Planning Services for Individuals: A New Challenge for the ... - UoM

Planning Services for Individuals: A New Challenge for the Planning Community Ioannis Refanidis1, Thomas Leo McCluskey2 and Yannis Dimopoulos3 1

University of Macedonia, Dept. of Applied Informatics, Thessaloniki, Greece University of Huddersfield, School of Computing and Engineering, Huddersfield, United Kingdom 3 University of Cyprus, Department of Computer Science, Nicosia, Cyprus E-mails: [email protected], [email protected], [email protected]

2

Abstract This paper describes preliminary work on an innovative application, intelligent mobile electronic assistants (IMEA), based, on its core, on AI planning/scheduling technology and task ontologies for planning. IMEA aims at assisting people to organize and groups of people to coordinate their everyday tasks. Its target group consists of people that have very tight personal daily plans with a variety of activities (e.g. managers, academics, etc) and have difficulties in remembing and organizing them. Its main objectives are to save user’s resources (time, money) and achieve better utilization of common resources. However, its side-effects could be to popularize artificial intelligence planning and knowledge engineering technology, thus increasing commercial and governmental interest on them. IMEA is based on a distributed open platform with a variety of devices (mainly wireless ones) for its user interface needs, running over the internet. Apart from AI planning and knowledge engineering, IMEA is based on technologies such as semantic web and intelligent agents.

Introduction Nowadays many people use electronic organizers, such as MS-Outlook, to schedule their tasks and meetings. These programs usually require from the user to manually enter her tasks or meetings, as well as details such as start/end time, recurrence, importance, execution status and reminder settings. Moreover, some programs allow the coordination with other people in order to arrange meetings; however, this again is based on manual communication between users, usually through e-mails. Users benefit from electronic organizers mainly in two ways: • They can identify conflicts between tasks or meetings and solve them by manually rearranging conflicting activities. • They are notified about upcoming tasks. The usability of modern electronic organizers has been improved with the advent of mobile devices and their ability of synchronization with desktop computers. However, they still do not exhibit any form of intelligent behavior. Thus, we could say that they do not offer any

additional functionality compared to traditional, paperbased organizers. In this paper we describe preliminary work on a new generation of electronic organizers, named intelligent mobile electronic assistants (IMEA), which extends the functionality of existing electronic organizers in the following ways: • They maintain task ontologies that help the user in managing her task list. • They have planning and scheduling abilities, in order to schedule the tasks in an efficient manner. • They cooperate with intelligent agents, in order to automatically or semi-automatically accomplish tasks. • They support automated coordination with other users, in order to reduce the burden of communication between them. • They manage common resources, trying to increase their utilization. Planning and scheduling technology lies in the core of IMEA. The system has to examine several alternative plans that result from different ways of grounding abstract tasks. Also, it may be necessary to insert new tasks in the user’s task list, in order to satisfy open preconditions of existing tasks. On the other hand, scheduling is necessary in order to produce efficient plans that satisfy all constraints. Several architectures for solving this planning and scheduling problem could be adopted, such as a traditional first-plan-then-schedule schema or more elaborated interleaving planning and scheduling approaches. IMEA is based on a distributed open platform running over the internet, consisting of planning servers and coordination servers. Various forms of interface devices, such as PDAs, smartphones or laptops could be used for the user-interface needs. IMEA’s target group consists of people that have very tight personal daily plans with a variety of activities (e.g. managers, academics, etc) and have difficulties in remembing and organizing them. Its main objectives are to save user’s resources (time, money) and achieve better utilization of common resources. However, its side-effects could be to popularize artificial intelligence planning and

knowledge engineering technology, thus increasing commercial and governmental interest on them. The rest of the paper is as follows: First we present some illustrative scenarios of potential IMEA’s use, then we present its architecture and the overall approach, then we identify scientific and technological challenges and its potential impact, subsequently we present related work and finally we conclude the paper.

