A task planning method, apparatus, medium, and product for a multi-robot system

Abstract

This invention discloses a task planning method, apparatus, medium, and product for a multi-robot system, relating to the field of robot task planning technology. The method includes: constructing a multi-robot temporal task and converting it into a nondeterministic Büchi automaton; modeling the workspace of each robot using a state transition system to obtain a tuple corresponding to each robot; constructing a search tree based on the nondeterministic Büchi automaton and determining several feasible paths based on the search tree; each feasible path is divided into a prefix and a suffix; calculating the cost of each feasible path using a cost objective function; and determining the optimal path planning based on the cost of all feasible paths. This invention constructs a search tree based on the nondeterministic Büchi automaton and considers the feasible region of each robot's task, improving the quality of sampling points and significantly enhancing the optimality, stability, and efficiency of the algorithm solution.

Description

Technical Field

[0001] This invention relates to the field of robot task planning technology, and in particular to a task planning method, apparatus, medium and product for multi-robot systems under temporal logic task constraints. Background Technology

[0002] Traditional motion planning involves generating a robot trajectory from a starting point to a target area while avoiding obstacles. Point-to-point navigation methods utilize potential fields and navigation functions, as well as sampling-based algorithms. Recently, a new class of planning methods has been proposed that can handle a wider range of tasks than classic point-to-point navigation. These tasks can include sorting or covering, data collection, etc., and can be captured using a formal language, namely Linear Temporal Logic (LTL) task planning.

[0003] In complex tasks, the task specification of a mobile robot is captured by an LTL formula. When each agent in the system has its own task specification, a bottom-up approach is generally considered. When a global LTL formula describing the overall task needs to be assigned to the robot team, a top-down approach is used. In practice, people often consider not only the task completion but also the robot's operating cost. The optimal task sequence problem under the LTL specification was solved in the work of Smith SL, Guo M, XLuo, and JWang, thus forming the most classic temporal logic task planning framework. Specifically, a synchronous product automaton can be constructed by taking the Cartesian product of a weighted state transition system (WTS) and a nondeterministic Büchi automaton (NBA). Then, the optimal path was successfully found in the synchronous product automaton using Dijkstra's algorithm, and the experimental results well demonstrate the feasibility of this algorithm. For multi-robot systems, Smith SL constructs a team-based weighted switching system by weighting the switching of individual robots, similarly building a team product automaton and using graph search techniques to obtain the optimal motion plan. D.Khalidi, in his work, proposed a novel planning algorithm T*, which opportunistically generates the optimal trajectory that satisfies the time-logic query using an A* search process. Experimental results show that it achieves an order-of-magnitude improvement over classical algorithms. However, with the increase in the number of robots or the expansion of the NBA scale, the state space of the product automaton grows exponentially, making graph search techniques difficult to solve effectively.

[0004] Unlike graph search approaches, Karaman S proposed a sampling-based algorithm that incrementally builds the Kripke structure until it has sufficient expressive power to generate motion plans that satisfy the task specification of deterministic μ-calculus representation. Inspired by the RRT* algorithm, Kantaros Y proposed an optimal task planning algorithm for multi-robot systems under a global time specification. This algorithm avoids explicitly building product automata between the transition system and the NBA, instead incrementally building a directed tree using a sampling-based approach. This directed tree approximately represents the state space and state transitions of the synchronous product automaton, and finally, the optimal task sequence is obtained through backtracking.

