A visual programming system for automated problem solving
Introduction
Planning has been an active research area since the early days of AI (Weld, 1999), and its applications have been extremely useful in cases when agents act in dynamic environments. Several formalisms for the representation of planning domains and problems have emerged over time, among which Planning Domain Definition Language (PDDL) (Ghallab et al., 1998) has dominated and become a standard. Over the same period of time, numerous algorithms, methods and techniques have been proposed (Hendler, Tate, & Drummond, 1990).
Most of the algorithms and techniques developed so far mainly focus on improving the efficiency of planning systems in terms of required time and resources. Although the results are in many cases impressive, there are not many successful examples of planning systems adapting to industrial use. This can be partly attributed to technical reasons such as the tendency of existing planning systems to be bound to a specific algorithm which might be overwhelmed by the amount of data they have to deal with, as it is becoming more and more immense. This brings on the requirement for systems that have the ability to incorporate multiple state-of-the-art algorithms before they are rendered outdated with respect to current problem demands. Therefore, there is a need not only for more efficient and scalable algorithms, but also for systems that, instead of being limited to a single algorithm, are flexile enough to adapt and employ new algorithms and techniques.
Furthermore, another important issue is the lack of systems that facilitate the development and deployment of planning domains and problems. Modelling and encoding them in some definition language such as PDDL, necessary as it is, can be tiresome and error-prone. It should not be assumed that the designer will always be a planning expert, and even in such cases, visual interfaces can save a lot of time and effort required, while reducing the number of errors and prevent inconsistencies. Visual interfaces in planning systems liberate the designers from the requirement to pay attention to syntax and facilitate focusing on the semantics and structure of the problem at hand.
This paper presents an attempt to approach the design and solving of planning problems while taking into account the aforementioned issues. The result of this research was the development of VLEPPO (Visual Language for Enhanced Planning Problem Orchestration), which is an integrated system for visually modeling and solving planning problems that aims at:
- •
offering a convenient and intuitive graphical interface which simplifies modeling, facilitates maintenance and promotes understanding of planning domains, even for non-expert users;
- •
conforming to the current standards for domain and problem representation, such as PDDL, thus facilitating communication with planning systems that comply with this standard;
- •
providing a high degree of flexibility in integration of different planning algorithms on top of the local one by employing the current web services technology.
The rest of the paper is organized as follows: Section 2 describes some formalisms and standards used for the representation of planning domains and problems, while Section 3 gives an overview of VLEPPO and its architecture. Sections 4 Visual design of planning problems with VLEPPO, 5 Import and export functions, 6 Obtaining solutions to planning problems elaborate on the features of the system, while a case study involving web service composition is described in Section 7. Related work in the area is presented in Section 8, and finally, Section 9 concludes the paper and poses future directions.
Section snippets
Representation of planning domains and problems
This section describes the most prominent formalisms and standards used for the representation of planning domains and problems, namely STRIPS and PDDL. The approach described in this work evolves around these formalisms, and particularly around PDDL, as clear correspondences exist between elements of the system and PDDL elements. In addition, PDDL is used in the VLEPPO system when importing and exporting domains and problems as a means of interoperability with other planning systems.
Overview and architecture of the system
VLEPPO is an integrated system intended to facilitate modeling and solving of planning problems. Among its key features is a convenient, intuitive and easy-to-use graphical interface, which allows design, comprehension and maintenance of planning domains and corresponding problems. The system accommodates for modularity, as domains and problems can be designed separately, and compatibility with standards, as most visual elements present in the system correspond to PDDL elements. Compliance with
Visual design of planning problems with VLEPPO
This section provides a description of the visual elements offered by the system for modeling planning domains and problems, while their correspondence to PDDL elements explained in Section 2.2 is pointed out. The features of the system are described using well-known planning examples.
Exporting to PDDL
The interoperability and increased flexibility of the system would not be possible without compliance with the PDDL standard. Visual elements taking part in domain definition, as well as comments, can be combined to formulate constructs and exported to a PDDL file, which is automatically enhanced with the appropriate requirements tag, as detected by the system. Some details clarifying export are presented in the remainder of the paragraph.
Upon export, the user is offered the option to use
Interface with planning systems implemented as web services
As VLEPPO is intended to be an integrated system not only for designing but for solving planning problems as well, an interface with planning systems is necessary. This is achieved by providing the ability to discover and communicate with web services which offer implementations of various planning algorithms. Moreover, existing planning systems can expose their functionality through web services and be utilized by VLEPPO (Gerevini and Long, 2005, HAP, xxxx).
