An intelligent agent that supports people helping each other.

Abstract

This invention provides a computer-implemented method and system for controlling intelligent agents to assist humans. [Solution] The method includes: acquiring information about multiple persons via at least one sensor of an intelligent agent; evaluating the state of each person other than the first person if at least one first person is having difficulty performing a task; determining each person's support capacity; determining the probability of support; selecting at least one second person from the multiple persons to optimize at least one of the support capacity and support probability; planning the intelligent agent's actions to optimize at least one of the support capacity and support probability of the at least one second person to support at least one first person in performing a task; and, based on the planned actions, the intelligent agent performing at least one of communication or physical action.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent agents and action plans for intelligent agents. In particular, a computer-implemented method and system for controlling an intelligent agent to assist a person are proposed.

Background Art

[0002] In a work environment, for example, on an assembly line in a factory, a team including multiple workers performs tasks, such as assembling parts of a vehicle body, in parallel with each other. Individual members of the team may encounter problems when performing their tasks, which may be difficult to overcome within the limited time available to complete the work. The difficulties in advancing the task may stem from the need for skills or knowledge to identify the correct actions to take to achieve the task, or simply the lack of useful or necessary tools to continue. Robots collaborating with the team may provide some support, but collaborative robots have a set of specialized skills and the required skills themselves may be lacking.

[0003] Similarly, in public spaces such as shopping malls or railway stations with multiple floors, the lack of escalators or elevators can pose significant problems when individuals with large bags, luggage, or strollers try to move between different floor levels. Sometimes, these problems may not be solvable for an individual, for example, due to physiological or physical limitations. An object may be too heavy for one person to lift, or a person with limited vision may not be able to read the signs guiding them towards the elevator.

[0004] Generally, people are prosocial and interested in helping others achieve their goals. However, sometimes others fail to notice that other people need help or that they possess the skills required to assist individuals struggling with their tasks. On the other hand, those who need support may not recognize their need for support, fail to notice others who are capable of helping, or do not approach others to ask for help, or a combination of these reasons. While there are robots that approach each person in need to detect the need for help and provide support, for example by providing navigation guidance, the specialized capabilities and knowledge bases of these robots limit their effectiveness in providing support across a wide range of tasks. [Overview of the project] [Problems that the invention aims to solve]

[0005] Therefore, even when intelligent agents themselves lack the ability to provide support, or when intelligent agents do not currently exist, it is necessary to identify the need for support for humans and provide that support. [Means for solving the problem]

[0006] The computer-implemented method described in independent claim 1 provides a favorable solution to the problem.

[0007] The dependent claims provide further advantageous embodiments.

[0008] A computer-implemented method for controlling an intelligent agent operating in an environment with multiple people includes acquiring information about the multiple people through at least one sensor of the intelligent agent. The processor determines whether at least one of the multiple people, a first person, is struggling to perform a task determined by evaluating the acquired information. If at least one first person is struggling to perform the task, the processor evaluates the state of each of the other people with respect to acting to support the at least one first person in performing the task. Evaluating the state of each other includes determining each person's ability to support the at least one first person and determining each person's probability of supporting the at least one first person. The processor then selects at least one second person from the multiple who optimizes at least one of the support ability and support probability, and plans the intelligent agent's actions to propose to the at least one second person in order to support the at least one first person in performing the task. The proposed actions optimize at least one of the support ability and support probability. Based on a planned action, the intelligent agent performs at least one of either a communication action or a physical action. The planned action may include at least one of a communication action and an action that involves performing several physical actions.

[0009] The proposed method includes a framework for detecting a first person who is struggling to perform a task. Struggling with the task includes making efforts to achieve a goal, but despite those efforts, only making limited progress toward achieving the goal. The goal is to successfully complete the task. An intelligent agent recognizes people performing different tasks and detects whether the person is struggling to make progress, or whether performing the task has negative effects on the person, and whether at least one of these effects could be mitigated by others, particularly others present in the environment. Thus, the method not only detects the struggling person but also uses this information in the process of identifying the task the first person is facing, whether and how they can be helped, and identifying supporters in the environment who are capable of and ready to support the first person.

[0010] The intelligent agent analyzes other people in the environment, or specific areas of the environment, and evaluates their state and their current activities in relation to the probability of acting as a supporter for the person in need. The intelligent agent calculates the probability that a potential supporter can provide input that reduces or avoids negative implications for the work of the person in need. Thus, the method can only approach those other people who have the potential to support the first person, for example, by their physical build. Other people, for example, those who apparently lack the physical build to carry heavy loads, will not be bothered or burdened by the intelligent agent requesting them to help the first person. Furthermore, the intelligent agent evaluates the probability that other people in the environment will accept a request for support based on their current situation and to what extent supporting the person in need would affect the supporter's current work.

[0011] The artificial agent plans actions, particularly communications, that maximize the opportunity for at least one potential supporter to be notified of the need for support, to be willing to provide the necessary support, to find the person in need of support, and to take supportive action. Therefore, the planned actions may include, for example, visual communication in a noisy environment, which increases the probability that a selected second person will understand the first person's need for support.

[0012] Comparing the computer-implemented method with existing methods for detecting individuals struggling with tasks using machine learning techniques on camera images, the proposed method significantly goes beyond simply tracking individual performance or progress, for example, by notifying designated instructors for support if an individual appears to be struggling. Unlike the proposed method, existing methods fail to provide a framework for considering the potential to help others.

[0013] The proposed method includes a framework for predicting opportunities for individuals to help others who are struggling with a task, whereas existing methods detect individuals who currently have no work or belong to a specific group with suitable attributes, and then predict opportunities for these individuals or groups of individuals to help others.

[0014] An intelligent agent (assistant) has the ability to communicate information to a target person or potential supporter (helper) in a social and public environment involving multiple people, using its cognitive abilities regarding its environment.

[0015] The step of determining at least one first person in a computer-implemented method may include using a trained model to evaluate the acquired information. The trained model is trained to classify a person's current behavior in relation to a task.

[0016] In embodiments, a computer-implemented method includes using a work model for evaluating acquired information by, in the step of a processor determining at least one first person, comparing at least one of the first person's current skills or knowledge with at least one corresponding of the first person's required skills or knowledge for performing the work. Based on the comparison, the method determines whether the first person's current skills or knowledge deviate from the first person's required skills or knowledge.