[0005] However, for many temporal logic constraints, there are sequential dependencies between tasks. Kantaros Y's algorithm does not consider the NBA constraint in its sampling method; instead, it samples and then checks, resulting in low sampling quality. This makes it difficult to effectively solve LTL constraints with many sequential dependencies, leading to unsatisfactory solution efficiency and performance. XLuo uses Büchi automata for biased sampling, but the task scenario is biased towards path planning and it does not fully utilize the automata information. Summary of the Invention

[0006] The purpose of this invention is to provide a task planning method, apparatus, medium, and product for multi-robot systems, which can improve the efficiency and quality of task planning.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A task planning method for a multi-robot system includes:

[0009] Construct multi-robot sequential tasks and convert them into nondeterministic Büchi automata; multi-robot sequential tasks involve multiple robots performing tasks within the same workspace;

[0010] The workspace of each robot is modeled using a state transition system, resulting in the tuple wTS corresponding to each robot. i ;

[0011] A search tree is constructed based on the nondeterministic Büchi automaton, and several feasible paths are determined based on the search tree; the feasible paths are divided into prefix parts and suffix parts;

[0012] For each feasible path, the cost cost corresponding to each feasible path is calculated using the cost objective function; the optimal path planning is determined based on the cost costs corresponding to all feasible paths.

[0013] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the task planning method for a multi-robot system described above.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the task planning method for a multi-robot system described above.

[0015] A computer program product includes a computer program that, when executed by a processor, implements the steps of the task planning method for a multi-robot system described above.

[0016] According to specific embodiments provided by the present invention, the following technical effects are disclosed: The present invention provides a task planning method, apparatus, medium, and product for a multi-robot system, constructs a multi-robot temporal task, and converts the multi-robot temporal task into a nondeterministic Büchi automaton; the multi-robot temporal task involves multiple robots performing tasks within the same workspace; a state transition system is used to model the workspace of each robot, obtaining the tuple wTS corresponding to each robot. i This invention constructs a search tree based on a nondeterministic Büchi automaton and determines several feasible paths based on the search tree. Each feasible path is divided into a prefix and a suffix. For each feasible path, the cost is calculated using a cost objective function. The optimal path plan is determined based on the cost of all feasible paths. This invention constructs a search tree based on a nondeterministic Büchi automaton and considers the feasible region of each robot's task, thus improving the quality of sampling points and significantly enhancing the optimality, stability, and efficiency of the algorithm's solution. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the task planning method for a multi-robot system provided in Embodiment 1 of the present invention;

[0019] Figure 2 This is a schematic diagram illustrating the specific implementation process of the task planning method for a multi-robot system provided in Embodiment 1 of the present invention;

[0020] Figure 3 This is a schematic diagram of a task scenario provided in Embodiment 1 of the present invention;

[0021] Figure 4 This is a schematic diagram of the reconstruction tree process provided in Embodiment 1 of the present invention;

[0022] Figure 5 This is a schematic diagram of a local optimization of the NBA guided sampling and auction algorithm provided in Embodiment 1 of the present invention;

[0023] Figure 6 This is a schematic diagram of the tree expansion and reconnection process provided in Embodiment 1 of the present invention;

[0024] Figure 7 This is a schematic diagram of the task planning results provided in Embodiment 1 of the present invention;

[0025] Figure 8 This diagram illustrates the comparison between the cost of the task planning method for the multi-robot system provided in Embodiment 1 of the present invention and the algorithm provided by the existing Kantaros Y, as the cost varies with the number of iterations.

[0026] Figure 9 A schematic diagram showing the comparison of the running time of the task planning method for the multi-robot system provided in Embodiment 1 of the present invention with that of the existing algorithm provided by Kantaros Y as a function of iteration number;

[0027] Figure 10 An internal structural diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0028] 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, and 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.

[0029] The purpose of this invention is to provide a task planning method, apparatus, medium, and product for a multi-robot system. It aims to construct a search tree based on a nondeterministic Büchi automaton, and takes into account the task-feasible region of each robot, thereby improving the quality of sampling points and significantly enhancing the optimality, stability, and efficiency of the algorithm solution.

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Example 1

[0032] like Figure 1As shown, this embodiment provides a task planning method for a multi-robot system, characterized by including:

[0033] S1: Construct a multi-robot time-series task and convert it into a nondeterministic Büchi automaton; the multi-robot time-series task involves multiple robots performing tasks within the same workspace.

