Work planning device, work planning method, and work planning program

Abstract

This work planning device comprises: a generation means for generating a second prompt for instructing a language model to acquire a text loss representing a problem of a first prompt that instructs acquisition of information representing a work plan with respect to problem information representing a problem relating to the work plan; and an update means for updating the first prompt on the basis of an output from the language model with respect to the second prompt.

Description

Work planning apparatus, work planning method and work planning program

[0001] The present disclosure relates to a work planning apparatus, a work planning method and a work planning program.

[0002] As a technique related to calculation of a work plan, Patent Document 1 describes a planning apparatus that generates relative arrangement information of a robot and a work table and an assembly task sequence, and formulates a layout plan.

[0003] Japanese Unexamined Patent Application Publication No. 2022-044127

[0004] However, the technique described in Patent Document 1 is a unidirectional processing flow. Therefore, a mechanism for iteratively improving the accuracy of a calculated plan is not sufficiently provided.

[0005] Accordingly, one object of the present disclosure is to provide a work planning apparatus, a work planning method and a work planning program that can improve the accuracy of a work plan.

[0006] A work planning apparatus according to the present disclosure includes: generation means for generating a second prompt that instructs a language model to acquire text loss representing problems with a first prompt that instructs acquisition of information representing a work plan corresponding to problem information representing a problem about the work plan; and updating means for updating the first prompt based on an output from the language model in response to the second prompt.

[0007] In a work planning method according to the present disclosure, a computer generates a second prompt that instructs a language model to acquire text loss representing problems with a first prompt that instructs acquisition of information representing a work plan corresponding to problem information representing a problem about the work plan, and updates the first prompt based on an output from the language model in response to the second prompt.

[0008] The work planning program provided in this disclosure causes a computer to perform a generation process that generates a second prompt that instructs a language model to obtain text loss representing problems with a first prompt that instructs the computer to obtain information representing a work plan for problem information representing problems with a work plan, and an update process that updates the first prompt based on the output from the language model for the second prompt.

[0009] This disclosure makes it possible to improve the accuracy of work plans.

[0010] This is a block diagram illustrating the functional configuration of a work planning system. This is a flowchart illustrating the operation of a work planning system. This is a diagram illustrating the environment of a work planning problem. This is a diagram illustrating a work planning problem given in PDDL format. This is a diagram illustrating a work planning problem given in JSON format. This is a block diagram illustrating the hardware configuration of a computer. This is a block diagram illustrating the main parts of a work planning system. This is a diagram illustrating an overview of the prompt optimization method described in reference 2.

[0011] To automate tasks such as pick-and-place using robots, a method utilizing Large Language Models (LLMs) has been proposed for calculating work plans. Large Language Models represent artificial intelligence models trained using text data and have the ability to perform tasks such as natural language processing. For example, Large Language Models can use neural networks, particularly transformer architectures, to enable the generation and interpretation of context-dependent text. Large Language Models do not necessarily have to be large; they can simply be language models. Furthermore, Large Language Models may be models trained using image data. Hereafter, for the sake of explanation, language models with the ability to process such language will be referred to as Large Language Models (LLMs).

[0012] For example, Document 1 (Yuchen Liu1, Luigi Palmieri, Sebastian Koch, Ilche ​​Georgievski, and Marco Aiello, "DELTA: Decomposed Efficient Long-Term Robot Task Planning using Large Language Models", [online], April 4, 2020, [Retrieved February 5, 2020], Internet<URL: https: / / arxiv.org / abs / 2404.03275> ) describes a technique for solving robot task planning problems using LLM. In the work planning method described in reference 1, a work plan is calculated for a single agent system using the following steps (1) to (3): (1) A natural language task given in natural language and a 3DSG (3D Scene Graph) representing the environment are input, and the LLM generates a formal language (Planning Domain Definition Language; PDDL). (2) The LLM decomposes the generated PDDL into subtask PDDLs. (3) The planner solves the decomposed subtask PDDLs. (4) The work plan results planned for each subtask are integrated to calculate the overall work plan. A formal language represents, for example, a language constructed based on grammatical rules so that a computer or artificial intelligence system can interpret instructions and data. The work plan result represents the plan of actions calculated for the work planning problem. The work plan result may include, for example, a sequence of actions, the time frame in which each action is performed, the total time required for the work, or an indicator showing how well the work plan satisfies the constraints.

[0013] However, the work planning method described in Reference 1 is a unidirectional processing flow. Therefore, it has the drawback of not having a sufficient mechanism for iteratively improving the accuracy of the calculated work plan.

[0014] Document 2 (Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, and James Zou, "TextGrad: Automatic “Differentiation” via Text ", [online], June 11, 2024, [retrieved February 5, 2025], Internet <URL: https: / / arxiv.org / abs / 2406.07496>) describes a prompt optimization method for LLMs.

[0015] FIG. 8 is a diagram illustrating the outline of the prompt optimization method described in Document 2. This method implements prompt optimization that imitates gradient descent. For example, when it is desired to optimize a system prompt for obtaining an "ideal answer" from a "question", the prompt is improved through the following steps (1) to (3): (1) A loss calculation unit 203 inputs the "question", an "answer", and the "ideal answer", lists problems related to the difference between the "answer" and the "ideal answer", and outputs the result as a text loss. (2) A gradient calculation unit 204 performs feedback analysis and criticism on the listed problems, infers how the system prompt should be improved, and outputs the inference result as a text gradient. (3) A variable update unit 205 updates the variable (the system prompt to be provided to an answer generation unit 202) based on the text gradient.