[0017] In embodiments, a computer-implemented method includes determining a first person, at least one of several, by using a trained model to evaluate acquired information. The trained model is trained to classify at least one of the following as a struggling behavior of at least one first person with respect to a task: human hesitation, a predetermined bodily expression, a predetermined facial expression, repetition of the same detected behavior, deceleration of detected activity, detected staggering behavior, or reclining posture.

[0018] The computer-implemented method of the embodiment includes determining a first person among several persons by detecting at least one of a work failure, a process error, repeated occurrences of work failures, and repeated occurrences of process errors, based on the acquired information.

[0019] A work failure is a lack of success in performing a task, or a lack of success in achieving the task's objectives. A process error is a failure in the sequence of steps for performing a task. The types of failures and errors discussed may be detected when monitoring the situation in the environment, providing a suitable starting point for determining the first person struggling with the task, determining the need for support, and indicating the quality of assistance required.

[0020] A computer-implemented method according to an embodiment includes a processor determining, based on a work model using a planning algorithm, the requirements for other persons to support at least one first person when performing a task.

[0021] Determining the requirements allows the system to identify potential supporters from among multiple people in the environment with a high probability of successfully supporting the first person.

[0022] According to one embodiment, a computer-implemented method includes determining each person's ability to support at least one first person by applying a trained model or by accessing a database to determine each person's skill level.

[0023] A trained model is a model that includes model parameters trained during the training phase, which is trained using machine learning algorithms and training data. Techniques exist for detecting people who are having trouble finding their way. Using trained models is particularly useful in public spaces as an environment for intelligent agents with a large number of unknown people spending time in, for example, a pedestrian area or a shopping mall. Databases can be a particularly efficient source of information about suitable support candidates from among multiple people in a scenario where others are known in advance, such as members of a closed group of people, for example, members of a team at a manufacturing site or employees of the same company with access to a facility.

[0024] The computer-implemented method may include determining at least one second person by an intelligent agent moving around the environment in order to detect further people for evaluation of the probability and ability to support at least one first person.

[0025] The other person is a person included in a plurality of people in the environment, other than the determined first person. The environment of the intelligent agent is the area around the intelligent agent that can be observed by at least one sensor. Thus, the environment essentially corresponds to the sensor range or sensor coverage, excluding, for example, areas where obstacles by objects in the environment prevent monitoring by at least one sensor.

[0026] A computer-implemented method according to an embodiment includes determining a support probability for each person to support at least one first person or to determine a support probability for each other person among a plurality of people other than the at least one first person, by applying a trained model, such as a model trained via a machine learning algorithm.

[0027] Using a trained model enables a computationally efficient implementation and is particularly useful in public spaces as the environment of an intelligent agent having a large number of unknown people spending time, for example, in a pedestrian area or a shopping mall.

[0028] In an embodiment of the computer-implemented method, determining the support probability includes at least one of determining whether the other person is currently performing other tasks, particularly whether performing other tasks and the urgency of the other tasks, whether in conversation, whether classified as being in a hurry, whether classified as being angry, and evaluating the cost to the other person for pausing other tasks and instead performing a support action for at least one first person.

[0029] Using specific criteria enables the system to plan actions for an agent targeted at potential supporters for the first person who are likely to support the first person in their actions. Thus, the intelligent agent operates based on actions planned in an efficient manner and arranges support for the first person early.

[0030] In a computer-implemented embodiment of the method, the method includes determining at least one second person, which includes determining the distance from another person to at least one first person.

[0031] Spatial distance allows the system to assess the probability of another person supporting a first person. The smaller the distance, the more likely another person is to provide support to the first person, because the cost of supporting the first person and returning to their own work decreases with decreasing distance.

[0032] In a computer-implemented embodiment of the method, determining at least one second person includes determining whether the other person is part of a group of people.

[0033] Members of a group of people are less likely to support the first person and are also less likely to gain attention through the actions of each intelligent agent. The probability of others supporting the first person decreases when those others are part of a group of people, for example, due to the bystander effect.

[0034] In a computer-implemented embodiment of the method, determining at least one second person includes determining the probability of support for each other based on the cost for the intelligent agent to approach the other person.

[0035] Therefore, the planned actions of an intelligent agent are efficient with respect to the intelligent agent's resources.

[0036] According to an embodiment of the computer-implemented method, the cost of approaching another person includes determining the distance from each person to the communication device for contacting the other person, each person's reachability, each person's attention to the communication device, and in particular the direction of each person's gaze to the communication device.

[0037] Therefore, the probability of successful communication increases the probability that the second person will come to support the first person.

[0038] A computer-implemented method according to an embodiment includes a processor planning the actions of an intelligent agent that optimizes at least one of the support capacity and support probability of at least one second person, further based on at least one of the cost and appropriateness of supporting at least one first person in performing the task.

[0039] The appropriateness of supporting the first person describes the quality of whether the potential support is appropriate or justified, for example, in accordance with social norms in a particular situation. Appropriateness may be determined based on predetermined information obtained from a database that stores a set of norms and rules for evaluation and application by method. Thus, support from the second person to the first person is more likely to conform to social standards among people and be favorably received and accepted by the first person.

[0040] In a computer-implemented embodiment of the method, performing at least one of communication or physical action by an intelligent agent includes at least one of the following: acoustically outputting text content describing the need for support in performing a task to at least one second person; visually outputting text content describing the need for support in performing a task to at least one second person; acoustically outputting information on what action at least one second person is required to take to support at least one first person; visually outputting information on what action at least one second person is required to take to support at least one first person; acoustically outputting text content guiding at least one second person to at least one first person; and visually outputting text content guiding at least one second person to at least one first person; and performing a physical action includes the intelligent agent moving to guide at least one second person to at least one first person.

[0041] A framework for action plans for intelligent agents to organize mutual support among people in the same environment works with intelligent agents that have a wide range of communication capabilities, including not only acoustic or visual communication but also other means of nonverbal communication to improve support among multiple people.

[0042] A system for controlling an intelligent agent operating in an environment with multiple people, comprising at least a processor and at least one sensor, wherein the system is configured to implement steps of a computer-implemented method, and the system achieves the corresponding advantageous effects as discussed with respect to the computer-implemented method.

[0043] The system may be implemented as part of an intelligent agent acting as a mobile robot, perhaps with at least partially additional remote processing capabilities. Alternatively, the system may be implemented as an intelligent agent in a distributed manner, for example, operating in the background of a local factory environment, using sensors and communication means placed in the factory environment to improve mutual support within a team of workers.

