation is to simplify subsequent logical processing. Since complex LTLf formulas often contain multiple layers of nested logic and negation operations, direct symbolic computation can easily lead to increased complexity in logical expansion and redundant state representation. Through NNF transformation, all negation operators are retained only at the atomic proposition level, avoiding complex logical inversion operations in subsequent symbolic recursion, thereby reducing the difficulty of state expansion and improving the efficiency of formula recursion.

[0033] S23. Perform recursive syntax tree decomposition on the transformed logical task formula, extract all sub-formulas, and construct a set of sub-formulas. .

[0034] Syntax tree decomposition is the process of breaking down a complex LTLf formula into smaller logical units, layer by layer. Let the input task formula be... Then the set of all its sub-formulas is defined as follows: This set is composed of the formula The system comprises all recursively decomposable logical components, including atomic propositions, logical combination expressions, and temporal sub-formulas. By fully extracting sub-formulas, complex formulas can be broken down into multiple local logical units, facilitating the subsequent establishment of symbolic state variables and recursive update mechanisms. The construction of the sub-formula set adopts a recursive extraction method: when a formula contains logical connection operators (such as AND, OR, NOT, implication) or temporal operators (such as final F, always G, until U, etc.), the system will automatically continue to recursively decompose the sub-expression until the atomic proposition level is reached.

[0035] S3. Perform binary bit vector encoding on the abstract region in the environmental topology transfer system to obtain the location state variables; perform binary bit vector encoding on the pickable resources in the environment to obtain the resource state variables; Position state variables Using a finite-width binary vector representation, the bit width of the position state variable is... , The number of abstract regions; the position state variable is used to uniquely identify the abstract region where the robot is currently located. Resource state variables Represented using binary vectors, each dimension of the resource state vector corresponds to a pickable resource, indicating whether the resource has been acquired by the robot.

[0036] To improve the solution efficiency of large-scale grid-based task planning problems, this invention employs a word-level symbolic representation for unified modeling of environmental dynamics, rather than the traditional bit-level encoding method based on Boolean variable expansion. Compared to bit-level modeling, word-level symbolic encoding can directly describe the environmental state using finite bit vectors, thereby reducing the number of variables and the scale of constraints, and improving the efficiency of subsequent symbolic solution. Let the size of the discrete state space in the environment be... Then the position state variable Represented using a finite-width binary vector, with its bit width set to... This uniquely identifies the current abstract position state of the robot. This method avoids the dimensionality expansion problem caused by traditional one-hot encoding and improves representation efficiency in large-scale state spaces. Furthermore, to describe the resource acquisition state during the planning process, this invention introduces a resource state vector. Each dimension of the vector corresponds to a resource object (such as a key, permission token, or task item). If the following conditions are met... If the result is positive, it means that the corresponding resource has been acquired and will remain valid in subsequent planning processes; otherwise, it means that the resource has not yet been acquired.

[0037] Unified state variables It was completed in the subsequent S4 build. and As a component, S3 encodes the location state and resource state, while the logical state is encoded and combined into a complete unified state variable in S4. S4. Based on the set of sub-formulas, construct logical state variables for each sub-formula. Logical state variables include current step logical variables, next step logical variables, and commitment variables. The current step logical variable indicates whether the corresponding sub-formula is true at the current moment; the next step logical variable is used to pass on unfinished logical constraints across time steps; and the commitment variable is used to mark unfinished sequential tasks. Combine the position state variables, resource state variables, and all logical state variables to construct the robot's unified state variables and set their initial values. Construct physical migration constraints based on the state transition function and dynamic constraints in the environmental topology transfer system. Construct logical evolution constraints based on the logical semantics of each sub-formula. Combine the physical migration constraints and logical evolution constraints to generate a unified state transition relationship. Specifically, it includes: First, construct three types of logical state variables.

[0038] Current step logical variable Used to represent the sub-formula corresponding to the current time. Whether it is true; when If the condition is met, it indicates that the current state satisfies the corresponding logical substructure; otherwise, it indicates that the logical condition in the current state has not yet been met. This variable allows for a unified description of the logical truth value of the current state.