To this end, a dynamic web service
Case study
In order to illustrate the use of the system, even on alternative domains, a case study concerning a real world problem has been selected, rather than the typical planning domains used in competitions and during the description of the system in the previous sections. The case study concerns web service composition; therefore, a brief introduction to web services along with a discussion about how a web service composition problem can be viewed as a planning problem is provided (Vrakas, Hatzi,
Related work
This section presents the most prominent experimental efforts to construct general-purpose planning tools that have appeared so far.
The GIPO system (McCluskey, Liu, & Simpson, 2003) is based on an object-centric view of the world. The main idea behind it is the notion of change in the state of objects throughout plan execution. Therefore, the domains are modeled by describing the possible changes to the objects included in them. The GIPO system is designed to work with both classical and
Conclusions and future work
This paper presented VLEPPO, a visual system for facilitating the design and maintenance of planning domains and problems through a convenient graphical user interface, as well as obtaining solutions to planning problems utilizing different planners. A high degree of flexibility is provided in cooperating with planners, as they may be either local, or implemented as web services, as the system offers a web service client in order to exploit such functionality. One of the main concerns while
Acknowledgments
This work was partially supported by a PENED program (EPAN M.8.3.1, No. 03ΕΔ73), jointly funded by the European Union and the Greek Government (General Secretariat of Research and Technology).
References (44)
- et al.
STRIPS: A new approach to the application of theorem proving to problem solving
Artificial Intelligence
(1971) - AIPS Planning Competition (2000)....
- Alfredo, M. (2003). Planning and scheduling for the web roadmap. Technical Coordination Unit for Planning and...
- et al.
Web service composition using a deductive XML rule language
Distributed and Parallel Databases
(2005) - et al.
The semantic web
Scientific American
(2001) - Booth, D. et al. (2003). Web services architecture, W3C working draft, August....
- Carman, M., Serafini, L., & Traverso, P. (2003). Web service composition as planning. In ICAPS 2003 workshop on...
- Cox, M. T., & Veloso, M. M. (1997). Supporting combined human and machine planning: An interface for planning by...
- Edelkamp, S., & Hoffmann, J. (2004). PDDL 2.2: The language for the classical part of IPC-4. In Proceedings of the...
- Fox, M., & Long, D. (2002). PDDL+: Modeling continuous time dependent effects. In Proceedings of the 3rd international...
PDDL2.1: An extension to PDDL for expressing temporal planning domains
Journal of Artificial Intelligence Research
AI planning: Systems and techniques
AI Magazine
Cited by (8)
Intelligent planning for allocating containers in maritime terminals
2012, Expert Systems with ApplicationsCitation Excerpt :The goal is that the most immediate containers to load are in the top of the towers, without indicating which containers must be in each tower. We can model our problem by using the standard encoding language for classical planning tasks called PDDL (Planning Domain Definition Language) (Ghallab et al., 1998) whose purpose is to express the physical properties of the domain under consideration and it can be graphically represented by means of tools as (Hatzi, Vrakas, Bassiliades, Anagnostopoulos, & Vlahavas, 2010). A classical AI planning problem can be defined by a tuple 〈A, I, G〉, where A is a set of actions with preconditions and effects, I is the set of propositions in the initial state, and G is a set of propositions that hold true in any goal state.
Integrated intelligent techniques for remarshaling and berthing in maritime terminals
2011, Advanced Engineering InformaticsCitation Excerpt :The goal is that the most immediate containers to load are in the top of the towers, without indicating which containers must be in each tower. We can model our problem by using the standard encoding language for classical planning tasks called PDDL (Planning Domain Definition Language) [7] whose purpose is to express the physical properties of the domain under consideration and it can be graphically represented by means of tools as [10]. A classical AI planning problem can be defined by a tuple 〈A, I, G〉, where A is a set of actions with preconditions and effects, I is the set of propositions in the initial state, and G is a set of propositions that hold true in any goal state.
Interactive visualization in planning and scheduling
2020, Knowledge Engineering Tools and Techniques for AI PlanningA Modelling and Formalisation Tool for Use Case Design in Social Autonomous Robotics
2020, Advances in Intelligent Systems and ComputingEnabling workflow composition within a social network environment
2014, Lecture Notes in Business Information ProcessingBuilding entity relationship models for PDDL and developing planners based on stored procedures
2013, Ruan Jian Xue Bao/Journal of Software