[0044] The following description of the embodiments will be accompanied by reference to the figures. [Brief explanation of the drawing]

[0045] [Figure 1] This is a simplified flowchart illustrating the processing steps of a computer-implemented method. [Figure 2] This is an overview diagram of the functional architecture of a system implementing the embodiment. [Figure 3] This figure shows application scenarios in which an intelligent agent is controlled according to the embodiment. [Figure 4] This is a simplified overview diagram of the hardware architecture for implementing the embodiment. [Modes for carrying out the invention]

[0046] In the diagrams, corresponding elements have the same reference numeral. Where possible, the diagrams should avoid discussing the same reference numerals in different diagrams, while avoiding unnecessary repetition for the sake of brevity without negatively impacting comprehensibility.

[0047] Figure 1 shows a simplified flowchart illustrating the processing steps of the computer-implemented method. The computer-implemented method controls an intelligent agent 11 operating in an environment with multiple people 13, 14, and 15.

[0048] The computer-implemented method is implemented in an (electronic) system 1 comprising at least one sensor 2 (sensing device), such as a camera, a processor (processing device), and at least one information communication means (communication device) for outputting acoustic information, such as sound via a speaker, or visual information, such as text via a display, or both.

[0049] In a specific implementation, the intelligent agent 11 is a robotic device, and in addition to its sensors 2, the human-machine interface 9 includes an actor 8 (actuator 8) that allows the intelligent agent 11 to move around the environment. Additionally or alternatively, the intelligent agent 11 uses its actor 8 for non-verbal communication.

[0050] The method includes step S1 of acquiring information about multiple people 12, 13, 14, and 15 via at least one sensor 2 of the intelligent agent 11. At least one sensor 2 may include a camera of the intelligent agent 11. Using the sensor 2, the intelligent agent 11 monitors the environment and acquires information about individual people 13, 14, and 15, and a group 15 of people present in the current scenario in the environment monitored by the sensor 2.

[0051] Sensor 2 provides the acquired information to at least one processor 23 of system 1.

[0052] In step S2, the method determines by the processor 23 whether the first person 12, at least one of the multiple people 12, 13, 14, 15, is having difficulty performing the task determined by evaluating information obtained from at least one sensor 2.

[0053] The processor 23 may determine whether at least one of the multiple people 12, 13, 14, 15 is struggling by determining whether person 12 is making progress when performing the task 20 that the processor 23 predicts person 12 will face.

[0054] Specifically, in step S2, the processor 23 uses the acquired information to infer the current state of at least one individual person 12 or a group of individuals 15 using a known machine learning algorithm. The processor 23 uses the acquired information to determine the current activity of at least one individual person 12 or a group of individuals 15 using a known machine learning algorithm. The processor 23 uses the acquired information to determine the current task (goal) that at least one individual person 12 or a group of individuals 15 is engaged in using a known machine learning algorithm.

[0055] The processor 23 assesses whether at least one individual person 12, or a group of individuals 15, is likely to request help in the near future, or in particular likely to request help, based on their current state, work, and activities, in order to reach their goal. The processor 23 may use one or a combination of the following:

[0056] The method may use a machine learning algorithm trained to classify the current actions of people 12, 13, 14, and 15 regarding the task. The processor 23 may use a task model to compare the current actions determined based on the acquired information with the correct actions to achieve the task's objective, and to detect discrepancies between the current actions and the correct actions.

[0057] The method may involve using a machine learning algorithm or a database 10 to understand the skill and knowledge level of at least one person 12 who is having difficulty performing task 20.

[0058] The method may use a work model to compare the current skills or knowledge attributable to at least one person 12 who is struggling to perform the work 20 with the skills or knowledge required to address the work, and to detect any discrepancies between the current skills or knowledge and the required skills or knowledge.

[0059] In step S2, the method may use a machine learning algorithm trained to classify human behavior or human state as struggling to work, such as prolonged hesitation, sadness, exhaustion, pain, or embarrassment, repetition of the same action, falling to the ground or lying down, or slowing down an activity performed by at least one person 12.

[0060] The method may determine whether at least one person 12 is having difficulty performing the task 20, based on detected work failures, process errors, and especially the repeated occurrence of such things.

[0061] Optionally, the method may use work models and planning algorithms to determine the need for others to support at least one person who is having difficulty performing work 20.

[0062] If the processor 23 determines in step S2 that there is at least one first person 12 in the environment who is having difficulty performing the task (20), the method proceeds to step S3 (yes). Alternatively, if the processor 23 determines in step S2 that there are no people 12, 13, 14, 15 in the environment who are having difficulty performing the task 20, the method returns to step S1 (no) and continues to monitor the environment via at least one sensor 2.

[0063] In step S3, the method determines, by the processor 23, at least one second person 13 from a group of people 13, 14, 15 to support the first person 12 who is struggling with his work 20, by evaluating the information obtained from step S1 regarding the environment.

[0064] The evaluation includes, in particular, evaluating multiple persons 13, 14, and 15 with respect to acting to support at least one first person 12 in order to perform task 20. Specifically, the state of each person 13, 14, and 15 includes determining each person 13, 14, and 15's support capacity to support at least one first person 12, and determining the probability of each person 13, 14, and 15 supporting at least one first person 12.

[0065] If at least one first person 12 who requires support is detected in step S2, the method evaluates all other people 13, 14, 15 in the environment, for example, all within the detection range of at least one sensor 2, in terms of their probability of supporting the first person 12 and their ability to support (support potential), by performing one or a combination of the following:

[0066] The method may use a machine learning algorithm or a database 10 to determine at least one of the following for each person 13, 14, or 15: skill level, knowledge level, or object possession, and compare it to the skills required to support the first person 12.

[0067] The method may use machine learning algorithms to detect the level of support availability for each person 13, 14, and 15, and the anticipated urgency of that task 20, such as whether people 13, 14, and 15 are currently performing their own task 20. For example, the task 20 of relevant potential supporters to assess their ability to support person 12 may include, for example, being in a conversation with other people 13, 14, and 15, or being classified as “urgent” or “angry.”

[0068] The method may involve temporarily suspending oneself from engaging in one's own work and instead using the determined work 20 of others 13, 14, and 15 to assess their individual costs for performing support actions for the first person 12.

[0069] Alternatively, the method may involve determining the spatial distance from each of the other people 13, 14, and 15 to the first person.

[0070] The method may also involve determining whether the other individuals 13, 14, and 15 are part of group 15, because, for example, the bystander effect may skew the likelihood of supporting the first person 12.

[0071] The method may evaluate the other persons 13, 14, and 15 in terms of the probability or cost for System 1 to approach these other persons 13, 14, and 15, for example, the distance to the communication device, the reachability to the artificial agent 11, and the attention or interest regarding the communication device.