[0016] However, when the prompt optimization method described in Document 2 is applied to a work planning problem to perform prompt optimization, the loss function that should be defined is unknown. That is, although a "question" (e.g., a work planning problem) and an "answer" (e.g., a work planning result) can be input to the loss calculation unit 203, the "ideal answer" (e.g., a correct work plan) is unknown.

[0017] Thus, the work planning method described in Reference 1 has the problem that it does not have a sufficient mechanism for iteratively improving the accuracy of the calculated work plan. Furthermore, even when attempting to apply the prompt optimization method described in Reference 2 to the work planning method described in Reference 1, difficulties arise because a loss function cannot be defined.

[0018] The work planning device of this disclosure calculates a loss function expressed in text from the results of the work planning. In this case, the loss function is information that represents the relationship between, for example, work information representing the content of the work and text representing information about the loss regarding the prediction results obtained by applying a model to the work information. Next, the work planning device infers how to improve the system prompts given to the LLM that calculates the work plan based on the loss function. Furthermore, the work planning device updates the system prompts based on the inference results and uses the updated system prompts to improve the accuracy of the work plan calculation by the LLM. The system prompts represent strings of characters used to give instructions to the Large-Scale Language Model (LLM) and control its responses and actions.

[0019] In other words, the work planning device of this disclosure makes it possible to update the system prompt while feeding back past work planning results during work plan calculation by LLM. As a result, the accuracy of work planning by LLM can be iteratively improved.

[0020] Furthermore, the work planning device described herein is applicable to calculating work plans in a variety of settings, such as logistics warehouses and manufacturing sites. However, the applications to which the work planning device described herein can be applied are not limited to these.

[0021] Embodiments of this disclosure will be described below with reference to the drawings. In each drawing, the same or corresponding elements are denoted by the same reference numerals, and redundant explanations are omitted as necessary for clarity. Unless otherwise specified, predetermined values ​​such as set values ​​and thresholds are stored in advance in a storage device accessible from the device that uses those values. Unless otherwise specified, the storage unit is composed of one or any number of storage devices.

[0022] Embodiment 1. [Description of Configuration] The work planning device of this embodiment will now be described. Figure 1 is a block diagram illustrating the functional configuration of the work planning device. The work planning device 100 includes an input unit 101, a work planning unit 102, a work planning result database 103, a loss function generation unit 104, a text loss calculation unit 105, a gradient calculation unit 106, a variable update unit 107, and an output unit 108. The text loss calculation unit 105, the gradient calculation unit 106, and the variable update unit 107 each have independent functions, but the configuration encompassing these functions is called the system prompt update unit 110.

[0023] The input unit 101 has the function of inputting a work plan problem and a system prompt (hereinafter also referred to as the work plan system prompt or first prompt PL) to be given to the work plan unit 102 as an initial variable to the work plan device 100. The input unit 101 inputs the work plan problem and the work plan system prompt to the work plan device 100, for example, based on user input operations. The first prompt PL can also be described as a prompt that instructs the acquisition of information representing a work plan for problem information representing a problem about the work plan.

[0024] The work planning problem addressed in this embodiment is, for example, a problem of finding a procedure (for example, a procedure that minimizes work time) to achieve an objective while satisfying constraints. The work planning problem can be expressed in a format such as natural language or PDDL. The work planning system prompt is an instruction for controlling the operation and output style of the LLM included in the work planning unit 102, and is composed of a string of characters.

[0025] The work planning unit 102 has the function of inputting a work planning problem, a work planning system prompt, and sorted past work planning results stored in the work planning result database 103, and inferring a work plan using LLM to calculate a work planning result. It can also be said that the work planning unit 102 has the function of acquiring information representing a work plan using a language model that operates based on a work planning system prompt (first prompt PL) for problem information representing a problem about a work plan.

[0026] The work planning unit 102 stores the calculated work plan results in the work plan result database 103. The work planning unit 102 also outputs the input work plan problem and the calculated work plan results to the loss function generation unit 104. If the input work plan problem is decomposed into subproblems (hereinafter also called subtasks), the work planning unit 102 also outputs information about the decomposed subtasks to the loss function generation unit 104.

[0027] The work planning unit 102 includes a Work Learning Manager (LLM), which operates based on the input work planning system prompts. The work planning unit 102 can utilize the LLM in various ways when calculating the work planning results. An example of how to use the LLM is shown below, but the operation of the work planning unit 102 is not limited to this.

[0028] The work planning unit 102, for example, uses LLM to break down the tasks defined in the work planning problem into subtasks. Then, the work planning unit 102 automatically plans the work planning problem for each subtask. Note that the work planning unit 102 may omit the process of breaking down the input work planning problem into subtasks.

[0029] Furthermore, the work planning unit 102 uses LLM to calculate a sequence of actions from the work planning problem (the input work planning problem, or the work planning problems for each subtask). Then, based on the obtained sequence of actions, the work planning unit 102 uses a simulator to verify the overall work time and the degree to which constraints are satisfied.