Functionality and Scenarios IMEA’s intended functionality is summarized in the following steps: • The user may enter her tasks using a variety of interface devices, such as Personal Digital Assistants (PDAs), SmartPhones, her laptop etc, probably powered with location identification capability. Tasks are selected from a predefined task ontology through a friendly user interface, however the user will have the possibility to define her own tasks and position them within the ontology. Task descriptions include details such as execution conditions (preconditions), duration, calendar constraints, potential decomposition in subtasks, location information, resource usage, need for cooperation with other users etc. • IMEA produces efficient plans wrt user preferences (e.g. time, cost, preference on having specific intervals - like evenings or weekends - free etc). IMEA may add new tasks in the user’s plan, in case this is necessary to satisfy executability conditions. For example, if the user’s task list has two consecutive tasks with location references A and B respectively, the system should plan a ‘move(A,B)’ action between the above two tasks. To handle such cases the system should be able to query a GIS server to obtain information about the distance between different locations, the time needed to travel between them etc. • Using intelligent agents, IMEA accomplishes specific tasks, whenever this is permitted by the user and a suitable agent is available. • IMEA handles requests for coordination with other users by synchronizing their plans, if possible. The system also manages common resources (e.g. the Meeting Hall of a company). • The solution (i.e. the personal plan) produced by IMEA is transmitted back to the user’s interface device. The user has the ability to modify her personal plan, by adding/deleting tasks or changing their properties. In case of inability to solve the problem, the user is asked to relax the tasks’ constraints. • IMEA monitors the execution of the plan, by taking into account the location of the user or asking for user intervention. In case of execution failure, replanning is initiated. Let us next present a non-exhaustive set of potential use cases of IMEA (the Greek name ‘Demitrios’ has been

adopted by the scenarios for ambient intelligence in 2010 (ISTAG, 2001) of IST Advisory Group - ISTAG) . Use case 1: Demitrios is the coordinator of a working group of the Department of Informatics, with the task to reformulate the department’s syllabus. The group has to meet once a month for the next six months. Using his PDA, Demitrios selects the task “Meeting” from the task hierarchy, he characterizes it as a periodic task with a monthly period and six repetitions, he selects the colleagues that should participate in this meeting. He estimates the duration of the meeting at 2 hours and accepts the default value “Meeting hall” for the meeting location. IMEA schedules all meetings, taking into account the personal plan of Demitrios as well as of the other participants, the availability of the meeting hall and the preferences of all involved people wrt to the time of the meeting. After finding time-slots for the meetings, the personal plans of all involved people, as well as the availability of the meeting hall, are updated accordingly. Additional tasks concerning preparation for the meeting (e.g. slides) and transportation are inserted in their plans as well. Note that future allocations (e.g. the last meeting) might change, since new tasks that might emerge in between could pose additional constraints in existing ones. Use case 2: The first meeting was scheduled for 10:00 am of this morning. However, due to the traffic congestion, one participant was unable to be there on time. IMEA, through the localization system of her PDA, determines that the participant will be late, and shifts the meeting to 11:00 am (hopefully, there was no conflict). All plans, as well as the availability of the meeting hall, are revised and the participants are notified accordingly. Use case 3: Demitrios wants to participate in a conference. He selects the abstract task “Conference participation” from the task ontology. The intelligent agent associated with this task asks for the conference’s URL. Hopefully, the site offers its functionality through a web service, so the intelligent agent retrieves critical dates and automatically inserts the necessary sub-tasks into Demetrios’ task list. Moreover, it automatically performs an early registration, using secure communication. Use case 4: One of the tasks implied by Demitrios’ interest in this conference is writing a paper. Demitrios estimates that he needs 40 hours for this task. IMEA allocates this amount of time, by splitting it in 5 portions of 8 hours each and in two fragments of consecutive days. For these days IMEA does not plan any other task (the system knows that Demitrios – as any other user – does not want to be distracted by other tasks when he is concentrated on intellectual activities). Use case 5: IMEA updated the profile of Demitrios based on the research topics of the conference. A few months later, IMEA locates the site of another conference with similar topics and proposes to Demitrios to insert again the task “Conference participation” in his task list. Demitrios accepts this proposal.