[0039] Next step logical variable Used to represent the corresponding sub-formula at the next moment The logical value of this variable is used to propagate unfinished logical constraints from the current time step to the next time step; this variable is mainly used for cross-time step logical propagation, enabling the finite trajectory conditions to be continuously updated during state transitions. In other words, the logical responsibilities in the current state can be... Variables are passed to subsequent states, thereby enabling the dynamic evolution of time-series constraints.

[0040] Commitment variables Used to mark the unfinished sequential tasks corresponding to the final operator, the start operator, and the until operator in the formula of a linear sequential logic task with a finite trajectory. The commitment variable takes the value of true when the corresponding sequential task has not been completed and takes the value of false when the corresponding sequential task has been completed. Commitment variables are introduced to handle sequential logic operators with future dependencies. This invention introduces commitment variables for final operators, always operators, until operators, and release operators. This is used to indicate that the corresponding logical task has not yet been completed, but still needs to be satisfied in subsequent finite trajectories. For example, for the formula... Its semantics are "finally reach region A". When region A has not yet been reached, set... This indicates that the corresponding task is still pending and needs to be satisfied in subsequent trajectories; once the robot successfully reaches area A, it will be updated to... This indicates that the logical commitment has been fulfilled and does not need to be propagated further. Through the commitment variable mechanism, this invention can naturally express future task requirements under finite trajectory conditions without the need for traditional infinite trajectory expansion methods or additional termination state compensation mechanisms, thereby effectively reducing model complexity.

[0041] Among them, the final operator The meaning is "at some future moment in the trajectory". "Established", always operator The meaning is "at every moment of the trajectory". "All are true", until the operator The meaning is " Continue to be established until "Until it is fulfilled." The commitment variable essentially transforms an unfinished future task (such as "finally reaching area B") into a currently traceable state marker, allowing the task to be passed on like a "baton" with each step the robot takes until it is fulfilled at a certain moment. This method of transforming future tasks into current state markers avoids the complex unfolding mechanism required to predict the state for each future moment in traditional automata construction methods, which is the key to achieving linear complexity in this invention.

[0042] All sub-formulas satisfy the logical consistency constraint: ,in Sub-formula The logical semantics at the current moment satisfy the conditions. The above constraints ensure that the logical variables at any time strictly satisfy the semantics of the corresponding finite trajectory, enabling the constraints of complex LTLf tasks to achieve consistent propagation within a unified state space.

[0043] Next, we will construct the physical state representation and the unified state variables.

[0044] Physical state is represented as That is, by jointly encoding the current location status and resource status, a complete expression of the physical state of the environment can be achieved.

[0045] The unified state variable is represented as ,in, This indicates the robot's spatial position state, used to describe its current location in the environment. This indicates the resource status, used to describe the resource acquisition situation during task execution. This is a set of current step logical variables, commitment variables, and next step logical variables corresponding to all sub-formulas, used to record the current task logic satisfaction status and future time-series responsibility propagation status; The initial values ​​of the unified state variables are set as follows: position state variables Retrieve the abstract region code corresponding to the robot's starting position and the resource state variables. All dimensions are set to false, the current step logical variable corresponding to the top-level formula is set to true, and all other logical state variables are set to false.

[0046] Next, we will construct physical migration constraints and logical evolution constraints.

[0047] S41. Based on the inter-region jump function Dynamic constraints on access control resources Construct physical migration constraints Physical migration constraints include movement constraints, resource acquisition constraints, and restricted area access constraints. The movement constraint is: the robot is allowed to perform a movement operation when the target abstract region and the current abstract region are adjacent. Resource acquisition constraint is: when there is a pickable resource at the robot's current location, the corresponding resource state variable is determined according to... Updated to the "acquired" status and maintained thereafter. For the next moment The status of a resource For the first The location of each resource; The restricted area access constraint is as follows: a robot is allowed to enter a restricted area only if the resource state variable corresponding to the restricted area is in an acquired state; restricted areas include doors, access checkpoints, or resource-restricted passages. This constraint ensures that a robot cannot enter a protected area before the resource conditions are met, thus realizing an environmental interaction mechanism with conditional constraints.

