DAI-List Digest Wednesday, 20 February 1991 Issue Number 26 Topics: DAI Research at AAII Please send submissions to DAI-List@mcc.com. Send other requests, such as changes in your e-mail address, to DAI-List-Request@mcc.com. ------------------------------------------------------------------------ From: Anand Rao Date: Wed, 20 Feb 91 09:12:40 EST Subject: DAI Research at AAII The Australian Artifical Intelligence Institute (AAII) is involved in various aspects of DAI research, including theoretical foundations, system implementations, experimental simulation, and applications of DAI. THEORETICAL FOUNDATIONS Research in this area is focused on the formalizations of a single agent's mental state in terms of his beliefs, goals, actions, intentions and commitment. The role played by decision theory in the deliberative process and its influence on the formation and revision of intentions and goals is also being examined. An extension of this enterprise is the formalization of the mental state of a group of agents. This involves notions such as mutual beliefs, joint goals, social actions, group intentions and joint commitment. The latter is being examined with respect to different organisational structures. With the above formalizations we are developing representations for global multiagent plans. The primary aim of this work is to develop techniques for collaborative planning among a group of agents and the associated communication and synchronisation required. SYSTEM IMPLEMENTATION The collaborative planning system is to be based on the existing embedded real-time reasoning system -- the Procedural Reasoning System (PRS). The philosophy of PRS, namely that agents do not plan from scratch but rather carry situation-specific plans that are instantiated and executed partially at the appropriate time, is carried through in the multiagent case. Thus, groups of agents have global plans which are adopted and executed when the group has the appropriate mutual beliefs and joint goals. The system will be written in Common Lisp. EXPERIMENTAL SIMULATION Recent work has employed a simulated environment to explore the interaction between agent and environment characteristics, concentrating particularly on the role of intentions and commitment in producing effective agent behaviour. This approach will be extended to multiagent domains, permitting experimental investigation of the effects of different organisational structures and reflective and cooperative strategies upon individual and group effectiveness. APPLICATIONS Some of the above DAI research will be applied to real-world domains. In particular work will be carried out on the following two projects: Distributed Air-Traffic Management System ----------------------------------------- The aim of this project is to develop a prototype distributed reasoning system that will coordinate the air-traffic within a group of airports and the associated air-space. Thus the air traffic management systems at the different airports have to cooperate with each other and revise their respective schedules so as to globally maximize the utilisation of the entire air-traffic network. The project is an extension of work currently being done for the Civil Aviation Authority of Australia on a single airport air-traffic flow management system. Beyond-Visual-Range Air Combat Modelling ---------------------------------------- The aim of this project is to model individual and team tactics of combat aircraft under different scenarios; namely, varying opposition team tactics and weapons capabilities and varying the certainty/uncertainty of the information available. The emphasis will be on team tactics along with the dynamic reorganisation of teams and the dynamic change of roles among the team members. This is a joint project with the Australian Defense Science and Technology Organisation's Aeronautical Research Laboratories. Apart from the above DAI projects which have recently commenced, the Institute has been working on multiagent embedded real-time systems over the past couple of years. Although not strictly DAI, (because the global communication/ synchronisation/negotiation is hard-wired and developed by the application developers) the following projects use a multiagent architecture that might be of some interest to the DAI community. OASIS (Optimal Aircraft Sequencing Using Intelligent Search) ------------------------------------------------------------ This prototype system is designed to perform air-traffic flow management for Australia's Sydney airport. It monitors the progress of aircraft destined to Sydney, sequences them based on different criteria and operating conditions, and issues directives to the controllers to implement these sequences. It monitors these sequences in real-time and revises them when necessary. Thus it is a real-time embedded reasoning and scheduling system. OASIS is based on a multiagent architecture with one aircraft agent for each aircraft (approximately 50 aircrafts at any one time) and a number of global agents including the sequencer, wind manager, coordinator, and trajectory checker. While each aircraft agent models the flight of an individual aircraft, the global agents are responsible for the global tasks of sequencing, feedback and monitoring. The system is currently running under a simulated environment and is expected to be trialed at Sydney airport in June 1991. IRTNMS (Interactive Real-Time Network Management System) -------------------------------------------------------- This prototype system developed for Telecom Australia manages the telephone network in real-time. It receives raw data about different parameters from the network in real-time and diagnoses different types of problems. The system then suggests and implements different control measures to reduce or remove these problems. It also continuously monitors these control measures suggesting the increase, decrease or removal of various control measures. The system is once again based on a multiagent architecture with three agents: the DIAGNOSIS agent responsible for diagnosing problems, the CONTROLS agent responsible for taking control measures, and the MONITOR agent responsible for monitoring the status of the network and the control measures. All three agents operate in parallel, with possibly different sets of data, and share information by message passing. Real-Time Diagnosis for the Space Shuttle's Reaction Control System (RCS) ------------------------------------------------------------------------- The RCS system was developed for NASA jointly at SRI International and AAII. It is an embedded reasoning system which monitors alarms, diagnoses problems, and suggests corrective measures for the reaction control system of the Space Shuttle. The system has two agents: the INTERFACE agent, which performs the low-level monitoring of temperatures and pressures and the RCS agent, which is responsible for the high-level diagnosis. Anyone interested in obtaining more details about the above projects can get in touch with me. Some of the above mentioned work has been published in conferences and the others are at different stages of development. Anand S. Rao Australian Artificial Intelligence Institute 1 Grattan Street Melbourne, Victoria-3053 AUSTRALIA Email: anand@aaii.oz.au