[0030] Furthermore, when performing inference using LLM, the work planning unit 102 can use past work planning results as needed. By utilizing past work planning results, the work planning unit 102 can further improve the accuracy of the work planning results.

[0031] The work plan results calculated by the work planning unit 102 include, for example, a list of actions, work time, and the degree of constraint satisfaction. Here, "action list" refers to a list of actions or operations necessary to accomplish a specific task, arranged in chronological order. "Work time" is the time required to complete the task. "Degree of constraint satisfaction" is an indicator showing the extent to which the constraints defined in the work plan problem are met.

[0032] The work plan result database 103 stores sorted past work plan results. Here, past work plan results are sorted in order from the best samples in terms of work time and constraint satisfaction, for example. The work planning unit 102 can perform inference efficiently and effectively by referring to these sorted past work plan results. Examples of sorting methods (1) to (3) are shown below, but the sorting method is not limited to these.

[0033] (1) In the case of a maximization (or minimization) problem with constraints: "Constraints are satisfied and the objective function value is high (or low)" > "Constraints are satisfied but the objective function value is low (or high)" > "Constraints are not satisfied"

[0034] (2) In the case of a maximization (or minimization) problem without constraints: "High (or low) objective function value" > "Low (or high) objective function value"

[0035] (3) In the case of a maximization (or minimization) problem where the degree of satisfaction of the constraints can be quantified: "Constraints are satisfied and the objective function value is high (or low)" > "Constraints are satisfied but the objective function value is low (or high)" > "Constraints are not satisfied but the degree of satisfaction is high" > "Constraints are not satisfied and the degree of satisfaction is low"

[0036] For example, consider a work planning problem where "multiple orders must be completed by a certain time, and the work time must be minimized." In this case, the orders may be sorted according to rules such as the following: "All orders meet the time constraint and have short work times" > "All orders meet the time constraint but have long work times" > "Some orders exceed the time constraint" > "Some orders significantly exceed the time constraint." For example, consider a work planning problem where "multiple orders must be completed by a certain time, and the work time must be minimized." In this case, the orders may be sorted according to rules such as the following: "All orders meet the time constraint and have short work times" > "All orders meet the time constraint but have long work times" > "Some orders exceed the time constraint, and the total excess time of the orders that exceed the time constraint is small" > "Some orders exceed the time constraint, and the total excess time of the orders that exceed the time constraint is large." However, "A > B" represents a list where A is placed before B.

[0037] The sorting process of past work plan results is performed, for example, by the work planning unit 102. In this case, the work planning unit 102 sorts the calculated work plan results in a predetermined order when storing them in the work plan result database 103. Furthermore, this process does not necessarily have to be performed by the work planning unit 102. For example, a sorting unit (not shown in Figure 1) dedicated to sorting may be provided, and the sorting unit may appropriately perform the sorting process of the work plan result database 103.

[0038] The loss function generation unit 104 has the function of inputting the work plan problem and the work plan result calculated by the work plan unit 102, and generating a system prompt for calculating text loss (hereinafter also referred to as the text loss calculation system prompt or second prompt LO). The second prompt LO is a prompt that instructs the language model to acquire the text loss representing the problem for the first prompt PL, which instructs to acquire information representing the work plan for the problem information representing the problem about the work plan. Therefore, it can also be said that the loss function generation unit 104 has the function of generating a second prompt LO that instructs the language model to acquire the text loss representing the problem for the first prompt PL, which instructs to acquire information representing the work plan for the problem information representing the problem about the work plan. In addition, if the work plan unit 102 decomposes the work plan problem into subtasks, the loss function generation unit 104 also inputs information about each subtask.

[0039] The loss function generation unit 104 is responsible for generating text loss calculation system prompts used by the text loss calculation unit 105. In this embodiment, the text loss is calculated by the LLM included in the text loss calculation unit 105. Therefore, the loss function generation unit 104 does not generate the text loss function itself, but rather generates text loss calculation system prompts to be given to the text loss calculation unit 105. The text loss calculation system prompts instruct the text loss calculation unit 105 on what should be extracted as loss.

[0040] The loss function generation unit 104 has the function of systematically generating text loss calculation system prompts. The loss function generation unit 104 may also generate text loss calculation system prompts using LLM.

[0041] The text loss calculation system prompt (second prompt LO) includes, for example, the following elements (1) to (5). Note that if the work planning unit 102 does not decompose the work planning problem into subtasks, i.e., does not perform task decomposition, element (2) can be omitted. (1) Work planning problem: What kind of work planning problem is given? (2) Task decomposition result: How the work planning problem was decomposed and assigned when task decomposition was performed? (3) Work planning result (action sequence): What kind of work plan was calculated? (4) Work planning result (work time and degree of constraint satisfaction): Total work time, degree of satisfaction if there are time constraints (5) Instructions: Instructions (instruction statements) that list what problems exist in the current work plan (i.e., the most recently calculated work plan) in order to minimize the total work time while satisfying the constraints.

[0042] The following is a specific example of a text loss calculation system prompt. In the example below, the loss function generation unit 104 systematically generates the text loss calculation system prompt by substituting the elements (work plan problem), (task decomposition result), and (work plan result) into a predetermined format. However, the format of the text loss calculation system prompt is not limited to the example below. Text loss calculation system prompt: "In (work plan problem), the task was decomposed into (task decomposition result) and the work plan was calculated, resulting in (work plan result). To shorten the work time while satisfying the time constraints of the order, please list the problematic areas."