[0034] S2: Model the workspace of each robot using a state transition system to obtain the tuple wTS corresponding to each robot. i .

[0035] S3: Construct a search tree based on the nondeterministic Büchi automaton, and determine several feasible paths based on the search tree; the feasible paths are divided into prefix parts and suffix parts.

[0036] S4: For each feasible path, calculate the cost cost corresponding to each feasible path using the cost objective function; determine the optimal path planning based on the cost costs corresponding to all feasible paths.

[0037] This embodiment will formally describe LTL by introducing its syntax and semantics. First, we will briefly review the basics of automata-based LTL model detection.

[0038] LTL is a form of formal logic whose basic components consist of a set of atomic propositions (APs) and Boolean operators such as conjunction (^), negation, etc. In addition, it includes unique timing operators, two of which are "next time" (○) and "until a certain time" (U). Linear temporal logic based on a set of atomic propositions can be constructed using the following syntax: Where π∈AP. For the sake of brevity, derivations of other Boolean and time operators, such as always □, eventually ◇, and implicit, are not given here. wait.

[0039] Defined in alphabet 2 AP An infinite string σ on the π / 2 is defined as a sequence σ = π0π1π2...∈(2π / 2) AP ) ω Where ω represents infinite repetition, and π k ∈2 AP .definition To satisfy the LTL formula The set, where (2 Π ) ω Indicates the alphabet is 2 Π All infinite strings, To satisfy the relationship.

[0040] Any LTL formula This can be translated into a definition in 2 AP Uncertain Buchi automata (NBA), defined as follows: defined in 2 AP The nondeterministic Buchi automata on the above are represented by tuples. Among them, Q B It is a set of states in a nondeterministic Buchi automaton. It is worth noting that the states of the automaton only represent different situations corresponding to the LTL task specification and have no clear practical meaning. Denotes the initial state of a nondeterministic Buchi automaton; Σ = 2 AP This is a character table that reflects possible subsets of the set of atomic propositions; This represents the state transition relationship. Whether the transition is satisfied depends on whether the label corresponding to the task sequence satisfies the transition condition Σ of the automaton. This indicates the acceptable or terminated state of the automaton.

[0041] if Then ρ B This is considered acceptable. If a running ρ B From an initial state Start, and in an accepting state. If the sequence ends, then it is in an accepting state. A finite sequence σ satisfies the LTL formula if at least one run is acceptable.

[0042] Step S1 is used for LTL task planning problem modeling, as follows:

[0043] The method in this embodiment is applicable to multi-robot systems. Consider a multi-robot system containing N mobile robots in a complex workspace W∧R. d It moves according to a certain dynamic law: Where x i (t) and u i (t) represents the position and control inputs associated with robot i, i ∈ {1, ..., N}. Assume there are multiple disjoint region values ​​in W. Each region is represented by W. j This indicates that it can be of any shape. Each robot has its specific capabilities, dynamics, or task functions, enabling it to perform specific tasks within its environment. Given robot dynamics or task capabilities, the robot's mobility within the workspace W can be represented by a weighted transition system (WTS), also known as a state transition system. The WTS of robot i is defined as follows:

[0044] The state transition system is a six-tuple, i.e. Among them, wTS i S is a tuple of i robots; iRepresents the set of states of the task environment. This represents the initial state of the agent. This represents the allowed state transition relationships in the task environment, where W i :S i ×S i →R + This represents the cost or weight of the corresponding state transition; AP i It is the set of atomic propositions for the i-th robot, where each atomic proposition corresponds to whether the agent has reached a certain state, and is true if it has. It represents a label function, which is a mapping from a specific state to a subset of the set of atomic propositions. It can be understood as an observation or label of a certain state.