[0072] Optionally, the method may determine an approximation of the cost for system 1 to guide the second person 13 to the first person 12, for example, by determining whether a line of sight exists between the second person 13 and the first person 12.

[0073] Optionally, the method assesses each of the other persons 13, 14, 15 with respect to the social expectations and comfort implications of the first person 12 in need of support, such as the fact that women may be more likely to receive support from other women.

[0074] The method then selects at least one second person 13 from the others 13, 14, and 15 that provides the optimal combination of help probability and calculated approach cost to support the first person 12.

[0075] The method then proceeds to step S4, in which the processor 23 plans the actions of an intelligent agent 11 to optimize at least one of the support capacity and support probability of at least one second person 13 in order to support at least one first person 12 in performing the task 20. Specifically, if a suitable pair of people including at least one first person 12 and at least one second person 13 is determined, the method plans the actions of the artificial agent 11 to approach the second person 13 and optimizes the efficiency, success probability and appropriateness of the support provided by the planned actions.

[0076] In step S5 of the method, the intelligent agent 11 performs at least one of communication or physical action based on a planned action to ensure that at least one first person 12 benefits from the support of at least one second person when performing the task 20.

[0077] For example, intelligent agent 11 may perform a selected action, such as outputting generated audio content to at least one second person 13 that describes the need for help for at least one first person 12. Actions performed by intelligent agent 11 may also include physically guiding the second person 13 “Helper” to the location where the first person 12 “Being Helped” needs support. Actions performed by intelligent agent 11 may also include displaying on a screen what actions need to be taken to help the first person 12 reach their goal. Actions performed by intelligent agent 11 may also include improving the second person 13’s ability to act in support of the first person 12 by, for example, holding an object for the second person 13 or the first person 12 to free their hands. Actions performed by intelligent agent 11 may also include supporting the second person 13 to cease any work currently being performed while it is supporting the first person 12.

[0078] Figure 2 provides an overview of the functional architecture of System 1 implementing the embodiment.

[0079] System 1 includes, as functional elements, a person / group detection module 3, a helper evaluation module 5, a help need detection module 4, a helper selection module 6, and a helper recruitment action planner module 7.

[0080] The person / group detection module 3 acquires information from the environment of system 1 from at least one sensor 2. The person / group detection module 3 evaluates the information from the environment acquired by sensor 2, detects people 12, 13, and 14 in the environment, and determines the group of people 12, 13, 14, and 15 in the environment based on the acquired information. The person / group detection module 3 generates information about the detected people 12, 13, 14, and 15, and, if any, information about the group of detected people, and provides the generated information to the helper need detection module 4 and the helper evaluation module 5 for further processing.

[0081] The assistance need detection module 4 acquires information from the environment of system 1 from at least one sensor 2. In addition, the assistance need detection module 4 obtains information from the person / group detection module 3 regarding the detected individuals 12, 13, 14, and 15, as well as information regarding the group of detected individuals.

[0082] The help need detection module 4 determines the tasks 20, activities, or goals that may be performed by individuals 12, 13, 14 and groups of individuals 15, as provided by the person / group detection module 3.

[0083] Task 20 includes, for example, explicit manual work. Manual work may include, for example, constructing or assembling objects, picking up, carrying, moving and transporting objects, using objects, organizing objects, and placing objects.

[0084] The object may be any kind of physical object, including, for example, a tool for handling other objects.

[0085] Task 20 may be a navigation task, such as attempting to reach a target location, or searching for an object, animal, or other person.

[0086] Task 20 may be social tasks, such as having conversations among people 12, 13, 14, and 15; arguing with others; supervising children, especially keeping them from running away; engaging in physical disputes with others; learning about others; or caring for objects, animals, or other people.

[0087] Task 20 may include at least one of the following cognitive tasks: understanding a theory, learning something, solving a puzzle, making a decision, answering a question, coding, and gathering information.

[0088] Task 20 may be associated with a time limit for performing Task 20, for example, to board the train while the doors are still open.

[0089] Task 20 may be associated with pressure from the implications of performing Task 20.

[0090] Task 20 may be a repetitive or basic task, and may create pressure to complete the basic task 20.

[0091] Task 20 may be a physiological state associated with person 12, 13, 14, or 15, such as hunger, thirst, injury, or incapacity. Task 20 may also be an emotional state, such as sadness, happiness, confusion, or fear.

[0092] The assistance need detection module 4 uses the determined tasks 20, activities, or objectives that may be performed by the individuals 12, 13, 14 and group 15 to determine if the first person 12 may require external assistance from other people 13, 14, 15, resulting in skill mismatches, knowledge mismatches, and work failures.

[0093] Examples of skill mismatches and work failures that the assistance need detection module 4 determines may require external assistance for the first person facing task 20 include, for example, physical limitations, such as the first person 12 being too short to reach a shelf high up on a wall. The first person 12 may also have limited mobility due to, for example, a limited speed when moving, limited strength when lifting objects, limited vision when reading information, or more fundamentally, a temporary or permanent handicap that leads to a single focus of attention for the first person 12.

[0094] Task 20 may require multiple people, for example, by carrying a large object or carrying many objects.

[0095] The completion of Task 20 may be adversely affected by the knowledge limitations of Person 12, such as errors in memory regarding the necessary tools or work sequence, an unknown route to the target location, or inadequate preparation for dealing with Task 20.

[0096] Completing Task 20 may be negatively affected by the skill limitations of the first person, such as being unfamiliar with certain dexterity movements or having to use previously unknown tools to perform Task 20.

[0097] The completion of Task 20 may be adversely affected by contextual factors, which may include, for example, external obstacles such as an additional Task 20, a blocked road preventing access to the navigation objective, or the unavailability of the tools necessary to perform Task 20.

[0098] The completion of task 20 may be affected by the misfortune of the first person 12, for example, slipping and getting injured, or mistaking an object during the process of completing the task.

[0099] The helper evaluation module 5 obtains information from the environment of system 1 from at least one sensor 2. In addition, the helper evaluation module 5 obtains information from the person / group detection module 3 regarding the detected individuals 12, 13, 14, and 15, as well as information regarding the group of detected individuals. The helper evaluation module 5 also obtains information from the help need detection module 4 regarding the task 20 and the first person 12 who is having difficulty performing the task.

[0100] The helper evaluation module 5 in Figure 2 also accesses database 10 to retrieve additional information related to evaluating the support capabilities of other people 13, 14, and 15.