[0048] S42. Based on the set of sub-formulas The temporal logical semantics of each sub-formula are used to construct logical evolution constraints; the logical evolution constraints include the semantic encoding of the next step operator, the final operator, the always operator, and the until operator; The essence of logical evolution constraints is to transform the logical meaning of each temporal operator defined in the LTLf formal language into computational rules that can be judged and passed on sequentially during state transitions. The semantic constraints of each temporal operator adopt the pattern of "judging the current step + continuing to pass on if not completed + handling the trajectory endpoint boundary", which enables complex temporal logic to propagate accurately on a finite trajectory.

[0049] For the next operator, the semantic constraints are: ,in For the next step operator and sub-formula The composite sub-formula represents "the sub-formula must be satisfied in the next time step". ; Representing the formula for the compound The logically satisfied state at the current moment, For sub-formula The next logical variable, This is a flag variable indicating whether the current state is the end point of the robot's trajectory. This indicates that the current state is not the end point of the trajectory.

[0050] The meaning of this semantic constraint is: ("The next moment must be satisfied") The following holds true at the current moment if and only if the current state is not the last step of the trajectory, and The next logical variable is true (i.e.) It is ready for execution in the next moment. If the current state is already the end of the trajectory, then there is no "next moment". Automatically expired.

[0051] For the final operator, the semantic constraints are: ,in, For the final operator and sub-formula The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, For sub-formula The next logical variable; The semantic constraints of the final operator can be understood as a recursive mechanism of "continue propagation if the task is not completed". If the current state already satisfies the condition... If the condition is met, the task is completed immediately; otherwise, the logical responsibility is passed on to the next state until the condition is met at some future step. This recursive passing mechanism allows the satisfaction of the final operator to be effectively determined within a finite trajectory, without needing to know in advance at which specific step the condition will be met.

[0052] For the time operator, the semantic constraints are: ,in, For the sake of the operators and sub-formulas The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, For sub-formula The next logical variable; The meaning of this semantic constraint is: ("Satisfying needs throughout the entire process") The following statement holds true at the current moment if and only if the following condition is met at the current moment: And (the next step is to continue to satisfy) (Or, the current step is the last step of the trajectory). If the current step has reached the end of the trajectory, then... It ended naturally.

[0053] For the until operator, the semantic constraints are: ,in, For sub-formula , sub formula The composite formula composed of operators up to the end, Representing the formula for the compound The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, For sub-formula The next logical variable; The meaning of this semantic constraint is: (“ Continue to be established until The condition "is true" is true at the current moment if and only if the condition has already been met at the current moment. , or (currently satisfied) And this is not the final step, and will The task responsibility is passed on to the next moment for further judgment. If the current condition is not met... Not satisfied Then it continues until the task fails at the current moment.

[0054] S43. Perform a logical AND operation between the physical migration constraints and all logical evolution constraints to generate a unified state transition relationship. ,in Sub-formula The logical value to be taken at the next moment. Sub-formula The logical semantics at the current moment satisfy the conditions; the unified state transition relationship forces the robot's physical displacement and logical task progress to be completed synchronously within a single state update cycle.

[0055] This unified state transition relationship forces the robot's physical displacement and logical task progress to be completed synchronously within a single state update cycle. Therefore, this invention constructs an integrated state system of spatial-logical co-evolution. Compared to the traditional two-layer product structure of "environment state space × automaton state space", this invention does not require maintaining an independent set of automaton states, fundamentally reducing the state scale and improving modeling efficiency and solution capability in complex task scenarios.

[0056] S5. Based on the target location in the environmental model, set all commitment variables to be in a completed state and construct a target state set; starting with the initial value of the unified state variables and using the unified state transition relationship as the jump rule, use an unbounded symbolic model detection solver to perform reachability calculation on the target state set; the unbounded symbolic model detection solver iteratively approximates the reachable state space layer by layer through inductive invariants until the state space converges to a fixed point; Specifically, in S5: Target state set The construction conditions are: and The robot's current position has reached the target abstract state. And the complete set of commitment variables All commitment variables are in a completed state; The unbounded symbolic model detection solver constructs a sequence of reachable state frames layer by layer using inductive invariants. ,in , The initial values ​​for the unified state variables, and , That is, from The state reached after performing a state transition step in any state is included in middle; in This indicates starting from the initial value of the unified state variable... The set of all unified states that can be reached within a step. Index for any number of steps; In order to make The step index at which it was established.