[0043] In the above text loss calculation system prompt, (Work Plan Problem) is the work plan problem entered by the input unit 101 and corresponds to element (1). (Task Decomposition Result) is the result of the work plan unit 102 decomposing the work plan problem into subtasks and corresponds to element (2). (Work Plan Result) is the work plan result calculated by the work plan unit 102 and corresponds to elements (3) and (4). The entire prompt corresponds to element (5) and is an instruction that tells the system to list the problems.

[0044] The system prompt updating unit 110 has a function of receiving the text loss calculation system prompt generated by the loss function generation unit 104 and outputting an updated work plan system prompt. It can also be said that the system prompt updating unit 110 has a function of updating the work plan system prompt (first prompt PL) based on an output from a language model for the text loss calculation system prompt (second prompt LO). The updating may be a process of generating a part of the prompt, or may be a process of generating the entire prompt. The updating process of the system prompt updating unit 110 is executed by the text loss calculation unit 105, the gradient calculation unit 106, and the variable updating unit 107.

[0045] The text loss calculation unit 105 has a function of inferring text loss based on the text loss calculation system prompt. The text loss calculation unit 105 enumerates problems of the work plan system prompt when optimizing a work plan through inference by an LLM. In the present embodiment, a set of enumerated problems of the work plan system prompt corresponds to text loss.

[0046] Inference settings of the LLM in the text loss calculation unit 105 are, for example, as follows. ・Role: Text loss calculation system prompt ・Instruction: Work plan system prompt ・Response: Text loss (a set of problems of the work plan system prompt based on work plan results for work plan problems and task decomposition)

[0047] The gradient calculation unit 106 has a function of inferring a text gradient based on text loss. The gradient calculation unit 106 analyzes text loss and performs feedback inference on how to improve the work plan system prompt through inference by an LLM. It can also be said that the gradient calculation unit 106 has a function of inferring parameters of the work plan system prompt (first prompt PL) based on text loss. In the present embodiment, text indicating how to improve the work plan system prompt corresponds to the text gradient.

[0048] The inference settings for the LLM in the gradient calculation unit 106 are, for example, as follows. ・Role: "You are part of an optimization system that improves given text (variables). Based on the text loss, please provide feedback on how the variables should be improved." ・Instruction: Text loss ・Response: Text gradient (text indicating how the work planning system prompt should be improved)

[0049] The variable updating unit 107 has a function of updating the work planning system prompt based on the text gradient. The variable updating unit 107 analyzes the text gradient through LLM inference and updates the work planning system prompt.

[0050] The inference settings for the LLM in the variable updating unit 107 are, for example, as follows. ・Role: "You are part of an optimization system that improves given text (variables). Based on the text gradient, please revise the work planning system prompt." ・Instruction: Text gradient ・Response: Updated work planning system prompt

[0051] The output unit 108 has a function of outputting the work planning result calculated by the work planning unit 102. For example, the output unit 108 can output and store the work planning result in a storage unit (not shown) of the work planning apparatus 100 or an external apparatus. Furthermore, the output unit 108 can output and display the work planning result on a display device such as a display apparatus (not shown). Note that the output unit 108 may be configured to be capable of outputting the updated work planning system prompt in the same manner as it outputs the work planning result.

[0052] Furthermore, the output unit 108 can output the work planning result to a robot or the like that executes work (for example, a robot or an apparatus that controls the robot). This enables the robot to operate based on the work planning result. That is, the work planning apparatus 100 can function as a control apparatus for a controlled object (for example, a robot), and can control the controlled object based on the work planning result calculated from the work planning problem.

[0053] [Explanation of Operation] Next, the operation of the work planning device 100 will be explained. Figure 2 is a flowchart illustrating the operation of the work planning device.

[0054] The input unit 101 receives the work planning problem and the work planning system prompt, which serves as the initial variable. The work planning system prompt here is, for example, an instruction that instructs the system to decompose tasks in a way that minimizes the work planning time while satisfying constraints.

[0055] The work planning unit 102 receives the work planning problem, the work planning system prompt, and sorted past work planning results stored in the work planning result database 103. Then, the work planning unit 102 uses an LLM that operates based on the work planning system prompt to decompose the work planning problem into subtasks (step S1). Hereinafter, the set of decomposed subtasks will also be called the subtask set.

[0056] Next, the work planning unit 102 starts processing to calculate the action sequence for each of the subtasks included in the subtask set (step S2).

[0057] First, the work planning unit 102 creates an automated planning problem in PDDL format to complete one of the subtasks (step S3). Then, the work planning unit 102 solves the automated planning problem using a PDDL planner and calculates the sequence of actions (step S4).

[0058] The work planning unit 102 repeats steps S3 and S4 for calculating the action sequence for all subtasks (step S5). After the action sequences for all subtasks have been calculated, the process moves on to step S6.

[0059] Once the action sequences for all subtasks have been calculated, the work planning unit 102 integrates the calculated action sequences and verifies the total work time required for the task and the degree to which the constraints are satisfied (step S6).

