Abstract An approach to intelligent process automation based on distributed planning agents is presented in this paper. According to this approach a higher-level agent-based automation layer supervises an ordinary process automation system. The purpose of the agent layer is to monitor the operation of the lower-level automation system and semi-autonomously reconfigure its control logic when needed. The agent layer operates as a distributed planning and plan execution system, which creates and runs reconfiguration sequences on the ordinary automation system in order to adapt it to various situations. In this way the approach aims to increase the operational flexibility of the whole automation system. The approach is demonstrated with a laboratory test process where process startup scenarios have been imitated. Experiences from the initial test runs are also described. 1 Introduction The research effort described in this paper is motivated by two factors: the gradually increasing functional requirements of process automation and the progress of agent technology. Although the requirements of process automation have traditionally focused on reliability, efficiency and quality [14], recently operational flexibility and easier system maintenance have been emphasized as well [20]. Agent technology [24] has been studied as possible means to reach such objectives as indicated by the research in other application domains, e.g. communication systems [7]. Our research objective is to specify a reasonable way to utilize agent technology in order to build a process automation system with the desired properties of enhanced flexibility and easier system maintenance. The purpose of this paper is to present an approach to intelligent process automation based on distributed planning agents. In this context the purpose of intelligence in process automation is to gain increased flexibility in process control operations and at the same time to enhance the maintenance of process automation systems. Our approach to intelligent process automation is based on multi-agent systems (acronym: MAS) performing distributed planning. We expect that the methods of the planning type of MAS will help with the presented objectives. The functional responsibility of the agents is to plan and execute higher-level control sequences, e.g. startup, shutdown, state change, batch process and abnormal situation handling procedures. This study complements our previous experiments with process automation agents based on a reactive operation scheme [16][17]. These experiments led us to a conclusion that agents with purely reactive operation scheme are only applicable to a limited set of process automation operations. A suitable combination of planning and reactive operation is needed for extended functionality. The paper is outlined as follows. Chapter 2 will discuss intelligent process automation, planning agents and their applications to process automation. Our concept of an agent-augmented process automation system containing planning agents is presented in Chapter 3. The distributed planning and plan execution processes of the system are described in Chapter 4. The test environment and an experiment with it are depicted in Chapter 5 followed by conclusions in Chapter 6. 2 Planning Agents and Intelligent Process Automation Intelligent supervisory control has been a research topic in process automation for already quite a time; see e.g. [22]. The basic idea of this concept is to extend process controllers with reasoning capabilities of artificial intelligence. The intelligence is in a supervisory controller that guides the operation of an ordinary primary controller. Experimented inference techniques include e.g. rule-based reasoning, fuzzy logic and qualitative modeling. However, regardless of the inference mechanism the model of operation in these approaches has been centralized. Distributed planning is one of the research topics of distributed artificial intelligence [24]. Several methods for planning in different situations with varying degree of Distributed Planning Agents for Intelligent Process Automation Ilkka Seilonen 1 , Teppo Pirttioja 2 , Pekka Appelqvist 2 , Aarne Halme 2 , Kari Koskinen 1 Helsinki University of Technology 1 Information and Computer Systems in Automation, 2 Automation Technology Laboratory Espoo, Finland 1 www.automationit.hut.fi, 2 www.automation.hut.fi 614 Proceedings 2003 IEEE International Symposium on Computational Intelligence in Robotics and Automation July 16-20, 2003, Kobe, Japan 0-7803-7866-0/03/$17.00 ©2003 IEEE