[0057] when When the state space search reaches a fixed point, it is determined that the search has been completed. If the target state set If the target state set is [value], then it is determined to be reachable; if the target state set is [value], then it is determined to be reachable. If so, it is determined to be unreachable.

[0058] S6. When the target state set is reachable, decode the unified state sequence output by the solver into a two-dimensional coordinate motion trajectory that the robot can execute; when the target state set is unreachable, output proof information that the task cannot be satisfied.

[0059] Specifically, after completing the unbounded symbolic model detection and solution, this invention further performs path decoding on the solution results to recover the executable planning trajectory that satisfies the LTLf task constraints. When the unbounded symbolic model detection process determines the target state set... When reachable, the solver outputs a finite state trajectory that satisfies the constraints.

[0060] Since the above trajectories belong to the abstract execution results in the unified symbolic state space, this invention further bases it on the environmental abstract mapping relationship. An inverse mapping operation is performed on the abstract state to recover the actual motion trajectory in the original mesh environment, that is, to convert the position encoding in the symbolic state back to the real position coordinates in the original mesh space. The final recovered trajectory of the robot in the environment is then obtained. ; in These are the grid coordinates corresponding to the robot's initial position. The grid coordinates represent the task completion position, and intermediate states represent the continuous movement positions of the robot during execution. Finally, the robot's actual execution trajectory is generated and sent to the robot motion control module.

[0061] When the solver returns an Unsatisfiable (UNSAT) result, it indicates that under the current environmental constraints and the LTLf task logic requirements, no finite-length trajectory can satisfy the planning objective. In this case, the present invention not only returns the planning failure result, but also further outputs the corresponding unreachability proof information to explain the reason for the task failure.

[0062] This embodiment also provides a robot task planning system based on unbounded symbolic model detection, used to execute the above-described method.

[0063] A robot task planning system based on unbounded symbolic model detection includes: The environment modeling module is used to build an environment model based on the grid map data of the robot's environment, and to merge and compress the connected domains of the passable areas in the environment model to construct an environment topology transfer system. The task logic parsing module is used to convert the task instructions to be executed by the robot into finite trajectory linear temporal logic task formulas, perform negation normal form transformation and syntax tree recursive decomposition on the logic task formulas, extract all sub-formulas, and construct a set of sub-formulas. The state encoding module is used to encode the abstract region in the environmental topology transfer system into a position state variable by binary bit vector encoding, to encode the pickable resources in the environment into a resource state variable by binary bit vector encoding, to construct a logical state variable for each sub-formula according to the sub-formula set, and to combine the position state variable, resource state variable and all logical state variables to construct a unified state variable for the robot, and to set the initial value of the unified state variable. The state transition coupling module is used to construct physical transition constraints based on the state jump functions and dynamic constraints in the environmental topology transition system, construct logical evolution constraints based on the logical semantics of each sub-formula, and combine the physical transition constraints and logical evolution constraints to generate a unified state transition relationship. The unbounded model detection and solution module is used to calculate the reachability of the target state set starting from the initial value of the unified state variables and using the unified state transition relationship as the jump rule. It iteratively approximates the reachable state space layer by layer through inductive invariants until the state space converges to a fixed point and outputs the search results. The target state set is set according to the target location in the environment model, and all committed variables are completed states. The trajectory decoding and execution module is used to decode the unified state sequence output by the solver into a two-dimensional coordinate motion trajectory that the robot can execute when the target state set is reachable; when the target state set is unreachable, it outputs proof information that the task cannot be satisfied.

[0064] The modules described above can be deployed on a robot-mounted industrial control computer or a cloud server. When deployed on a vehicle-mounted industrial control computer, the system can achieve real-time online planning and is suitable for autonomous decision-making in dynamic environments; when deployed on a cloud server, the system can handle larger-scale computing tasks and is suitable for multi-robot collaborative scheduling scenarios. The modules communicate with each other through standardized data interfaces, and the modular design facilitates system maintenance and functional upgrades.