[0045] Multi-robot systems require defining a Team Product Transformation System (PTS) based on the above. Building upon the definition of a given Workload Switch (WTS), the Team Product Transformation System (PTS) is defined, which captures the robots' performance within their respective WTSs. i All possible combinations of states are shown below:

[0046] Given N state transition systems wTS i , It can be represented by the following tuples: Q PTS =Q1×Q2×…Q N It is a set of states. This represents the initial state. The state transition rules are defined by the following rules: Based on this, the state q′ is defined. PTS This transformation rule means that if there exists a transformation from q... i to q i The transformation of ' for all i∈{1,...,N}, then there exists a transformation from q PTS to q′ PTS The conversion. w PTS :Q PTS ×Q PTS →R + It is the cost function that represents the weight / cost assignment for each possible transformation in the PTS, defined as follows: in q PTS Represents state q PTS In WTS i Projection onto state space; It is a set of atomic propositions; and, It is an observation / output relationship, given in state q PTS ∈Q PTS The set of atomic propositions that satisfy the condition.

[0047] Define an infinite path τ in a PTS as an infinite sequence of states, τ = τ(1)τ(2)τ(3)... such that... Here, k is the index pointing to the k-th term of τ, denoted as τ(k). A finite path in a PTS can be defined accordingly. The only difference from an infinite path is that a finite path is defined as a finite sequence of states in the PTS. Given weight w... PTS Given the definition, the cost of a finite path τ is expressed as: Its definition is:

[0048]

[0049] Where |τ| represents the number of states in τ.

[0050] Task requirements: The task constraints in W consist of a global LTL semantic sequence. This indicates that L(S)∈2 in PTS AP It consists of Boolean operators. If the corresponding robot reaches the marked area in the PTS, the corresponding atomic proposition is true, and according to its action, it indicates that the task of the corresponding state can be completed. By considering other robots and state points together, a task path that satisfies the acceptance definition described above can be generated.

[0051] Given the LTL formula If there exists a satisfying If the path is such that there exists a path, then there is a path. This path can be represented in a finite form, called a prefix-suffix structure, i.e., τ = τ pre [τ suf ] ω , where the prefix τ pre Executes only once, followed by the suffix τ. suf The task is executed indefinitely. Therefore, the goal is to compute a task sequence τ = τ pre [τ suf ] ω =Π| PTS p pre [Π| PTS p suf ] ω , among which Π| PTS Representing the state space Q PTS The projection onto the surface is used to minimize the following cost objective function, which is as follows:

[0052]

[0053] The cost objective function captures the total cost incurred by all robots during the execution of the prefix and a single execution of the suffix, where, The cost incurred by all robots during the execution of the prefix portion; Costs incurred by all robots during a single execution of the suffix portion.

[0054] The problems considered in this embodiment can be summarized as follows, without loss of generality: Figure 3 The combat scenario is described below:

[0055] The problem that needs to be solved is task allocation for robots to cooperate in reconnaissance, detection, and attack missions. The characteristics of this problem include:

[0056] 1) The establishment of WTS only focuses on the meaningful key nodes in the task execution of each robot, and does not perform redundant modeling of intermediate processes;

[0057] 2) Due to the complexity of the tasks, each task point may contain multiple task contents, and each robot may have different abilities to perform the tasks;

[0058] 3) This embodiment focuses on issues that do not contain (always) atomic propositions, i.e., it focuses on the prefix part. This is because the problems encountered in this scenario almost never have a suffix part.

[0059] Based on the above description, the robots must collectively satisfy the global LTL task specification defined on a set of tasks. N robots operate in an environment W, assuming that the mobility and capability i of each robot are modeled as a weighted transition system wTS. i For all robots i, determine if the following conditions are met. The discrete team plan τ, i.e. And minimize the cost function (i.e., formula (2)).

[0060] Step S3 is used to construct the search tree within the NBA-guided heuristic sampling optimal planning algorithm framework. The specific process is as follows:

[0061] This embodiment proposes a heuristic sampling-based algorithm. The process incrementally constructs a directed tree and redesigns both the sampling and tree construction processes, enabling efficient solution of discrete motion plans τ while satisfying the given global LTL specification. The entire task planning process is as follows: Figure 2 As shown.