[0060] The work planning unit 102 calculates a work planning result for a work planning problem through the processing in steps S1 to S6. This work planning result includes the work planning problem, the set of subtasks decomposed in step S1, the sequence of actions calculated in steps S2 to S5, and the work time and degree of satisfaction of constraints calculated in step S6.

[0061] Next, the work planning unit 102 determines whether or not predetermined termination conditions are met (step S7). Here, predetermined termination conditions include, for example, some or all of the following conditions: - When a loop process is executed to update the work planning system prompt and calculate the work planning result until a certain number of iterations is reached. - When the calculated work time is less than or equal to a certain threshold. - When a work planning result is calculated that satisfies all constraint conditions.

[0062] If the predetermined termination conditions are met (step S7:Y), the work planning device 100 terminates its processing. On the other hand, if the termination conditions are not met (step S7:N), the work planning unit 102 stores the work planning result, including the work time and the degree of satisfaction of the constraints calculated in step S6, in the work planning result database 103 (step S8).

[0063] Next, the work planning unit 102 or sorting unit (not shown) sorts the work storage results stored in the work planning result database 103 based on the work time and the degree to which constraints are satisfied (step S9).

[0064] Next, the loss function generation unit 104 receives the work plan problem and the work plan result calculated by the work plan unit 102 as input and generates a text loss calculation system prompt to instruct what should be extracted as loss (step S10).

[0065] Next, the text loss calculation unit 105 infers the text loss using LLM based on the text loss calculation system prompt (step S11). The text loss is represented as a string listing the problems in the work planning system prompt.

[0066] Next, the gradient calculation unit 106 infers the text gradient using LLM based on the text loss (step S12). The text gradient is represented as a string of characters indicating points for improving the work planning system prompt.

[0067] Next, the variable update unit 107 updates the work planning system prompt using LLM based on the text gradient (step S13). Then, the variable update unit 107 inputs the new work planning system prompt to the work planning unit 102. After that, the process proceeds to step S1.

[0068] As shown in Figure 2, the work planning device 100 repeatedly performs steps S1 to S6 for calculating a work plan result from a work planning problem, and steps S8 to S13 for updating the work planning system prompt, until a predetermined termination condition is met. The work planning system prompt represents a string of characters used to instruct the model of specific work instructions or conditions when calculating a work plan using a large-scale language model (LLM).

[0069] The example of operation shown in Figure 2 does not limit the operation of the work planning device 100 of this disclosure. For example, the work planning device 100 may omit the processing of step S1, which decomposes the input work planning problem into subtasks, and be configured to directly calculate the action sequence from the work planning problem. Also, the work planning device 100 can perform the processing of step S9 at any time before next referring to the work planning result database 103.

[0070] Furthermore, after a predetermined termination condition is met in step S7, the output unit 108 may execute a process to output the work plan result. In this case, the output unit 108 may output the work plan result to a storage unit (not shown) for storage, or to a display device (not shown) such as a display device for display. Alternatively, the output unit 108 may output the work plan result to a robot or the like that will perform the work (for example, a robot or a device that controls the robot). This will enable the robot to operate based on the work plan result.

[0071] [Description of Effects] Next, the effects of this embodiment will be described. In this embodiment, the work planning unit 102 calculates a work planning result (i.e., information representing the work plan) from a work planning problem (i.e., problem information representing a problem about the work plan) using a large-scale language model (LLM) that operates based on a work planning system prompt (first prompt PL). The loss function generation unit 104 generates a text loss calculation system prompt (i.e., a second prompt LO that instructs the large-scale language model (LLM) to obtain a text loss representing a problem about the first prompt PL) based on the work planning problem and the work planning result. The system prompt update unit 110, which includes a text loss calculation unit 105, a gradient calculation unit 106, and a variable update unit 107, updates the work planning system prompt (first prompt PL) based on the output from the large-scale language model (LLM) for the text loss calculation system prompt (second prompt LO).

[0072] Furthermore, the work planning device 100 generates a text loss calculation system prompt containing the following elements from the loss function generation unit 104, and provides this to the LLM included in the text loss calculation unit 105 to calculate a text-based loss function. (1) Work planning problem: What kind of work planning problem is given (i.e., problem information representing the problem about the work plan) (2) Task decomposition result: How the work planning problem was decomposed and assigned when the task was decomposed (i.e., information about subtasks) (3) Work planning result (action sequence): What kind of work plan was calculated (i.e., information representing the acquired work plan) (4) Work planning result (work time and degree of constraint satisfaction): Total work time, degree of satisfaction if there are time constraints (5) Instructions: Instructions that list the problems with the current work plan (i.e., information representing the most recently calculated work plan) in order to minimize the total work time while satisfying the constraints

[0073] In the system prompt update unit 110, the text loss calculation unit 105 infers the text loss based on the text loss calculation system prompt. Next, the gradient calculation unit 106 infers how to improve the work plan system prompt (i.e., the parameters of the work plan system prompt (first prompt PL)) based on the text loss. Then, the variable update unit 107 updates the work plan system prompt based on the gradient inferred by the gradient calculation unit 106.

[0074] In other words, the work planning device 100 does not simply feed back the results of past work plans and instruct the LLM to improve the accuracy of the work plan, but rather analyzes the problems based on the results of past work plans and incorporates the results of its inferences on how to improve them into the work planning system prompt. With this configuration, the work planning device 100 can iteratively and efficiently improve the accuracy of the work plan.