[0065] The unbounded model detection and solution module integrates a formal verification engine based on the IC3 / PDR algorithm. IC3 / PDR (Incremental Construction of Inductive Clauses for Property Directed Reachability) is an efficient symbolic model detection algorithm that achieves efficient search of the state space through incremental construction of inductive invariants.

[0066] Example 2 This embodiment provides a specific application of a robot task planning method based on unbounded symbolic model detection in an intelligent warehousing environment.

[0067] This embodiment can be applied to an autonomous mobile robot task planning system in an intelligent warehousing environment. In this application scenario, the warehousing area consists of multiple storage areas, transportation channels, access control areas, and task execution areas. The mobile robot needs to autonomously generate an executable motion path based on the task requirements issued by the scheduling system, while satisfying spatial access constraints, safety constraints, and task timing constraints.

[0068] Specifically, intelligent warehouse robots not only need to complete simple navigation tasks such as moving from a starting position to a target position, but also complex tasks involving task sequence, area access restrictions, and resource dependencies. For example: (1) The robot first moves to the designated shelf area A to complete the task of retrieving goods; (2) Then move to area B to complete the cargo scanning or information registration task; (3) The robot should always avoid entering the manual operation area C during the execution of the task; (4) Access gates are set up in some storage areas. Robots must obtain the corresponding electronic key or access token in advance before they are allowed to enter the restricted areas. (5) After completing all tasks, transport the goods to the designated unloading area D.

[0069] The above tasks include spatial movement constraints, resource access constraints, and time sequence constraints. Traditional path planning methods based solely on the shortest distance principle cannot guarantee that all of these constraints are met.

[0070] Therefore, this invention uses finite trajectory linear temporal logic (LTLf) to formally describe the task requirements of warehouse robots.

[0071] For example, the task requirement that "the robot needs to reach the cargo area A first, then the scanning area B, and cannot enter the danger area C during the entire task execution process" can be expressed as: ; in: This indicates that the robot has arrived at goods retrieval area A; This indicates that the robot has arrived at scanning area B; This indicates a dangerous area, C, where entry is prohibited. This indicates that a certain task condition is ultimately met; This indicates that a certain safety constraint is continuously satisfied throughout the entire process.

[0072] For warehouse areas with access restrictions, such as "robots must obtain an electronic key before entering a special warehouse area," further resource dependency constraints are established: ; in Indicates whether the robot has acquired the first... Individual access resources This indicates the corresponding restricted access area.

[0073] The resource dependency constraint mentioned above is reflected in the physical migration constraint as: only if At that time, from the current area to the access control area Only after a certain number of redirects are allowed. This ensures that bots cannot enter protected areas without the necessary permissions.

[0074] The above methods can be used to uniformly transform the movement behavior, task sequence, safety requirements, and access permissions of warehouse robots during actual execution into computable task constraints.

[0075] The specific implementation steps of the above method in a smart warehousing environment are as follows: Step S101: Obtain warehouse environment data and establish a task environment model.

[0076] First, the system acquires relevant data about the intelligent warehousing environment, including: (1) two-dimensional map data of the warehousing area; (2) robot initial position; (3) shelf area position; (4) target transportation area position; (5) prohibited areas; (6) access gate position; and (7) electronic key or access resource position. A discrete grid environment model is established based on the above data.

[0077] in: G represents a grid map of the warehouse environment; O indicates an obstacle area; R represents a set of regions with task semantics, such as cargo regions, scanning regions, and unloading regions; D represents a set of restricted access areas, such as access control areas; K represents the set of available resources, such as electronic keys or access tokens; This indicates the correspondence between resources and restricted areas.

[0078] For example, in a warehousing environment: Area A: Goods picking area; Area B: Scanning and detection area; Area C: Manual work area; Area D: Cargo unloading area; Door M: Special warehouse access door; Key K: Electronic access required to open door M.

[0079] Through the above environmental modeling process, the real warehouse environment is transformed into a discrete state space that can be processed by a computer.

[0080] Step S102: Establish robot task logic constraints.

[0081] Based on the task requirements generated by the warehouse scheduling system, the natural language task is converted into an LTLf task formula.