[0101] Helper Assessment Module 5 evaluates the information obtained, and in particular assesses the potential for help from the other 13, 14, and 15 people among the group of 12, 13, 14, and 15, excluding the first person 12. Helper Assessment Module 5 determines the support capacity of each of the other 13, 14, and 15 people.

[0102] Assessing support capabilities may include matching the skill levels or knowledge of other persons 13, 14, 15 to the requirements of the task 20 and the support situation. Required skill levels or knowledge may include, for example, needing local residents to help Person 12 who is having difficulty finding their way to a specific point in the environment, or having medical knowledge to help Person 12 who is injured.

[0103] Assessing support capabilities may include matching the object possessions of each individual person 13, 14, 15 with the requirements of the work 20 and support situation detected in the environment, such as carrying water when it is thought that the first person 12 needs a drink, or organizing a flashlight to help retrieve lost items from under a parked vehicle in difficult lighting conditions.

[0104] Assessing the ability to provide support may include matching the physical capabilities of the other persons 13, 14, and 15 against the requirements for the support situation, such as determining the strength of the other persons 13, 14, and 15 to lift heavy objects, their mobility, for example, their mobility on rough ground, for example, whether they are not in a wheelchair to help the first person 12 descend stairs 21.

[0105] Helper assessment module 5 evaluates the information obtained, and in particular assesses the probability or claimed willingness of the other persons 13, 14, and 15, excluding person 12, among the group of persons 12, 13, 14, and 15. Helper assessment module 5 determines the probability of support for each of the other persons 13, 14, and 15.

[0106] Evaluating the probability of each other 13, 14, and 15 providing support may include, for example, evaluating the availability of help if the other 13, 14, and 15 are currently doing their own work, which may include, for example, having a conversation, being on their way to a particular place, or watching over a child, all of which may decrease the probability of supporting the first person 12 in doing the work 20 that the first person is struggling with.

[0107] Evaluating the probability of support for each of the other persons 13, 14, and 15 may include determining how likely or costly it is to temporarily suspend the work of the candidate for the role of second person 13, or how costly it is to change the status of the candidate for the role of second person 13.

[0108] Assessing the probability of each other 13, 14, 15 supporting others may include determining the states of others 13, 14, 15 such as being stressed and tired, sad, happy, or angry; cognitive states such as being in a hurry or exploring; and transitions between tasks that may indicate a reduced willingness or probability to support the first person 12.

[0109] Evaluating the probability of support for each of the other persons 13, 14, and 15 may include estimating the time, effort, or cost of taking a helping action for each of the other persons 13, 14, and 15. The time, effort, or cost for each of the other persons 13, 14, and 15 may be assessed by determining the distance to the first person 12 for each of the other persons 13, 14, and 15.

[0110] Evaluating the probability of each other 13, 14, and 15 providing support may include estimating the required skill level of each of the other 13, 14, and 15, for example, that highly skilled individuals can provide the necessary support actions in a relatively short amount of time. Therefore, highly skilled individuals may be more likely to support the first person 12 than the other 13, 14, and 15 who possess only a limited set of skills regarding the necessary support actions.

[0111] Evaluating the probability of each of the other persons 13, 14, and 15 supporting the first person 12 may include estimating the time, effort, or cost for any additional steps each of the other persons 13, 14, and 15 may need to take to return to their original activities after supporting the first person 12. This is especially true if supporting the first person 12 may result in the other persons 13, 14, and 15 moving further away from their original positions.

[0112] Evaluating the probability of support from each of the other persons 13, 14, and 15 may include considering the additional cost if the candidate for the role of the second person 13 is part of the group of persons 15, because this can negatively bias the probability of support from others helping the first person 12, for example, due to the bystander effect. The bystander effect describes how, in emergency situations, people are more likely to take action when there are few or no other witnesses. Being part of a group of persons means that no individual person is responsible for the action.

[0113] Evaluating the probability of each other supporting others13,14,15 may include evaluating the additional benefit that reduces the estimated cost if it is recognized or known that a person will help others, or it may include adding a further penalty to the cost if it is recognized or known that a person will refuse the support request at an earlier opportunity.

[0114] Evaluating the probability of support for each other 13, 14, and 15 may include evaluating the additional benefit to the cost if the candidate for the role of second person 13 is known to have a specific relationship with first person 12, e.g., responsible supervisor, team leader, school, university, or workplace acquaintance.

[0115] Evaluating the probability of support for each other 13, 14, and 15 may include evaluating the additional benefit to cost if the candidate's claimed occupation for the role of second person 13, e.g., police officer, firefighter, or medical worker, is known and, for example, expected to help others, at least while on duty.

[0116] The helper evaluation module 5 evaluates the information obtained and, in particular, evaluates the probability and ability of the other persons 13, 14, and 15 (excluding the first person 12) to play the role of the second person 13 supporting the first person 12. The helper evaluation module 5 provides the helper selection module 6 with the determined probability and ability of each of the other persons 13, 14, and 15.

[0117] Subsequently, the helper selection module 6 selects at least one second person 13 from among the others 13, 14, and 15 as a candidate to support at least one first person 12, optimizing at least one of the support capabilities and support probabilities of at least one second person 13 to support at least one first person 12 when performing the work.

[0118] Selecting at least one second person 13 that optimizes at least one of support ability and support probability includes, for example, determining the weighted sum of each determined support ability and support probability for each of the other people 13, 14, and 15 among a group of people 12, 13, 14, and 15. The weights may be predetermined weights stored in advance. System 1 then selects the other person 13, 14, and 15 that has the largest weighted sum of their respective determined support ability and support probability as the second person 13. It is also possible to use alternative selection criteria other than the metrics discussed.

[0119] The determined task may require system 1 to select multiple second persons 13 from other persons 13, 14, and 15.

[0120] The helper selection module 6 provides the helper recruitment action planner module 7 with at least one selected second person 13 to support at least one first person 12.

[0121] Subsequently, the helper recruitment action planner module 7 executes an action plan for the intelligent agent 11, and the planned action maximizes the probability that the second person 13 will support the first person 12 when performing task 20.

[0122] A planned action for at least one second person 13 may include, for example, at least one of the described actions for execution by the intelligent agent 11.

[0123] Exemplary actions of intelligent agent 11 may include directly asking a second person 13 whether they can help the first person 12. Optionally, intelligent agent 11 may indicate to the second person 13 who to support, where to support the first person 12, and how to support the first person 12.

[0124] The exemplary behavior of intelligent agent 11 may include optionally explaining to the second person 13 why the second person 13 has been chosen to support the first person 12.