[0075] The following describes a specific example of the operation of the work planning device realized in the above embodiment.

[0076] Example 1. Figure 3 illustrates an example of the environment for a work planning problem. Figure 3 visually shows an example of a pick-and-place problem in a logistics warehouse.

[0077] In this embodiment, the input unit 101 receives a work planning problem that includes the following elements: • Two robots, Agent 1 and Agent 2, perform move, pick, and place actions (PDDL description). • An order is given (transport items to the shipping shelf within a specified time). • Order 1: Transport Obj1 and Obj2 to Shipping_shelf within 10 minutes. • Order 2: Transport Obj3 and Obj4 to Shipping_shelf within 20 minutes.

[0078] In this embodiment, the work planning problem is given in PDDL and JSON format. Figure 4 is an example of a work planning problem given in PDDL format. Figure 5 is an example of a work planning problem given in JSON format.

[0079] The PDDL consists of a domain file shown in Figure 4(A) and a problem file shown in Figure 4(B). The PDDL shown in Figures 4(A) and (B) provides predicate definitions representing relationships between objects, definitions of actions that the robot can use, definitions of objects, the initial state of objects, and the goals that must ultimately be met. Figure 5 also shows an example of orders 1 and 2, which indicate transporting items to the shipping shelves within a specified time, written in JSON format.

[0080] In this embodiment, the input unit 101 also inputs a work planning system prompt as an initial variable. This work planning system prompt is defined as follows: Work planning system prompt: "Please decompose the task in a way that reduces the work time while satisfying the time constraints of the order."

[0081] The work planning unit 102 receives the work planning problem and the work planning system prompt. Then, using the LLM that operates based on the work planning system prompt, the work planning unit 102 breaks down the work planning problem into the following subtask 1 and subtask 2: Subtask 1: Transport Obj1 and Obj2 to Shipping_shelf -> Assign to Agent 1 Subtask 2: Transport Obj3 and Obj4 to Shipping_shelf -> Assign to Agent 2

[0082] The work planning unit 102 generates an automated planning problem in PDDL format from the subtasks. Next, the work planning unit 102 solves the automated planning problem using a PDDL planner and calculates the action sequence. After that, the work planning unit 102 integrates each calculated action sequence and verifies the overall work time and the degree to which the constraints are satisfied. Through this process, the work planning unit 102 calculates a work plan result that includes the following elements: Work plan result: ・Action sequence: (move Agent1 D)[0.2min] -> (move Agent1 E)[0.3min] -> … -> (place Agent2 obj4)[0.9min] ・Work time: 26min ・Degree to which constraints are satisfied: Order 1 is completed with a 1min buffer, Order 2 is completed with a 6min overtime.

[0083] The loss function generation unit 104 receives the work planning problem and the work planning result calculated by the work planning unit 102 as input and generates the following text loss calculation system prompt: Text loss calculation system prompt: "In (work planning problem), the task was broken down into (subtasks 1 and 2) and a work plan was calculated, resulting in (work planning result). To shorten the work time while satisfying the time constraints of the order, please list the problematic areas." The loss function generation unit 104 systematically generates the above text loss calculation system prompt by substituting values ​​into each element of a predefined format.

[0084] The text loss calculation unit 105 infers text loss using LLM based on the text loss calculation system prompt. Here, text loss is represented as a string listing the problems in the work planning system prompt. The inference settings for this LLM are, for example, as follows: • Role: Text loss calculation system prompt • Instruction: Work planning system prompt as initial variable (updated work planning system prompt from the second week onwards) • Response: 1. Imbalance in agent load balancing: While Agent 1 was able to complete task order 1 with ample margin, … 2. Subtask order and timing: The current task order may not be efficient, …

[0085] The gradient calculation unit 106 infers the text gradient using LLM based on the text loss. Here, the text gradient is represented as a string of characters indicating points for improving the work planning system prompt. The inference settings for this LLM are, for example, as follows: • Role: "You are part of an optimization system that improves the given text (variable). Please provide feedback on how to improve the variable." • Instruction: Text loss • Response: "The current work planning system prompt has abstract instructions and does not include specific improvement measures. Please provide detailed…"

[0086] The variable update unit 107 updates the work planning system prompt using LLM based on the text gradient. The inference settings for this LLM are as follows, for example: • Role: "Part of an optimization system that improves the given text (variable). Modify the work planning system prompt based on the text gradient." • Instruction: Text gradient • Response: Updated work planning system prompt "Perform a task decomposition that satisfies the time constraints of the given order, evenly distributes the workload among agents, and promotes cooperation and efficiency. Aim to optimize the order and timing of subtask transitions, make effective use of available time, and introduce parallel work as much as possible..."

[0087] The variable update unit 107 inputs the updated work plan system prompt to the work plan unit 102. The work plan unit 102 then calculates the work plan result using LLM based on the updated work plan system prompt.

[0088] The work planning device 100 repeatedly performs the process of calculating a work plan result from a work planning problem and the process of updating the work planning system prompt until a predetermined termination condition is met. After the predetermined termination condition is met, the output unit 108 outputs the work plan result. At this time, the output unit 108 may output the work plan result to a storage unit (not shown) for storage, or to a display device (not shown) such as a display device for display. The output unit 108 may also output the work plan result to a robot or the like that will perform the work (for example, a robot or a device that controls the robot). As a result, the robot (i.e., Agent 1 and Agent 2) can operate based on the work plan result.