[0082] For example: "The robot sequentially completes the tasks of acquiring, scanning, and transporting goods, and is prohibited from entering the human area throughout the entire process." Corresponding task constraints: ; in: This indicates that the robot has completed the acquisition of the goods; This indicates that the robot has completed scanning the goods; This indicates that the robot has arrived at the unloading area; This indicates the area for manual operation.

[0083] For access control tasks, "the robot can only enter the special warehouse after obtaining the electronic key," corresponding to the following resource constraints: At that time, entry is permitted. area.

[0084] The above method transforms the task requirements of warehouse robots into formal logical constraints.

[0085] Step S103: Construct a unified symbolic state model for the robot.

[0086] This invention further unifies the encoding of robot spatial state, resource state, and task logic state. Defines a unified robot state: ; in: (1) This indicates the robot's current position and is used to describe the area where the robot is currently located. (2) This indicates the resource status, used to show whether the robot has acquired the corresponding authorized resources; (3) It represents the logical state of a task, used to indicate the current task execution progress and future task constraints.

[0087] For example, the robot's initial state is: ; This indicates that the robot is at its starting position, has not yet acquired any authorized resources, and its task logic is incomplete. When the robot moves to the area where the electronic key is located... This indicates that the robot has obtained the first... Each resource has a specific permission level, which grants subsequent access to the corresponding permission area.

[0088] Step S104: Detect and generate task planning paths based on unbounded symbolic models.

[0089] After completing the construction of the unified state model, this invention uses the unbounded symbolic model detection method to solve the task.

[0090] Unlike traditional methods (LTLf formula → automaton → product state space search), this invention does not require explicit construction of an automaton, but instead proceeds synchronously in a unified symbolic state space: (1) changes in robot spatial position; (2) changes in resource state; (3) changes in LTLf task logic state.

[0091] During the solution process: if the target state is reachable, output the robot motion path that meets the task requirements; if the target state is unreachable, output the proof that the task cannot be satisfied.

[0092] Step S105: Decode the path and control the robot to perform the task.

[0093] Once the solver outputs a state sequence that satisfies the task constraints, the system further performs path decoding.

[0094] Convert the symbolic state into the robot's actual motion path: ; in Indicates the robot's initial position. The current state indicates the completed position, while intermediate states represent the continuous movement positions of the robot during execution. Finally, the robot's actual execution trajectory is generated and sent to the robot motion control module.

[0095] The robot completes the following steps according to the planning results: (1) moves to the cargo area; (2) acquires the target cargo; (3) moves to the scanning area; (4) avoids the human work area; (5) acquires the authorized resources; (6) passes through the authorized gate; (7) arrives at the unloading area.

[0096] Therefore, this invention employs a robot task planning method and system based on unbounded symbolic model detection. By introducing unbounded symbolic model detection and LTLf native logic modeling mechanism, it achieves efficient solution to robot task planning problems under complex temporal constraints, outperforming existing technologies in planning capability, solution efficiency, scalability, and reliability. This invention can be effectively applied to practical intelligent equipment systems such as intelligent warehousing robots, autonomous inspection robots, and automated delivery robots, enabling autonomous task planning and control that meets temporal constraints in complex environments.

[0097] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A robot task planning method based on unbounded symbolic model detection, characterized in that, Includes the following steps: S1. Obtain grid map data of the robot's working environment, establish an environment model, and merge and compress the connected components of the passable areas in the environment model to construct an environment topology transfer system. S2. Obtain the task instructions to be executed by the robot, convert the task instructions into finite trajectory linear temporal logic task formulas, perform negation normalization transformation and syntax tree recursive decomposition on the logic task formulas, extract all sub-formulas, and construct a set of sub-formulas. S3. Perform binary bit vector encoding on the abstract region in the environmental topology transfer system to obtain the location state variables; perform binary bit vector encoding on the pickable resources in the environment to obtain the resource state variables; S4. Based on the set of sub-formulas, construct logical state variables for each sub-formula. The logical state variables include the current step logical variable, the next step logical variable, and the commitment variable. Combine the position state variables, resource state variables, and all logical state variables to construct the robot's unified state variables, and set the initial values ​​of the unified state variables. Construct physical migration constraints based on the state transition functions and dynamic constraints in the environmental topology transfer system. Construct logical evolution constraints based on the logical semantics of each sub-formula. Combine the physical migration constraints and logical evolution constraints to generate a unified state transition relationship. S5. Based on the target location in the environmental model, set all commitment variables to be in a completed state and construct a target state set; Starting with the initial values ​​of unified state variables and using unified state transition relations as the jump rules, an unbounded symbolic model detection solver is used to calculate the reachability of the target state set. The unbounded symbolic model detection solver iteratively approximates the reachable state space layer by layer through inductive invariants until the state space converges to a fixed point. S6. When the target state set is reachable, decode the unified state sequence output by the solver into a two-dimensional coordinate motion trajectory that the robot can execute; when the target state set is unreachable, output proof information that the task cannot be satisfied.