[0125] System 1 may use a human-machine interface 9 of an intelligent agent 11, for example, to output text information on a display screen, voice messages from a speaker, or send messages to a specific device, such as a second person 13's smartphone.

[0126] An exemplary planned action of the intelligent agent 11 may include moving along a trajectory that brings the intelligent agent's human-machine interface 9 to a certain position relative to a second person 13 selected to support a first person 12, thereby increasing the likelihood that the second person 13 will perceive any information provided through the human-machine interface 9.

[0127] An exemplary planned action of the intelligent agent 11 may include displaying a symbol, such as a help sign, on a display screen, for example, on the display screen of the intelligent agent 11's human-machine interface 9.

[0128] Exemplary planned actions of intelligent agent 11 may include visual simulation, for example, using an actor of intelligent agent 11 to wave to a second person 13, for example, to follow intelligent agent 11 or to come to intelligent agent 11.

[0129] Exemplary planned actions of the intelligent agent 11 may include leading the second person 13, moving forward to guide them, and pointing arrows in the relevant directions toward the first person 12's location.

[0130] Exemplary planned actions of intelligent agent 11 may include attention guidance, for example, using actor 8 of intelligent agent 11 to point to a first person 12, and using shared attention to direct a second person 13 towards the first person 12.

[0131] Exemplary planned actions of intelligent agent 11 may include performing supportive actions for the second person 13, such as suggesting to the second person 13 that they help pause their current activity. Exemplary planned actions of intelligent agent 11 may include performing supportive actions for the second person 13, such as holding an object so that it does not become a burden when the second person 13 is supporting the first person 12. Exemplary planned actions of intelligent agent 11 may include supportive actions such as watching over or monitoring a child or a dog.

[0132] Figure 3 shows an application scenario in which the intelligent agent 11 is controlled according to the embodiment. The example in Figure 3 is a specific example illustrating the advantages of System 1. The artificial agent 11 is located in a public space, such as a shopping mall or train station, or a pedestrian intersection. A staircase 21 is located in the environment monitored by the sensor 2 of the artificial agent 11.

[0133] The intelligent agent 11 detects multiple people 12, 13, 14, and 15 in the environment based on sensor information acquired from sensor 2.

[0134] Several people, 12, 13, 14, and 15, including a woman with a stroller approaching the top of the stairs 21.

[0135] System 1 predicts the intention of the woman descending the stairs 21. System 1 determines the task 20 that the woman is facing as she descends the stairs 21 with a stroller, and determines that the woman is the first person 12 who needs support when performing the task 20 of descending the stairs 21 with a stroller.

[0136] The persons 12, 13, 14, and 15 in the environment monitored by the intelligent agent 11 include persons 13, 14, and 15 other than the first person 12, and include a single person 13, a person with a small child 14, and a group of three persons 15.

[0137] System 1 then proceeds to evaluate others in terms of their ability to support a first person 12 descending stairs 21 by carrying a stroller, and in terms of their availability to support the first person 12.

[0138] Person 13, acting alone, is a promising candidate to support Person 12. Person 13 is closer to the top of the stairs 21 than the others 14, 15, and Person 12, who is determined to be in need of support. Person 14 is at a greater distance from the top of the stairs 21 than Person 13 and is holding hands with a small child. As a result, Person 14 is determined to have less support capacity due to being occupied by the child and being at a greater distance from Person 12 compared to Person 13. The group of people 15 is also at a greater distance from Person 13 to Person 12 at the top of the stairs 21, and it may be difficult for the intelligent agent 11 to attract the attention of the three people making up the group of people 15.

[0139] Based on the scene evaluation, System 1 determines that Person 13 is the most suitable and promising candidate from among Persons 13, 14, and 15, who represent all candidates to support the first Person 12, and selects Person 13 as the second Person 13.

[0140] System 1 plans the actions of intelligent agent 11, including approaching the second person 13 and guiding the second person 13 along the path 16 to the top of the stairs 21, as well as communicating the need for support for the first person 12 as they descend the stairs 21 with the stroller.

[0141] The basic scenario in Figure 3 may be modified in different ways to illustrate further aspects and advantages of computer-implemented methods for controlling intelligent agents.

[0142] In a concrete example, the intelligent agent 11 is an unmanned aerial vehicle (UAV) capable of flying and hovering over a main road where an entrance is located, for example, for an underground pedestrian crossing or to a subway. This example of the intelligent agent 11 provides an elevated monitoring position for the UAV-mounted sensor 2, which has a particularly good view and only limited obstacles, when determining a first person 12 and assessing other people 13, 14, 15 and their support capabilities and availability of support. Furthermore, the active search for further supporter candidates to determine a second person 13 can be particularly simple due to the flight capabilities of the intelligent agent 11.

[0143] Some of the other 13, 14, and 15 may be carrying bags, but that reduces their ability and availability to provide support.

[0144] Some of the other 13, 14, and 15 may just be entering the building, which reduces their availability. This can be further predicted by System 1 if, for example, signs of haste or stress are detected in the walking patterns of the other 13, 14, and 15.

[0145] System 1 and its computer-implemented methods include different but highly advantageous application scenarios, including System 1, which is an electronic system that controls different sensing and communication systems in a factory environment to record the work of a group (team) of workers on an assembly line. The information obtained reveals that an inexperienced worker is using the wrong tool for the current assembly task being performed on the assembly line. The inexperienced worker struggles to successfully join two required parts using an unsuitable tool.

[0146] System 1 determines that an inexperienced worker in the team is struggling to perform his current task 20. System 1 then proceeds to check whether other workers on the team who have access to System 1 can support the inexperienced worker representing Person 12. System 1 may have access to a database 10 containing information about the team's workers, their experience, and their respective skills. Some of the other workers on the team are newcomers. The system concludes that the other inexperienced workers may not be able to identify the problem with Person 1. Others 13, 14, 15 may now be detected as unavailable because they are in different rooms. Alternatively, System 1 may classify at least some of the other workers on the team as having performed a longer task, and therefore selecting them as Person 2 to support Person 13 might delay the overall work time of the team. Based on the evaluation of the scenario, System 1 selects one of the more skilled workers who has just finished his current task as the most suitable supporter (Person 2) for the inexperienced worker (Person 12) struggling with his task.

[0147] The following action plan for System 1 may be simple if System 1's only ability to interact with people 12, 13, 14, and 15 is to send messages to workers, for example, to their personal devices. In the example discussed, System 1 generates and sends a short message to a selected worker (second person 13) asking them to go to the struggling worker (first person 12) in order to identify errors and guide the inexperienced worker toward achieving further progress in performing task 20.