[0062] Used in trees To represent a search tree, use express The root node. The set contains Q that has been sampled and added to the tree structure. P The state. If there exists a state... to state The transformation of edge set E T capture The transformation between nodes, i.e. (q) P ,q′ P )∈E T .function Indicates from the root of the tree arrive The cost of a node, i.e. Where, τ T It is a tree Connect the root to q P The path.

[0063] The tree construction method proposed in this embodiment utilizes the temporal state information of the currently sampled node (i.e., the current sampling point). Therefore, the current situation is clear. Corresponding to states in the NBA, these can be directly added to the tree. Furthermore, existing nodes in the tree selected for this sampling are utilized. like Figure 4 As shown in Algorithm 1, performing a reconnection operation based on the node in each iteration can more efficiently guarantee the optimality of the search tree connection relationship.

[0064]

[0065] Step S3 involves constructing a search tree based on the nondeterministic Büchi automata, specifically including:

[0066] Step S31: Sample the nodes in the search tree using the nondeterministic Büchi automaton to obtain the current sampling point; the current sampling point and its parent node satisfy the transition conditions in the nondeterministic Büchi automaton, and the transition conditions in the nondeterministic Büchi automaton are:

[0067] Step S31 is used for NBA-guided heuristic sampling:

[0068] In the LTL task planning problem, the newly sampled node must satisfy the transition conditions in NBA 2K with its parent node, i.e. Only then can it be allowed to be added to the tree. The sampling steps proposed in this embodiment are shown in Algorithm 2:

[0069]

[0070]

[0071] First, sample a node from the search tree according to the given distribution that satisfies q(B) ≠ accept, denoted as q(B) ≠ accept. Then use S childRecord each robot from The states that can be reached from the given state, i.e., those that satisfy... The state. Next, based on... In the NBA state section, the buchi_inspired function is used to find nondeterministic Büchi automata from... Set of target states that can be transitioned Furthermore, noting the system's robot count limitation, only states where the number of APs simultaneously required to be true in the transfer condition does not exceed the number of robots are stored. Based on the above, random selection... A timing state As the target state to be transferred ( Figure 5 (This vividly reflects the above process). Combining the above two parts, Algorithm 2 is used to... child Choose one that matches from arrive The conditions and the optimal result. If all The result of the guided transfer in S child If a corresponding state can be found and the number of such states equals N, then the result is directly set to N. If the number is less than N, then the corresponding robot position is set to Other positions remain unchanged. If there are... The result of the guided transfer in S child The corresponding state, i.e., flag, cannot be found. trans If the result is False, it indicates that the timing transition may not be feasible and resampling is required.

[0072] Step S32: Using the auction algorithm, determine the optimal transition result based on the list of atomic propositions for the current sampling point; add the current sampling point to the search tree based on the optimal transition result. Specifically:

[0073] According to the pending From the guided AP list, select the optimal transfer outcome and determine whether that outcome can be achieved in S. child The algorithm is found within this framework. Its core consists of two parts: first, constructing a value matrix, which is represented row-wise. Each guided AP is represented by a column indicating the robot's sequence number. If a robot can transition from its current state to the state corresponding to when the AP is true, its value is recorded as:

[0074]

[0075] Otherwise, it is recorded as 0. Where V m This represents the maximum cost limit for all state transitions. This indicates that the cost of completing this transfer is the same as the cost recorded in the PTS. The second part is... S child The auctioneer and the value matrix are respectively input into the auction algorithm as the auctioneer, the auctioned item, and the value matrix, ultimately yielding the optimal allocation result for the current step. The reason for defining the value matrix in this way is as follows: Figure 5 Therefore, a final selection result of 0 indicates that the transition is not feasible. It is important to note that since the PTS state and AP are not actually equivalent, a conversion is required during the algorithm process. This is the purpose of pts2AP and AP2pts in Algorithms 2 and 3.