User-1

User-4

User-2

Planning Server 1

Planning Server 2 User-3

Planning Server 3

Internet Coordination Server A

Coordination Server B

Figure 1: The general architecture of the IMEA platform

Architecture Figure 1 presents the general architecture of IMEA, which consists of the interface devices, the planning servers and the coordination servers. IMEA will support a variety of interface devices, including PDAs, SmartPhones and desktop/laptop computers. Interface devices are dummy; they just send and receive data. They communicate with the planning servers using common protocols and wired or wireless connections (e.g. Bluetooth, Wi-fi, GPRS etc). Planning servers constitute the heart of the system. Their functionality is the following: • Provide an updateable task ontology • Store the user’s profile • Host intelligent agents • Communicate with the user’s interface device • Solve the planning problems The rationale of employing planning servers (instead of the interface devices themselves) to store information and solve the planning problems is to enable the user to use a variety of interface devices, without the need to synchronize them to each other. On the other hand, not all types of interface devices provide the necessary resources (e.g. processing power, storage) for the above tasks. All user-relevant data, such as her task list or her profile, are stored in the planning servers. The user can access her

planning server using a usual secure login procedure. A planning server can host many users. Coordination servers handle requests from the planning servers for synchronization between users (e.g. meetings) and for common resource allocation. Each user and common resource is registered in a single coordination server, thus having a universally unique identity. Coordination servers are widely distributed across internet and communicate to each other on a peer-to-peer basis. It is possible for coordination and planning servers to be physically hosted in the same machine. However, hosting planning and coordination servers in different machines is preferable for two reasons: • Solving planning problems may be a very hard job from a computational point of view, so dedicating one planning server to one (or at most a few) user increases the efficiency of the whole system. • It is recommended for planning servers to be private machines, i.e. the user’s desktop computer. This increases the protection of sensitive private data (e.g. the user’s profile or her plan), which are stored in the planning server, and consequently the system’s acceptation by end users. Note that, in case of multiple users using the same planning server, the server may be linked statically to various coordination servers (like several e-mail accounts handled by the same client program).

Overall Approach In this section we will concentrate on the planning and scheduling aspects of the IMEA system. Of course there are numerous other issues to be solved, such as friendly and adaptable user interfaces, networking, communication protocols, security issues etc, but these are out of the scope of this paper. Solving the planning problem is the most critical task of the system. There are mainly two types of tasks: • Simple tasks, like ‘Visit to the dentist’, which are indivisible activities. The user has to describe properties such as duration, location, resource demands and calendar constraints. • Time-demand tasks, like e.g. ‘Reading a book’. These tasks demand a prespecified amount of time, e.g. 30 hours. They can be fragmented and accomplished in smaller portions. In addition to the simple tasks, the user can set constraints on the minimum and maximum size of these portions or the presence of other tasks in between them. In addition to the above general types of tasks, a task may be characterized as periodic, i.e. a task that is repeated on a regular basis, like e.g. a weekly lecture. For periodic tasks the user has to determine the number of the repetitions or the end date; infinite repetitions are allowed as well. Furthermore, a task might be characterized as a coordination task, i.e. a task where other people are involved, like a business meeting. In this case the task has to be scheduled for all involved users simultaneously, updating their individual plans. Finally, any task might be an abstract task. Abstract tasks can be decomposed into a partial ordered set of subtasks. There might exist several alternative ways of decomposition. Abstract tasks enhance the usability of the system, in that they reduce the data-entry requirements for the user and let the planner to decide the more efficient way to accomplish the task. Calendar constraints impose constraints on the allowable days and times a task can be executed. The user will be able to determine specific days and hours within these days (possibly different hours for different days), where the task can be scheduled. In order to reduce the burden of information entry, the intersection of a variety of higher level declarations, such as earliest start time and latest end time, days of week and hours of a day could be used. The user will be able to express her preferences on a variety of ways. There are several objectives that should be maximized, such as time, cost, user satisfaction (e.g. preference on having specific time-intervals free) etc. The exact way of modeling and aggregating preferences remains an issue. One approach could be to ask the user to set tradeoffs between alternative objectives, project them on a common scale and use utility based aggregation methods (Keeney and Raiffa, 1976). Other more sophisticated methods, such as outranking methods (Roy,

