Periodic Reporting for period 1 - AIPlan4EU (Bringing AI Planning to the European AI On-Demand Platform)
Okres sprawozdawczy: 2021-01-01 do 2022-06-30
Automated Planning and Scheduling is a central research area in AI that has been studied since the inception of the field and where European research has been making strong contributions over decades. Planning is a decision-making technology that consists in reasoning on a predictive model of a system being controlled and deciding how and when to act in order to achieve a desired objective. It is a relevant technology for many application areas that need quick, automated and optimal decisions, like agile manufacturing, agrifood or logistics.
The AIPlan4EU project will bring AI planning as a first-class citizen in the European AI On-Demand (AI4EU) Platform by developing a uniform, user-centered framework to access the existing planning technology and by devising concrete guidelines for innovators and practitioners on how to use this technology. To do so, we will consider use-cases from diverse application areas that will drive the design and the development of the framework, and include several available planning systems as engines that can be selected to solve practical problems. We will develop a general and planner-agnostic API that will both be served by the AI4EU platform and be available as a resource to be integrated into the users' systems. The framework will be validated on use-cases both from within the consortium and recruited by means of cascade funding; moreover, standard interfaces between the framework and common industrial technologies will be developed and made available.
The technical idea is to develop a “Unified Planning Framework” (UPF) as a single-access-point service for using and experimenting with planning technology. Planning engines can be used through the UPF by means of a dynamic plugin system. This means that a user can pose planning queries in an engine-independent way and the UPF is capable of discovering, among the available engines, the ones capable of answering the query and can adapt the query to the specificities of the selected engine by means of the engine-dependent plugin.
We will also realize solutions prototypes for the use-cases elicited within the project. Each use-case can use a specific technology to perform its business logic and there is the need of interfacing the specificities of such technologies to the UPF API/language. Hence, we design and develop Technology Specific Bridges (TSBs) as reusable software components that are used to realize the solution of the various use-cases.
Finally, an overarching objective of the project is to bring this grand view to AI4EU. We aim at providing the components of our architecture as downloadable AI-assets and allowing experimentation with our planning technology and use-cases directly on the platform.
The AIPlan4EU project will bring AI planning as a first-class citizen in the European AI On-Demand (AI4EU) Platform by developing a uniform, user-centered framework to access the existing planning technology and by devising concrete guidelines for innovators and practitioners on how to use this technology. To do so, we will consider use-cases from diverse application areas that will drive the design and the development of the framework, and include several available planning systems as engines that can be selected to solve practical problems. We will develop a general and planner-agnostic API that will both be served by the AI4EU platform and be available as a resource to be integrated into the users' systems. The framework will be validated on use-cases both from within the consortium and recruited by means of cascade funding; moreover, standard interfaces between the framework and common industrial technologies will be developed and made available.
The technical idea is to develop a “Unified Planning Framework” (UPF) as a single-access-point service for using and experimenting with planning technology. Planning engines can be used through the UPF by means of a dynamic plugin system. This means that a user can pose planning queries in an engine-independent way and the UPF is capable of discovering, among the available engines, the ones capable of answering the query and can adapt the query to the specificities of the selected engine by means of the engine-dependent plugin.
We will also realize solutions prototypes for the use-cases elicited within the project. Each use-case can use a specific technology to perform its business logic and there is the need of interfacing the specificities of such technologies to the UPF API/language. Hence, we design and develop Technology Specific Bridges (TSBs) as reusable software components that are used to realize the solution of the various use-cases.
Finally, an overarching objective of the project is to bring this grand view to AI4EU. We aim at providing the components of our architecture as downloadable AI-assets and allowing experimentation with our planning technology and use-cases directly on the platform.
WP1:
Administrative activities, management and documentation
WP2:
Requirements definition in order to identify and describe the planning problem emerging in each use-case and clarify its objective. Support for the evaluation of the open-calls for use-cases.
WP3:
Design and develop of the UPF prototype, making it ready to be integrated with planning engines and TSBs. The UPF is publicly developed and available as an open-source python library.
WP4:
Integration of several planning engines as plug-ins to the UPF. Definition of the test cases repository. Analysis and identification of the targeted planning engines available in the literature.
WP5:
Definition of the general integration principles and a general architecture. Discussion of the design of the various demonstrators for the use-cases emerging from project partners in preparation of the development.
WP6:
Definition of the general methodology for evaluation and requirements for the UPF.
WP7:
Deep collaboration and coordination with the other ICT49 projects, publication of resources in the European AI On-Demnad Platform and preparation of a demo of integration of the UPF within the AI4EU Experiments Platform.
WP8:
Communication platforms. Dissemination of open-calls. Participation to the ICAPS demo session. Write articles related to planning. Promote AIPLAN4EU to international conferences.