2. The method according to claim 1, characterized in that, S1 specifically includes: S11. Obtain a grid map of the robot's environment, and extract the coordinate sets of obstacles, regions of interest, access control areas, pickable resources, and the unlocking mapping table between resources and access control. S12. Obtain the robot's starting position and target position in the grid map; S13. Establish an environmental model ,in For raster maps, A collection of obstacles. For a set of regions of interest, Gathering for access control areas, For the set of pickable resources, This establishes a mapping relationship between resources and access control unlocking. This is a function that maps the location of resources in a raster map. For the robot's starting position, For the target location; S14, Through abstract mapping functions Connectivity merging is performed on the passable grid regions in the environment model, and interconnected free grids are clustered into a topological node to obtain an abstract region set. ; S15. Construct an environmental topology transfer system ,in For the robot's abstract state set, It is the set of actions that a robot can perform. As the starting point, abstract state Abstract state for the target. This is a function for switching between regions. For the set of atomic propositions, This is a region label mapping function. These are the dynamic constraints for access control resources.

3. The method according to claim 1, characterized in that, S2 specifically includes: S21. Read the task instructions to be executed by the robot and convert them into finite trajectory linear time-series logic task formulas. ; S22, Regarding the logical task formula By performing a negation paradigm shift, using De Morgan's law and the equivalence rule of temporal logic, all logical NOT operators are moved to the front of the atomic proposition, so that the transformed formula does not contain any negation symbols that act on the composite logic structure or temporal operators. S23. Perform recursive syntax tree decomposition on the transformed logical task formula, extract all sub-formulas, and construct a set of sub-formulas. .

4. The method according to claim 3, characterized in that, In S3: Position state variables Using a finite-width binary vector representation, the bit width of the position state variable is... , The number of abstract regions; the position state variable is used to uniquely identify the abstract region where the robot is currently located. Resource state variables Represented using binary vectors, each dimension of the resource state vector corresponds to a pickable resource, indicating whether the resource has been acquired by the robot.

5. The method according to claim 4, characterized in that, In S4: Current step logical variable Used to represent the sub-formula corresponding to the current time. Is it valid? Next step logical variable Used to represent the corresponding sub-formula at the next moment The logical value is used to pass the logical constraints that are not completed at the current time to the next time. Commitment variables Used to mark the unfinished sequential tasks corresponding to the final operator, the start operator, and the until operator in the formula of a linear sequential logic task with a finite trajectory. The commitment variable takes the value of true when the corresponding sequential task has not been completed and takes the value of false when the corresponding sequential task has been completed. All sub-formulas satisfy the logical consistency constraint: ,in Sub-formula The logical semantics at the current moment satisfy the condition.

6. The method according to claim 5, characterized in that, In S4: Physical state is represented as ; The unified state variable is represented as ,in This is the set of current-step logical variables, commitment variables, and next-step logical variables corresponding to all sub-formulas; The initial values ​​of the unified state variables are set as follows: position state variables Retrieve the abstract region code corresponding to the robot's starting position and the resource state variables. All dimensions are set to false, the current step logical variable corresponding to the top-level formula is set to true, and all other logical state variables are set to false.