[0089] As described above, the work planning device 100 of this embodiment can iteratively and efficiently improve the accuracy of the work plan. Therefore, in the above embodiment, as a result of improving the accuracy of the work plan, it becomes possible to operate the robots (i.e., Agent 1 and Agent 2) based on a suitable work plan result.

[0090] Each component in the above embodiments and examples can be configured with one piece of hardware, but can also be configured with one piece of software. Furthermore, each component can be configured with multiple pieces of hardware, or with multiple pieces of software. In addition, some of each component can be configured with hardware, and other parts with software.

[0091] Each function (each process) in the above embodiment can be implemented by a computer having a processor, memory, etc. For example, a program for implementing the method (process) in the above embodiment may be stored in a storage device (storage medium), and each function may be implemented by executing the program stored in the storage device with a processor.

[0092] Figure 6 is a block diagram illustrating the hardware configuration of computer 1000. Computer 1000 is any computer. For example, computer 1000 is a stationary computer such as a personal computer or a server machine. Alternatively, computer 1000 is a portable computer such as a smartphone or a tablet terminal. Computer 1000 may be a dedicated computer designed to implement the work planning device 100, or it may be a general-purpose computer.

[0093] The computer 1000 includes a processor 1001, a storage device 1002, memory 1003, a bus 1004, an input / output interface 1005, and a network interface 1006.

[0094] The processor 1001 is a variety of processing unit, such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), FPGA (Field-Programmable Gate Array), or DSP (Digital Signal Processor).

[0095] The storage device 1002 is, for example, a non-transitory computer-readable medium. Non-transitory computer-readable media include various types of tangible storage media. Specific examples of non-transitory computer-readable media include semiconductor memory (e.g., mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM).

[0096] Memory 1003 is a main memory system implemented using RAM (Random Access Memory) or the like. Memory 1003 temporarily stores data when the processor 1001 executes processing.

[0097] Bus 1004 is a data transmission path for the processor 1001, memory 1003, storage device 1002, input / output interface 1005, and network interface 1006 to send and receive data to and from each other. However, the method of connecting the processor 1001 and the others to each other is not limited to bus connection.

[0098] The input / output interface 1005 is an interface for connecting the computer 1000 with input / output devices. For example, input devices such as a keyboard and output devices such as a display device are connected to the input / output interface 1005.

[0099] The network interface 1006 is an interface for connecting the computer 1000 to a network. This network may be a LAN (Local Area Network) or a WAN (Wide Area Network).

[0100] The storage device 1002 stores a program that implements each of the functional components in the above-described embodiments and examples. The processor 1001 reads this program into the memory 1003 and executes it to implement each of the functional components in the above-described embodiments and examples.

[0101] The work planning device 100 may be implemented by one computer 1000 or by multiple computers 1000. In the latter case, the configuration of each computer 1000 does not need to be the same and can be different.

[0102] Each functional component in the above embodiments and examples may be implemented by a combination of the hardware and software described above, or by hardware (for example, a hardwired electronic circuit).

[0103] Next, an overview of the present disclosure will be described. Figure 7 is a block diagram illustrating the main components of a work planning device. The work planning device 10 shown in Figure 7 (corresponding to, for example, a work planning device 100) includes a generation means 11 (implemented by a loss function generation unit 104 in an embodiment) that generates a first prompt (corresponding to, for example, a work planning system prompt, or first prompt PL) that instructs the language model to acquire text loss representing problems with a first prompt (corresponding to, for example, a text loss calculation system prompt, or second prompt LO loss function generation unit 104 in an embodiment) that updates the first prompt based on the output from the language model to the second prompt (corresponding to, in an embodiment, a system prompt update unit 110 including a text loss calculation unit 105, a gradient calculation unit 106, and a variable update unit 107). With this configuration, the work planning device 10 can iteratively and efficiently improve the accuracy of the work plan.

[0104] Although the present disclosure has been described above with reference to embodiments, the present disclosure is not limited to the embodiments described above. Various modifications to the structure and details of the present disclosure are possible, as can be understood by those skilled in the art within the scope of the present disclosure. Each embodiment can be combined with other embodiments as appropriate.

[0105] Each drawing is merely illustrative to illustrate one or more embodiments. Each drawing may be associated with one or more other embodiments, rather than being associated with only one specific embodiment. As those skilled in the art will understand, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings, for example, to create embodiments not explicitly shown or described. Not all features or steps shown in any one drawing to illustrate an exemplary embodiment are necessarily required, and some features or steps may be omitted. The order of steps described in any of the drawings may be changed as appropriate.

[0106] Some or all of the above embodiments may also be described as follows, but are not limited to the following:

[0107] (Note 1) A work planning device comprising: a generation means that generates a second prompt that instructs a language model to acquire text loss representing problems with respect to a first prompt that instructs the acquisition of information representing a work plan for problem information representing problems with a work plan; and an update means that updates the first prompt based on the output from the language model for the second prompt.