1991), need pair wise comparisons and are are harder to compute due to their exponential requirements. The initial planning engine will be based on hierarchical task network algorithms (HTN) with causal link information on all levels of abstraction, enhanced with constraint satisfaction and local search algorithms and powered by a bundle of heuristics. HTN partial-order planning and constraint satisfaction techniques will be used to quickly find an initial, sub-optimal plan. Local search algorithms will be used to gradually improve the quality of the initial plan. Periodic tasks will be handled by considering each individual repetition as an independent task. Infinite periodic tasks need special treatment, as e.g. considering only an initial finite number of repetitions (actually, in real life there are no infinite tasks for human beings) or performing a loose (e.g. pairwise) constraint satisfiability test among the infinite tasks. We believe that periodic tasks will not constitute a hard problem from a computational point of view, since the user’s task list in the mid- or long-term is not expected to be tight enough. Coordination tasks will be based on finding the intersection of free intervals for all involved people and selecting one that maximizes their average satisfaction. However, local search algorithms could again be employed, since rearranging already scheduled tasks of the users might be proved quite beneficial for all them. Finally, common resources management will be performed on a first-come-first-served basis. however, subsequent requests for allocating the same resource for the same time period will also be taken into account, especially in the case where an alternative allocation is accepted by the ‘first’ user. Tasks can be accompanied by intelligent agents, which will be launched to perform tedious jobs, like seeking information or using web services. IMEA will also retain a profile for each end-user that will be used to assign default values to the task attributes, as well as to propose new tasks to be inserted in the user’s task list. The system could also employ machine learning techniques to acquire knowledge about the user’s preferences, thus updating her profile. Special attention will be given to the open architecture of the whole system. The interface specifications of all system’s components will get distributed, thus allowing third organizations and private companies to develop their own task ontologies, user interface modules, planning and coordination servers, intelligent agents etc. Existing standards (as e.g. RDF/RDFS, DAML+OIL for the ontologies) will be used whenever possible for greater interoperability with existing protocols and systems.

Challenges and Impact IMEA involves research in four directions: • Planning and scheduling technology: IMEA requires advances in the following aspects of planning technology: o Multiobjective planning, with a variety of preferences and methods to aggregate them.