[0148] A different but highly advantageous application scenario for System 1 and its computer-implemented method involves an intelligent agent 11, which is a mobile robot moving through a shopping mall. System 1 detects a person sitting on a bench in the shopping mall who exhibits behavioral signs of dehydration, for example, System 1 detects the person's rapid breathing and fatigued posture based on information obtained from the environment.

[0149] System 1 is unable to detect other people 13, 14, and 15 in the environment of Person 12, who is showing signs of dehydration. System 1 plans the actions of Intelligent Agent 11, including moving around the shopping mall to find other people 13, 14, and 15 who may be able to help Person 12. Of the detected candidates, System 1 detects a group 15 of people who are just about to leave a supermarket in the shopping mall. System 1 determines that this group has a high probability that at least one person in group 15 may have a drink. System 1 further determines that approaching multiple people increases the probability that someone is willing to help. Based on the actions planned by System 1, Intelligent Agent 11 approaches the group by moving over to them and asking group 15 if they have water and if they can stay with Intelligent Agent 11 for a while to help someone who is in urgent need of help. Subsequently, the intelligent agent's planned actions include immediately beginning to move along a trajectory from the point where it approached the group of people 15 toward the location of the first person 12 where the assumed dehydrated person is sitting, assuming that at least some of the people in group 15 will follow the intelligent agent 11 and provide it to the first person 12 if they have some water. If system 1 determines that no one is following the intelligent agent 11, system 1 plans updated actions, including searching for new candidates for the role of the second person 13 to support the first person 12.

[0150] Figure 4 shows a simplified overview of the hardware architecture for implementing System 1 to control the intelligent agent 11.

[0151] System 1 may be implemented using at least partially the resources of the intelligent agent 11, such as the intelligent agent 11's sensor 2, processor 23, memory 24, human-machine interface 9, and actor 8.

[0152] The intelligent agent 11 may include at least one of the following: a mobile robot, a stationary robot, a group of robots, an unmanned aerial vehicle (UAV) or drone, or one or more screens, a distributed computing system, or a virtual avatar on a smart home system.

[0153] System 1 may also use resources different from each of the intelligent agents 11, such as external sensors, or at least one server 27 that communicates with at least one intelligent agent 11 via a network 28.

[0154] Server 27 may include data storage capacity and processing capacity for a distributed implementation of System 1. In particular, Server 27 may include at least one processor and at least one memory for storing, for example, the database 10.

[0155] The intelligent agent 11 may perform at least a portion of its processing on the server 27 via the communication interface 21 and the communication network 28.

[0156] The intelligent agent 11 includes memory 24 for locally storing the database 10. Alternatively, or additionally, the processor 23 of the intelligent agent 11 accesses the database 10 stored on the server 27 via the intelligent agent 11's bus 25 and communication interface 21, and further via the network 28.

[0157] The intelligent agent 11 may also acquire further sensor information from sensors not shown in Figure 4 via the intelligent agent 11's bus 25 and communication interface 21, as well as via network 28, or a different communication link separate from the communication network 21.

[0158] The intelligent agent 11 further includes a human-machine interface 9 (HMI 9) that enables the intelligent agent to receive input from people 12, 13, 14, and 15 and to output information to people 12, 13, 14, and 15.

[0159] Inputting and outputting information via the HMI9 may include communicating information via at least one of acoustic signals, visual signals, and tactile signals. The HMI9 includes at least one microphone for acquiring acoustic information, at least one speaker for outputting acoustic information, and at least one display including a display screen for visually outputting information to people 12, 13, 14, 15.

[0160] The display and projector, as part of the HMI9 of the intelligent agent 11, may be installed in various locations, including, for example, on a mobile robot or within the environment, such as a wall-mounted display in a shopping mall or train station, or a projector that projects visual information onto the floor or wall.

[0161] The speaker, as part of the HMI9 of the intelligent agent 11, may be installed in various locations, including, for example, on a mobile robot or within the environment, such as a wall-mounted speaker in a shopping mall or train station.

[0162] HMI9 may include the ability to send messages to a personal device, such as a smartphone of person 12, 13, 14, or 15, via at least one of email, SMS, MMS, video call, or picture.

[0163] In specific embodiments, the planned actions may include communicating information, for example, by pointing, through the facial expressions and body posture of the intelligent agent 11, or by physical interaction, for example, by using the manipulator arm of the intelligent agent 11 to grasp the hands of people 12, 13, 14, and 15 under the control of the motion controller 26.

[0164] The processor 23 obtains information about the environment from at least one sensor 2 of the intelligent agent 11, for example, via the bus 25 shown in Figure 4.

[0165] The processor 23 may use the intelligent agent motion controller 26, which controls the actor 8 of the intelligent agent 11, to execute the planned actions of the intelligent agent 11, including the movement of the intelligent agent.

[0166] Actor 8 may include electric motors that drive the legs or arms of intelligent agent 11, similar to a humanoid robot.

[0167] Alternatively, or additionally, actor 8 may include wheels or tracks for the ground-mobile intelligent agent 11 to move the intelligent agent 11 within the environment based on planned actions. Alternatively, or additionally, actor 8 may include propellers for the air-mobile intelligent agent 11 to move the intelligent agent 11 within the environment based on planned actions.

[0168] Sensor 2 includes at least one of the following: a camera sensor, an infrared (IR) sensor, an ultrasonic sensor, a LiDAR sensor, a radar sensor, a touch sensor, such as a capacitive touch sensor, and a device motion sensor, such as a torque sensor. Sensor 2 includes a main actuating device, a sensor on a mobile robot, an environment sensor on other moving or stationary devices accessible to System 1, such as a surveillance camera, such as an in-vehicle camera, or as part of a wearable device carried by a human user, such as a mobile phone sensor or a smartwatch.

[0169] System 1 is not limited to using a mobile intelligent agent 11. Alternative embodiments may include an intelligent agent 11 implemented in a distributed manner, for example, using sensors 2, HMI 9, processor 23, memory 24, bus 25, and communication interface 21 distributed across an environment (domain) of an industrial manufacturing site.

[0170] System 1 may be implemented using multiple intelligent agents 11 linked via a communication network 28.

[0171] All steps performed by the various entities described in this disclosure, and functions described to be performed by the various entities, are intended to mean that each entity is adapted or configured to perform its respective step and function.