[0108] (Note 2) The work planning apparatus according to Note 1, wherein the update means includes a loss calculation means for inferring the text loss based on the second prompt, a gradient calculation means for inferring the parameters of the first prompt based on the text loss, and a variable update means for updating the first prompt based on the gradient which is the inference result of the gradient calculation means.

[0109] (Note 3) The work planning device according to Note 1 or Note 2, wherein the second prompt includes an element indicating problem information representing problems with the work plan, an element indicating information representing the acquired work plan, an element indicating the degree to which the work time and constraints are satisfied, and an element indicating that problems with the information representing the acquired work plan should be listed in order to minimize the overall work time while satisfying the constraints.

[0110] (Note 4) The work planning apparatus according to Note 3, wherein the second prompt includes an element indicating how the problem about the work plan was broken down when acquiring information representing the work plan.

[0111] (Note 5) A work planning device according to any one of Notes 1 to 4, comprising a work planning means that acquires information representing a work plan using a language model that operates based on the first prompt, in response to problem information representing a problem about a work plan, wherein the work planning means uses previously acquired information representing a work plan when acquiring information representing a work plan.

[0112] (Note 6) The work planning means is a work planning device as described in Note 5, which uses information representing previously acquired work plans sorted based on the degree to which work time and constraints are satisfied.

[0113] (Note 7) The work planning means is a work planning device according to Note 5 or Note 6, which decomposes a work planning problem into sub-problems and acquires information representing the work plan for each of the decomposed sub-problems.

[0114] (Note 8) The work planning device according to any one of Notes 5 to 7, wherein the work planning means acquires information representing a work plan using a language model that operates based on the first prompt updated by the update means, with respect to problem information representing a problem about the work plan.

[0115] (Note 9) A work planning method characterized in that the computer generates a second prompt which instructs a language model to obtain text loss representing problems for a first prompt which instructs the computer to obtain information representing a work plan for problem information representing problems for a work plan, and updates the first prompt based on the output from the language model for the second prompt.

[0116] (Note 10) A work plan program for causing a computer to perform a generation process that generates a second prompt that instructs a language model to obtain text loss representing problems with a first prompt that instructs the computer to obtain information representing a work plan for problem information representing problems with a work plan, and an update process that updates the first prompt based on the output from the language model for the second prompt.

[0117] Some or all of the elements (e.g., configuration and function) described in Appendices 2 to 8 that are dependent on Appendice 1 may also be dependent on Appendices 9 and 10 in the same way as those described in Appendices 2 to 8. Some or all of the elements described in any appendice may be applied to various hardware, software, recording means, systems, and methods for recording software.

[0118] 10,100 Work planning device 11 Generation means 12 Update means 101 Input unit 102 Work planning unit 103 Work planning result database 104 Loss function generation unit 105 Text loss calculation unit 106 Gradient calculation unit 107 Variable update unit 108 Output unit 110 System prompt update unit 1000 Computer 1001 Processor 1002 Storage device 1003 Memory 1004 Bus 1005 Input / output interface 1006 Network interface

Claims

1. A work planning device comprising: a generation means that generates a second prompt that instructs a language model to acquire text loss representing problems with a first prompt that instructs the language model to acquire information representing a work plan for problem information representing problems with a work plan; and an update means that updates the first prompt based on the output from the language model for the second prompt.

2. The work planning apparatus according to claim 1, wherein the update means includes a loss calculation means for inferring the text loss based on the second prompt; a gradient calculation means for inferring the parameters of the first prompt based on the text loss; and a variable update means for updating the first prompt based on the gradient which is the inference result of the gradient calculation means.

3. The work planning device according to claim 1 or 2, wherein the second prompt includes an element indicating problem information representing a problem with the work plan, an element indicating information representing the acquired work plan, an element indicating the degree to which the work time and constraints are satisfied, and an element instructing the device to list what problems exist with the information representing the acquired work plan in order to minimize the overall work time while satisfying the constraints.

4. The work planning apparatus according to claim 3, wherein the second prompt includes an element indicating how the problem about the work plan was broken down when acquiring information representing the work plan.

5. A work planning device according to claim 1 or 2, comprising a work planning means that acquires information representing a work plan using a language model that operates based on the first prompt, in response to problem information representing a problem about a work plan, wherein the work planning means uses previously acquired information representing a work plan when acquiring information representing a work plan.

6. The work planning device according to claim 5, wherein the work planning means uses information representing previously acquired work plans sorted based on the degree to which work time and constraints are satisfied.

7. The work planning device according to claim 5, wherein the work planning means decomposes a problem concerning work planning into sub-problems and acquires information representing a work plan for each of the decomposed sub-problems.

8. The work planning device according to claim 5, wherein the work planning means acquires information representing a work plan using a language model that operates based on a first prompt updated by the update means, with respect to problem information representing a problem about the work plan.

9. A work planning method characterized in that a computer generates a second prompt that instructs a language model to obtain text loss representing problems with a first prompt that instructs the computer to obtain information representing a work plan for problem information representing problems with a work plan, and updates the first prompt based on the output from the language model for the second prompt.

10. A work plan program that causes a computer to perform a generation process that generates a second prompt that instructs a language model to obtain text loss representing problems with a first prompt that instructs the computer to obtain information representing a work plan for problem information representing problems with a work plan; and an update process that updates the first prompt based on the output from the language model for the second prompt.

Images