o Coordination planning, where plans of different agents need to get synchronized. o Planning with new types of tasks, such as periodic tasks and time demand tasks. o New types of constraints. The problem of selforganization introduces new types of constraints, which demand new algorithmic solutions. For example, a simple task might be scheduled on specific days of a week and specific time-intervals of these days (possibly different for the various days), whereas it has to be completed before a deadline. On the other hand, a time-demand task might have constraints on the minimum and maximum size of the individual fragments or on the minimum number of consecutive fragments. • Knowledge engineering for planning: There are research problems that need to be overcome, such as knowledge engineering of an initial task ontology for particular user groups, its configuration to a particular user, and its adaptation to the user's needs over time. These will involve the development of a generic, structured HTN task definition language that has builtin primitives for this kind of application area based on a 'calendar-oriented' temporal ontology. The definition language will have to represent information gathering tasks, as well as events and possibly processes. It will require translators to produce a user friendly form for viewing tasks and plans, and a method, likely to be based on some form of inductive technique, for the user to define new tasks in a non-mathematical fashion. • Mobile user-interface for planning technology: Currently, most of the planning systems use text files for their input/output needs (Long and Fox, 2003). There are just a few attempts to create generic user interfaces for planning, but they are still very basic (McCluskey et. al, 2003). IMEA involves research on: o User interfaces for planning, supporting task ontologies, constraints, user preferences, alternative views of plans, user profiles etc. o Adaptable planning user interfaces for various types of mobile and non-mobile devices. • Protocols: A variety of protocols have to be implemented, in order to assure a seamless communication between the various modules of the system. Specifically, protocols have to be defined for the interface between: o Interface devices and planning servers o Planning serves and coordination servers o Between coordination servers o Planning servers and intelligent agents IMEA has the potentiality to reinforce interest in the research area of planning and scheduling algorithms. In case of acceptation of intelligent mobile electronic assistants by end-users, companies would likely invest in developing better algorithms for planning and scheduling. On the other hand, the new technology could also be adapted to other application domains, such as rescue

planning, space-mission planning or the planning of elderly-care. Apart from AI planning, IMEA has the potentiality to reinforce interest in the development of intelligent agents that will help users to automatically accomplish their tasks. Furthermore, the creation of a critical mass of intelligent agents would also push forward the proliferation of web services. However, IMEAs main effect should be expected from the benefits from its real use. Our personal experience after a long time of real use of modern, non-intelligent electronic organizers to fulfill self-organization needs suggests that: • Existing digital organizers significantly improve selforganization. • Users are more confident that they are fulfilling their obligations. • After long use of digital organizers, users acquire new practices such as e.g. consulting their digital assistant at regular times. • Users tend to increasingly rely on their digital assistant, thus they stop thinking of non-significant details. • Some times, users are not sure whether specific tasks have been inserted in their task list, which might cause inconvenience to them. • Users tend to overload their task list with nonsignificant tasks. • Users tend to give priority in non-significant but short tasks, just to shorten their task list. IMEA attempts to rationally organize the tasks, giving priority to the most important ones, whereas simultaneously trying to efficiently schedule as many as possible from the less significant ones. However, its acceptance by end users, conditioned on the systems usability, efficiency and privacy protection, remains an issue.

Related work During the last years researchers have focused on organizing people and coordinating groups of people, using planning, scheduling and intelligent agents technologies. In the following paragraphs we briefly present three representative research efforts and pinpoint their differences from IMEA. The I-X architecture at University of Edinburgh aims at cooperation between human and computer systems in the synthesis and modification of a product, such as a plan, design or physical entity (Tate, 2000). Two aspects of I-X are relevant to intelligent mobile assistance applications: IP2 and I-Plan. I-P2 (I-X Process Panel) aims to act as a workflow aid, providing users with reporting and messaging. I-Plan is a mixed-initiative planning system based on plug-ins; this means that extensions to the basic planning engine are needed in order to use the system in other domains. IMEA differs from I-X architecture in that it concentrates mainly on the single user rather than on cooperative environments. This is achieved by providing a