WP9:
Helped project partners and external stakeholders in the definition of an exploitation plan for the innovative solutions based on AIPlan4EU.
WP10:
Issued 4 open-calls (out of 5), evlauated them and organized the workshops for the calls for use-cases.
Administrative activities, management and documentation
WP2:
Requirements definition in order to identify and describe the planning problem emerging in each use-case and clarify its objective. Support for the evaluation of the open-calls for use-cases.
WP3:
Design and develop of the UPF prototype, making it ready to be integrated with planning engines and TSBs. The UPF is publicly developed and available as an open-source python library.
WP4:
Integration of several planning engines as plug-ins to the UPF. Definition of the test cases repository. Analysis and identification of the targeted planning engines available in the literature.
WP5:
Definition of the general integration principles and a general architecture. Discussion of the design of the various demonstrators for the use-cases emerging from project partners in preparation of the development.
WP6:
Definition of the general methodology for evaluation and requirements for the UPF.
WP7:
Deep collaboration and coordination with the other ICT49 projects, publication of resources in the European AI On-Demnad Platform and preparation of a demo of integration of the UPF within the AI4EU Experiments Platform.
WP8:
Communication platforms. Dissemination of open-calls. Participation to the ICAPS demo session. Write articles related to planning. Promote AIPLAN4EU to international conferences.
WP9:
Helped project partners and external stakeholders in the definition of an exploitation plan for the innovative solutions based on AIPlan4EU.
WP10:
Issued 4 open-calls (out of 5), evlauated them and organized the workshops for the calls for use-cases.
The project developed (and is developing) a novel platform, called Unified Planning Framework (UPF) providing an abstraction layer for planning services that aims to facilitate and streamline the use of planning technology in practice. The development of the UPF follows an open-source, agile methodology with a public GitHub repository (https://github.com/aiplan4eu/unified-planning) where issues and user-stories are discussed, and the developers discuss publicly about design choices and APIs. For the design of the UPF we followed user-centered design principles, where the various features are backed by user-stories detailing the intended use and serve as informal documentation of the requirement. The UPF is incarnated as an open-source Python3 library, which offers a unified API for the specification and manipulation of planning problems and integrates a number of planning engines (integrated by the project and available at https://github.com/aiplan4eu) which offer planning technology services. The library has been developed following the best practices for open-source python development and offers a comprehensive API for classical, numeric, temporal and HTN planning problems. Extensions to model task-and-motion planning and multi-agent planning problems are underway. The UPF is well documented at code level and we created a number of Python Notebooks to showcase its use. In the next period, we will work on the planning “cookbook” creating a wiki for guiding prospecting users and innovators in the use of planning technology.
We worked for the integration of the UPF and other planning materials in the European AI On-Demand Platform (see https://www.ai4europe.eu/ai-community/projects/aiplan4eu) in two main directions, namely the publication of planning related resources in the AI on demand platform and the realization of an AI4EU Experiments pipeline showing how to use planning technology via the UPF in the AI4EU Experiments Marketplace and AI4EU Playground. The latter is still not available in the public release of the platform, but only in the beta version. This first demo will serve as a template for all the use-cases we will demonstrate on the platform: the idea is to provide a page for each demonstration in the platform CMS and a pipeline that, when executed either on the Playground or at the user premises, will provide an interactive demo of the planning technology integrated in the specific use-case. This will give a first-hand view of the usefulness and applicability of planning technology to the platform users.
The development of reusable components interfacing the UPF with industrially-relevant technologies starts in the next period: so far, we focused on the elicitation of use-cases from all the partners and from the open-call winners and in the definition of a general, shared architecture for the development of demonstrators for the use-cases provided by partners of the consortium. Some use-cases already developed some preliminary code for internal experimentation and validation of the UPF.
We worked for the integration of the UPF and other planning materials in the European AI On-Demand Platform (see https://www.ai4europe.eu/ai-community/projects/aiplan4eu) in two main directions, namely the publication of planning related resources in the AI on demand platform and the realization of an AI4EU Experiments pipeline showing how to use planning technology via the UPF in the AI4EU Experiments Marketplace and AI4EU Playground. The latter is still not available in the public release of the platform, but only in the beta version. This first demo will serve as a template for all the use-cases we will demonstrate on the platform: the idea is to provide a page for each demonstration in the platform CMS and a pipeline that, when executed either on the Playground or at the user premises, will provide an interactive demo of the planning technology integrated in the specific use-case. This will give a first-hand view of the usefulness and applicability of planning technology to the platform users.
The development of reusable components interfacing the UPF with industrially-relevant technologies starts in the next period: so far, we focused on the elicitation of use-cases from all the partners and from the open-call winners and in the definition of a general, shared architecture for the development of demonstrators for the use-cases provided by partners of the consortium. Some use-cases already developed some preliminary code for internal experimentation and validation of the UPF.