[0076]

[0077] This yields heuristic sampling guided by the NBA. The sampling result is determined based on the transition conditions in NBA, and it is guaranteed that the transition can be obtained from the current state. Furthermore, the auction algorithm ensures the optimal matching relationship within the process. Therefore, sampling using the above method can greatly improve the solution efficiency and quality.

[0078] After obtaining the current sampling point, the process also includes node expansion and reconnection, as detailed in steps S33 and S34.

[0079] Step S33: Use the Extend function to add the current sampling point to the search tree and construct the set. According to the set Determine whether to add the current sampling point to the search tree; if the set If empty, the node will not be... Add to the search tree; if the set If it is not empty, then the node will be... Add to the search tree; set Defined as in, Includes Q that has been sampled and added to the search tree structure P The set of states; This is the current sampling point.

[0080] In each iteration, after obtaining the sampling points Next, the node needs to be added to the tree, ensuring asymptotic optimization of the connections from the root node to the terminal node. The Extend function will... Add to the tree. Its construct set. Defined as:

[0081]

[0082] The set contains all states that satisfy the transition rules. That is, all those that can be reached directly The state of the result set. If empty, the node will not be... Add to the tree. If Then the state Add to the tree and select The parent node is:

[0083]

[0084] Step S34 is used for reconnection, and the reconnection process is as follows:

[0085] Rewiring In, nodes can be used Connect to root For nodes, if such reconnection can reduce their cost (q) P The reconnection process is as follows: Figure 6 As shown. In the rewire function, the reachable set is first constructed. Defined as

[0086]

[0087] gather Satisfy transfer rules All states, i.e. All directly reachable states. Then, for all states... Check their current cost Cost(q) P Whether they are greater than them The cost of connecting to the root node. If the node... In this case, then q P Set the new parent node to And delete the connection q P The edge between it and its original parent node. Node q P The cost was updated to

[0088] The sampling, expansion, and reconnection process is repeated until the maximum number of iterations is reached. After constructing the search tree, all feasible paths in the tree can be easily found. τ is constructed by tracing the sequence of parent nodes of nodes starting from all nodes in accepting states and ending at the root of the tree. Finally, the optimal plan τ is obtained by selecting the path with the minimum total cost among all feasible paths using the cost objective function. * (i.e., optimal path planning).

[0089] To demonstrate the effectiveness of the task planning method for multi-robot systems proposed in this embodiment, a specific task is simulated. In this embodiment, a task planning problem with a total of 14,580 states for both PTS and NBA is considered. In this problem study, the task planning method for multi-robot systems provided in this embodiment is compared with the method in Kantaros Y's work, and the performance of the proposed algorithm is examined in terms of running time and optimality.

[0090] The specific problem task scenario studied in this embodiment is roughly as follows: Figure 3 The descriptions are consistent. The PTS definitions for each robot are shown in Tables 1 and 2:

[0091] Table 1. System state space settings in simulation.

[0092]

[0093] Table 2 shows the system state space settings in the simulation.

[0094]

[0095] Consider a robot system with N=4 robots, where the state spaces of two robots are described by Table 1 and the state spaces of the other two are described by Table 2. This indicates that the robots in the system have different task capabilities. The task capabilities described in Table 1 represent the robots' ability to detect all task points in the environment, while Table 2 represents the attack task, which reaches r. i The mission points have no practical use (the corresponding AP is meaningless), and can only be attacked at mission points g1 and g2 (and the AP is different from that in Table 1).

[0096] The LTL task constraints to be considered are as follows:

[0097]

[0098] &&(!g1U r1 )&&(!g1U r2 )&&(!g2U r2 )&&(!g2U r3 )

[0099] &&(!g1cUg1)&&(!g2cUg2)

[0100] The sequential task means that g1 and g2 need to be scouted and attacked in the end. Before scouring g1, r1 and r2 need to be scouted first, and before scouring g2, r2 and r3 need to be scouted first. Before attacking g1, g1 needs to be scouted first, and before attacking g2, g2 needs to be scouted first. This means that r4 and r5 in the state space are invalid task points in this task.