7. The method according to claim 6, characterized in that, The physical migration constraints and logical evolution constraints in S4 specifically include: S41. Based on the inter-region jump function Dynamic constraints on access control resources Construct physical migration constraints Physical migration constraints include movement constraints, resource acquisition constraints, and restricted area access constraints. The movement constraint is: the robot is allowed to perform a movement operation when the target abstract region and the current abstract region are adjacent. Resource acquisition constraint is: when there is a pickable resource at the robot's current location, the corresponding resource state variable is determined according to... Updated to the "acquired" status and maintained thereafter. For the next moment The status of a resource For the first The location of each resource; The restricted area access constraint is: the robot is allowed to enter the restricted area only if the resource state variable corresponding to the restricted area is in the acquired state; S42. Based on the set of sub-formulas The temporal logical semantics of each sub-formula are used to construct logical evolution constraints; the logical evolution constraints include the semantic encoding of the next step operator, the final operator, the always operator, and the until operator; For the next operator, the semantic constraints are: ,in For the next step operator and sub-formula The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, For sub-formula The next logical variable, This is a flag variable indicating whether the current state is the end point of the robot's trajectory; For the final operator, the semantic constraints are: ,in, For the final operator and sub-formula The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, For sub-formula The next logical variable; For the time operator, the semantic constraints are: ,in, For the sake of the operators and sub-formulas The composite formula, Representing the formula for the compound The logically satisfied state at the current moment, For sub-formula The next logical variable; For the until operator, the semantic constraints are: ,in, For sub-formula , sub formula The composite formula composed of operators up to the end, Representing the formula for the compound The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, Sub-formula The logically satisfied state at the current moment, For sub-formula The next logical variable; S43. Perform a logical AND operation between the physical migration constraints and all logical evolution constraints to generate a unified state transition relationship. ,in Sub-formula The logical value to be taken at the next moment. Sub-formula The logical semantics at the current moment satisfy the conditions; the unified state transition relationship forces the robot's physical displacement and logical task progress to be completed synchronously within a single state update cycle.

8. The method according to claim 1, characterized in that, In S5: Target state set The construction conditions are: and The robot's current position has reached the target abstract state. And the complete set of commitment variables All commitment variables are in a completed state; The unbounded symbolic model detection solver constructs a sequence of reachable state frames layer by layer using inductive invariants. ,in , The initial values ​​for the unified state variables, and , ; when When the state space search reaches a fixed point, it is determined that the search has been completed. If the target state set If the target state set is [value], then it is determined to be reachable; if the target state set is [value], then it If so, it is determined to be unreachable.

9. A robot task planning system based on unbounded symbolic model detection, used to execute the method according to any one of claims 1-8, characterized in that, include: The environment modeling module is used to build an environment model based on the grid map data of the robot's environment, and to merge and compress the connected domains of the passable areas in the environment model to construct an environment topology transfer system. The task logic parsing module is used to convert the task instructions to be executed by the robot into finite trajectory linear temporal logic task formulas, perform negation normal form transformation and syntax tree recursive decomposition on the logic task formulas, extract all sub-formulas, and construct a set of sub-formulas. The state encoding module is used to encode the abstract region in the environmental topology transfer system into a position state variable by binary bit vector encoding, to encode the pickable resources in the environment into a resource state variable by binary bit vector encoding, to construct a logical state variable for each sub-formula according to the sub-formula set, and to combine the position state variable, resource state variable and all logical state variables to construct a unified state variable for the robot, and to set the initial value of the unified state variable. The state transition coupling module is used to construct physical transition constraints based on the state jump function and dynamic constraints in the environmental topology transition system, construct logical evolution constraints based on the logical semantics of each sub-formula, and combine the physical transition constraints and logical evolution constraints to generate a unified state transition relationship. The unbounded model detection and solution module is used to calculate the reachability of the target state set starting from the initial value of the unified state variables and using the unified state transition relationship as the jump rule. It iteratively approximates the reachable state space layer by layer through inductive invariants until the state space converges to a fixed point and outputs the search results. The target state set is set according to the target location in the environment model, and all committed variables are completed states. The trajectory decoding and execution module is used to decode the unified state sequence output by the solver into a two-dimensional coordinate motion trajectory that the robot can execute when the target state set is reachable; when the target state set is unreachable, it outputs proof information that the task cannot be satisfied.

Images