single planning engine, which allows the users to define their own tasks in a declarative way, without the need for modifications. On the other hand, IMEA will support coordination between people from different organizations, through communication/negotiation between the coordination servers. Finally, IMEA will possess a rich task ontology with various task types, a rich set of declarative constraints and a variety of user preferences. Autominder (Pollack et.al., 2003) focus on the design of an autonomous robot for assisting an elderly client in carrying out the required activities of daily life by providing her with timely and appropriate reminders. In generating these reminders, the goal is to balance three objectives: (i) maximizing the client’s compliance in performing activities of daily life; (ii) maximizing the level of caregiver and client satisfaction with the system; and (iii) avoiding making the client overly reliant on the system. In Autominder, the caregiver initially inputs a description of the activities, as well as any constraints on, or preferences regarding the time or manner of performance. Updates to the plan can be made by the caregiver, or with certain restrictions, by the client herself. Apart from having totally different target groups, IMEA differs from Autominder in that it provides a rich task ontology with various task types, it allows for coordination between users and management of common resources and it uses mainly wireless personal digital assistant for interface needs. Finally, the Electric Elves project (Chalupsky et. al., 2003) applies agent technology in service of the day-to-day activities of human organizations. The system consists of agents deployed into human organizations to help with organizational tasks. The agents are tied to individual user workstations, fax machines, voice, mobile devices such as cell phones and palm pilots. The Electric Elves agents assist humans in routine tasks, such as rescheduling meetings, selecting presenters for research meetings, tracking people's locations, organizing lunch meetings, etc. Moreover these agents can interact directly with humans both within the organization and outside the organization by using email, wireless messaging, and faxes. The agents ensure that activities within an organization run smoothly and that events are planned and coordinated to maximize the productivity of the individuals of an organization, thus performing a level of monitoring that would be impractical for human assistants. IMEA differs from Electric Elves in that it is mainly focused on a human rather than on an organization. Plans are produced and optimized in order to satisfy individual needs and preferences. Of course, coordination between users is also supported, so in this way IMEA can also serve the needs of a company. Actually, intelligent agents similar to those of the Electric Elves project could accompany task descriptions of IMEA. Finally, IMEA is based on a task ontology and not on monitoring agents, thus making easier for end-users to define their own tasks.

Conclusions This paper presented preliminary work on intelligent mobile electronic assistants (IMEA), powered with planning and scheduling capabilities. We proposed a distributed architecture, with high-performance planning servers being used to store user profiles and task ontologies, as well as to solve the planning problems, coordination servers being used to synchronize plans of different users and manage common resources and, finally, interface devices (mainly mobile) being used for communication with the user. The main goal is to help users to organize and groups of people to coordinate their every day tasks, in a resource-efficient manner. As far as we known, IMEA is the first attempt to bring planning technology in the service of everyday people. Thus it has the potentiality to reinforce commercial and governmental interest in planning and scheduling technology, as well as in knowledge engineering for planning. In any case, the success or failure of this project will be determined by its acceptance by end-users, conditioned on criteria such as efficiency, usability, privacy etc.

References Chalupsky, H., Gil, Y., Knoblock, C., Lerman, K., Oh, J., Pynadath, D., Russ, T., and Tambe, M. Electric Elves: Agent Technology for Supporting Human Organizations. AI Magazine, 23 (2), 2002, 11-24. ISTAG Report. Scenarios for Ambient Intelligence in 2010. February 2001, http://www.cordis.lu/ist/istag.htm. Keeney R.L., and Raiffa H. Decision with multiple objectives. John Wiley, New York (1976). Long, D., and Fox, M. The 3rd International Planning Competition: Results and Analysis. Journal of Artificial Intelligence Research, 20 (2003), 1-59. McCluskey, Y.L., Liu, D. and Simpson, R.M. GIPO II: HTN Planning in a Tool-supported Knowledge Engineering Environment. Proc. of the International Conference on Automated Planning and Scheduling (ICAPS), 2003 , AAAI Press. Pollack, M.E., Brown, L., Colbry, D., McCarthy, C., Orosz, C., Peintner, B., Ramakrishnan, S., and Tsamardinos, I. Autominder: an intelligent cognitive orthotic system for people with memory impairment. Journal of Robotics and Autonomous Systems, 44 (2003), 273-282. Roy B. The outranking approach and the foundation of ELECTRE methods. Theory and Decision, 31 (1991), 4973 Tate, A. Intelligible AI Planning, in Research and Development. Proc. of 20th British Computer Society Special Group on Expert Systems International Conference on Knowledge Based Systems and Applied Artificial Intelligence, 2000, pp. 3-16