[0172] In the claims and specification, the word “comprising” does not exclude the presence of other elements or steps. The indefinite article “a” or “an” does not exclude the plural.

[0173] A single element or other unit may perform the functions of several entities or items described in the claims. The mere fact that different dependent claims describe several measures and features of the method does not preclude combinations of these measures and features from being combined in a favorable implementation.

[0174] Features described in the description of specific embodiments and illustrated in the drawings may be combined with respect to the invention defined in the attached claims. [Explanation of Symbols]

[0175] 1 System 2 sensors 3 Person / Group Detection Module 4. Help Need Detection Module 5. Helper Evaluation Module 6. Helper Selection Module 7. Helper Recruitment Action Planner Module 8 Actuators 9. Human-Machine Interface 10 Databases 11 Intelligent Agents 12, 13, 14, 15 people 16 Travel Route 20 tasks 21 stairs 21 Communication Interface 23 processors 24 memory 25 buses 26 Motion Controller 27 servers 28 Network

Claims

1. A computer-implemented method for controlling an intelligent agent (11) operating in an environment with multiple people (12, 13, 14, 15), wherein the method is S1) involves acquiring information about the plurality of people (12, 13, 14, 15) via at least one sensor (2) of the intelligent agent (11), The processor (23) determines whether at least one of the multiple persons (12, 13, 14, 15), the first person (12), is having difficulty performing the task determined by evaluating the acquired information (S2), If at least one of the first persons (12) is having difficulty performing the task (20), The processor (23) operates in support of the at least one first person (12) in order to perform the task (20), and in this regard, the state of each of the multiple persons (12, 13, 14, 15) other than the at least first person (12) (13, 14, 15) is evaluated. Evaluating the state of each person (13, 14, 15) includes determining each person's (13, 14, 15) ability to support the at least one first person (12), and determining the probability of each person (12, 13, 14, 15) supporting the at least one first person (12). The processor (23) selects at least one second person (13) from the plurality of people (12, 13, 14, 15) who optimizes at least one of the support ability and the support probability (S3), The processor (23) plans the actions of the intelligent agent (11) to optimize at least one of the support capacity and support probability of the at least one second person (13) in order to support the at least one first person (12) when performing the task (S4), The intelligent agent (11) performs at least one of the following based on the planned actions (S5): Computerized methods, including those mentioned above.

2. The computer-implemented method according to claim 1, wherein the step (S2) of determining the at least one first person (12) includes using a work model for evaluating the acquired information by comparing at least one of the first person (12)'s current skills or knowledge with the corresponding at least one of the first person (12)'s required skills or knowledge for performing the work, and determining whether the first person (12)'s current skills or knowledge deviate from the first person (12)'s required skills or knowledge.

3. The aforementioned method, Using a trained model to evaluate the acquired information, determine which of the multiple persons (12, 13, 14, 15) is the first person (12) (S2). The trained model includes at least one of the following: human hesitation behavior, a predetermined bodily expression, a predetermined facial expression, repetition of the same detected behavior, slowing of detected activity, detected staggering behavior, or reclining posture. Regarding the aforementioned task (20), the actions of at least one first person (12) that are causing difficulties are as follows: A computer-implemented method according to claim 1, which is trained to classify.

4. The aforementioned method, Based on the acquired information, the first person (12) among the multiple persons (12, 13, 14, 15) is determined by detecting at least one of the following: work failure, process error, repeated occurrence of work failure, and repeated occurrence of process error (S2). The computer-implemented method according to claim 1, including the method described in claim 1.

5. The aforementioned method, The processor (23) determines, based on a work model using a planning algorithm, the requirements for other persons (13, 14, 15) to support the at least one first person (12) when performing the work (20). The computer-implemented method according to claim 1, including the method described in claim 1.

6. The aforementioned method, To determine the skill level of each other (13, 14, 15), or to determine the support capabilities of each person (12, 13, 14) to support the at least one first person (12) by applying a trained model or by accessing a database (10). The computer-implemented method according to claim 1, including the method described in claim 1.

7. The aforementioned method, Determining the probability of each person (13, 14, 15) supporting the at least one first person (12) by applying a trained model or by determining the availability of each other person (13, 14, 15). The computer-implemented method according to claim 1, including the method described in claim 1.

8. The computer-implemented method according to claim 1, wherein determining the probability of support includes at least one of determining whether other work is currently being performed, in particular whether other work is being performed and whether the other work is urgent, whether it is in a conversation, whether it is classified as hurried, or whether it is classified as angry, and evaluating the cost of the other person(s) (13, 14, 15) to pause the other work and instead perform a support action for the at least one first person(s) (12).

9. The computer-implemented method according to claim 1, wherein determining the at least one second person (13) includes determining the distance from each other person (13, 14, 15) to the at least one first person (12).

10. The computer-implemented method according to claim 1, wherein determining the at least one second person (13) includes determining, for each other person (13, 14, 15) whether it is part of a group of people (15).

11. The computer-implemented method according to claim 1, wherein determining the at least one second person (13) includes determining the support probability for each other person (13, 14, 15) based on the cost for the intelligent agent (11) to approach the other person (13, 14, 15).

12. The computer-implemented method according to claim 11, wherein the cost of approaching the other persons (13, 14, 15) includes determining the distance from each other person (13, 14, 15) to a communication device for contacting the other persons (13, 14, 15), the reachability of each other person (13, 14, 15), the attention of each other person (13, 14, 15) regarding the communication device, in particular the direction of each other person's gaze regarding the communication device.

13. The aforementioned method, The processor (23) plans the actions of the intelligent agent (11) to optimize at least one of the support capabilities and support probabilities of the at least one second person (13), based on at least one of the cost and appropriateness of supporting the at least one first person (12) when performing the task (20) (S4). The computer-implemented method according to claim 1, including the method described in claim 1.

14. The intelligent agent (11) performing at least one of communication or physical action (S5) The text content explaining the need for support when performing the aforementioned task (20) is to be audibly output to at least one second person (13), To visually output text content explaining the need for support when performing the above task (20) to at least one second person (13), The system provides acoustic output of information indicating what action the at least one second person (13) is required to take in order to support the at least one first person (12), To visually output the information of what actions the at least one second person (13) is required to take in order to support the at least one first person (12), The system provides audible output of text content that guides the at least one second person (13) to the at least one first person (12), Visually output text content that guides the at least one second person (13) to the at least one first person (12). Includes at least one of the following: The computer-implemented method according to claim 1, wherein performing the physical action includes the intelligent agent (11) moving to guide the at least one second person (13) to the at least one first person (12).

Images