[0101] The simulation results using the method proposed in this embodiment are shown in Table 3:

[0102] Table 3. Sequence of Task Planning Results

[0103]

[0104] The above is the optimal task sequence (task planning result sequence) under the requirements of this task, which is visualized as follows: Figure 7 The above results indicate that the algorithm can find a plan sequence τ that meets the LTL task constraints and achieves the optimal value.

[0105] Simulations were also performed using the method described in Kantaros Y's work. The comparison of algorithm optimality and runtime is as follows: Figure 8 and Figure 9 As shown.

[0106] The results of the optimality comparison show that, under multiple experiments, the algorithm proposed in this embodiment has a faster convergence speed and a lower average cost for the searched solutions. Furthermore, the method proposed in this embodiment is not affected by non-critical task points in the state space.

[0107] The results of the runtime comparison show that the algorithm proposed in this embodiment runs significantly faster than the method in Kantaros Y's work (this is already an order of magnitude faster than the direct search method), and is less affected by the number of iterations. Finally, it is worth mentioning that the comparison shows that the method in Kantaros Y's work successfully found a feasible solution in all cases. According to statistics, under 100 iterations, the method found a feasible solution in only 40% of cases, while the method proposed in this embodiment found a feasible solution in 100% of cases. This means that in the task planning scenario defined in this embodiment, the method in Kantaros Y's work is almost ineffective, while the method proposed in this embodiment basically meets the practical requirements.

[0108] This embodiment also tested scenarios with more critical task points or more complex LTL task constraints in the environment, and the results were similar to those described above, so they will not be repeated here.

[0109] This embodiment proposes a novel sampling planning algorithm based on NBA-guided algorithm. First, a class of LTL task planning problems is defined, applicable to scenarios where task points have different meanings and robots have varying execution capabilities. A new sampling algorithm framework is proposed, employing a more efficient method for constructing search trees, reducing the number of tree nodes. Simultaneously, sampling points are based on NBA guidance, introducing an auction algorithm to optimize local matching results, significantly improving the quality of sampling points. Finally, numerical simulations verify the algorithm's optimality, stability, and significant advantages in time consumption. Future work will consider scenarios where the environment and task are dynamic.

[0110] This embodiment has the following beneficial effects:

[0111] First, this embodiment considers and defines a certain LTL task planning problem, which mainly focuses on the meaningful key nodes of each robot in the task execution process, and there are many sequential dependencies of task constraints.

[0112] Secondly, this embodiment proposes a novel sampling method where sampling points gradually reach an acceptable state based on NBA guidance. Simultaneously, considering the task-feasible region for each robot, an auction algorithm is introduced to adjust local matching relationships. This significantly improves the quality of sampling points and substantially enhances the optimality and stability of the algorithm's solution.

[0113] Existing technologies often waste a large number of sampling nodes and have low sampling efficiency by simply sampling the next state arbitrarily within the scalable space of the current state and then performing the aforementioned checks. Therefore, this embodiment proposes an improved search tree construction framework that does not require traversing all NBA states in each sampling. The algorithm's time loss is significantly better than the method proposed by Kantaros Y, and it is less affected by the number of iterations and the number of NBA states.

[0114] Example 2

[0115] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a task planning method for a multi-robot system according to Embodiment 1.

[0116] Example 3

[0117] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a task planning method for a multi-robot system as described in Embodiment 1.

[0118] Example 4

[0119] A computer program product includes a computer program that, when executed by a processor, implements the steps of a task planning method for a multi-robot system as described in Embodiment 1.

[0120] Example 5

[0121] A computer device, which may be a database, may have an internal structure diagram as shown below. Figure 10As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores pending transactions. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a task planning method for a multi-robot system as described in Embodiment 1.

[0122] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this invention are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0123] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided by this invention may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided by this invention may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0124] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0125] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method of task planning for a multi-robot system, characterized by, include: Construct a multi-robot system time-series task and convert the multi-robot time-series task into a nondeterministic Büchi automaton; Multi-robot system sequential tasks involve multiple robots performing tasks within the same workspace. The workspaces of each robot are modeled using a state-transition system, resulting in a multiset corresponding to each robot ; A search tree is constructed based on the nondeterministic Büchi automaton, and several feasible paths are determined based on the search tree; The feasible path is divided into a prefix part and a suffix part; Constructing a search tree based on the nondeterministic Büchi automata specifically includes: The nodes in the search tree are sampled using the nondeterministic Büchi automaton to obtain the current sampling point. The current sampling point and its parent node satisfy the transition conditions in the nondeterministic Büchi automaton. The transition conditions in the nondeterministic Büchi automaton are as follows: ; Indicates state transition relationships; This represents the allowed state transition relationships in the task environment; Indicates a sum; sample a sum from the search tree according to a given distribution that satisfies... The node is denoted as ;use Record each robot from The states that can be reached from the given state satisfy the following conditions: The state; Represents the state transition rules; according to The NBA status section in the article uses... Functions searching for nondeterministic Büchi automata from Set of target states that can be transitioned Furthermore, noting the system's robot count limitation, only states where the number of APs (Atomic Propositions) simultaneously requiring a transition condition to be true are stored, with the number of such states not exceeding the number of robots; APs represent atomic propositions; and they are randomly selected. A timing state As the target state to be transferred; in Choose one that matches from arrive The conditions and the optimal result; if all The results of the guided transfer are in If a corresponding state can be found and the number of such states equals N, then the result is directly set to N. If the number is less than N, then the corresponding robot position is set to... Other positions remain in their original state; if there are The results of the guided transfer are in The corresponding state cannot be found. This indicates that the time transfer may not be feasible and resampling is required. ; Using an auction algorithm, the optimal transition result is determined based on the list of atomic propositions for the current sample point; based on the optimal transition result, the current sample point is added to the search tree. Specifically: based on the undetermined... From the guided AP list, select the optimal transfer outcome and determine whether that outcome can be achieved. If the robot can transition from its current state to the state corresponding to when AP is true, its value is recorded as follows: Otherwise, it is recorded as 0; where, This represents the maximum cost limit for all state transitions. This indicates the cost of completing this transfer; For each feasible path, the cost corresponding to the feasible path is calculated using the cost objective function; the optimal path planning is determined based on the costs corresponding to all feasible paths.

2. The task planning method for a multi-robot system according to claim 1, characterized in that, Also includes: Define a nondeterministic Büchi automaton as a tuple of variables. ;in, It is the set of states in a nondeterministic Büchi automaton; This represents the initial state of a nondeterministic Büchi automaton; For character tables; Represents the state transition relations in a nondeterministic Büchi automaton; This indicates the acceptable or terminated state of a nondeterministic Büchi automaton.

3. The task planning method for a multi-robot system according to claim 1, characterized in that, Also includes: Define the tuple corresponding to the robot as follows: ； in, for A multi-group of robots; Represents the set of states of the task environment. This represents the initial state of the agent; This represents the allowed state transition relationships in the task environment. This represents the cost or weight of the corresponding state transition; It is the first A set of atomic propositions for a robot; This represents the label function.

4. The task planning method for a multi-robot system according to claim 1, characterized in that, After obtaining the current sampling point, the following is also included: The Extend function is used to add the current sample point to the search tree and construct a set. According to the set Determine whether to add the current sampling point to the search tree; if the set If empty, the node will not be... Add to the search tree; if the set If it is not empty, then the node will be... Add to the search tree; set Defined as ;in, Includes samples that have already been sampled and added to the search tree structure The set of states; This is the current sampling point.

5. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of a task planning method for a multi-robot system according to any one of claims 1-4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the task planning method for a multi-robot system as described in any one of claims 1-4.

7. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the task planning method for a multi-robot system as described in any one of claims 1-4.

Images