Multi-Agent Programming: Languages, Platforms and Applications. Multiagent Systems, Artificial Societies, and Simulated Organizations, Vol. 15
MULTI-AGENT PROGRAMMING Languages, Platforms and Applications Rafael H. Bordini Mehdi Dastani Jii... more MULTI-AGENT PROGRAMMING Languages, Platforms and Applications Rafael H. Bordini Mehdi Dastani Jiirgen Dix Amal El Fallah Seghrouchni 4y Sprini ... Multi-Agent Programming Languages, Platforms and Applications ' ... Multiagent Systems, Artificial Societies, and ...
Uploads
Papers by Mehdi Dastani
To make this discussion more concrete, consider a personal as- sistant agent that manages a user’s calendar and tasks. One goal the agent may have is booking a meeting. This would typically be modelled as an achievement goal that aims to bring about a state where all required participants have the meeting in their calendar. However, in practice, diaries change, and we want to ensure that the meeting remains in participants’ diaries, and that should a key participant become unable to attend, a new time will be negoti- ated. This is not captured by an achievement goal. Rather, it is better modelled by a combined “achieve then maintain” goal which achieves a certain condition, and then maintains it over a certain time period. Another task that we might want the agent to under- take is to ensure that booking travel is not done until the budget is approved. Note that budget approval may be under the control (or perhaps just influence) of the agent, i.e. the agent may have plans for attempting to have the budget approved. Alternatively, it may be completely outside the agent’s control, in which case the agent can just wait for it to happen and then enable the travel booking process.
A number of papers have taken this line of research a step fur- ther by taking arbitrary Linear Temporal Logic (LTL) formulae as goals [1, 2, 12, 13, 16], rather than considering specific goal types. The advantage of this approach is that it does not restrict the goal types that can be used. However, a possible disadvantage is that it requires extensive alterations of a more basic agent programming framework, the practical implications of which are not yet clear.
Temporal logic is also used by MetateM [10], but it is used directly for agent execution, whereas we use temporal logic as a design framework for specifying goal types which are mapped to existing implementations of achieve and maintenance goals. Addi- tionally, MetateM requires a particular format for its rules: all rules are in one of the three forms: start → φ, or ψ → φ or ψ → φ where φ is a disjunction of literals, ψ is a conjunction of literals, and φ is a positive literal.
In this paper, we propose an approach that is somewhere in be- tween those focusing on a limited set of goal types and those in which arbitrary LTL formulae can be taken as goals. We propose an approach in which goals that are represented by relatively com- plex LTL formulae are operationalized by translating these LTL formulae to more basic achieve and maintain goals. The advantage of this approach is that the goal types can be integrated in existing
Rich Goal Types in Agent Programming
Mehdi Dastani Utrecht University The Netherlands [email protected]
M. Birna van Riemsdijk Delft University of Technology The Netherlands [email protected]
Michael Winikoff University of Otago
New Zealand [email protected]
Goals are central to the design and implementation of intelligent software agents. Much of the literature on goals and reasoning about goals in agent programming frameworks only deals with a limited set of goal types, typically achievement goals, and some- times maintenance goals. In this paper we extend a previously proposed unifying framework for goals with additional richer goal types that are explicitly represented as Linear Temporal Logic (LTL) formulae. We show that these goal types can be modelled as a com- bination of achieve and maintain goals. This is done by providing an operationalization of these new goal types, and showing that the operationalization generates computation traces that satisfy the temporal formula.
To make this discussion more concrete, consider a personal as- sistant agent that manages a user’s calendar and tasks. One goal the agent may have is booking a meeting. This would typically be modelled as an achievement goal that aims to bring about a state where all required participants have the meeting in their calendar. However, in practice, diaries change, and we want to ensure that the meeting remains in participants’ diaries, and that should a key participant become unable to attend, a new time will be negoti- ated. This is not captured by an achievement goal. Rather, it is better modelled by a combined “achieve then maintain” goal which achieves a certain condition, and then maintains it over a certain time period. Another task that we might want the agent to under- take is to ensure that booking travel is not done until the budget is approved. Note that budget approval may be under the control (or perhaps just influence) of the agent, i.e. the agent may have plans for attempting to have the budget approved. Alternatively, it may be completely outside the agent’s control, in which case the agent can just wait for it to happen and then enable the travel booking process.
A number of papers have taken this line of research a step fur- ther by taking arbitrary Linear Temporal Logic (LTL) formulae as goals [1, 2, 12, 13, 16], rather than considering specific goal types. The advantage of this approach is that it does not restrict the goal types that can be used. However, a possible disadvantage is that it requires extensive alterations of a more basic agent programming framework, the practical implications of which are not yet clear.
Temporal logic is also used by MetateM [10], but it is used directly for agent execution, whereas we use temporal logic as a design framework for specifying goal types which are mapped to existing implementations of achieve and maintenance goals. Addi- tionally, MetateM requires a particular format for its rules: all rules are in one of the three forms: start → φ, or ψ → φ or ψ → φ where φ is a disjunction of literals, ψ is a conjunction of literals, and φ is a positive literal.
In this paper, we propose an approach that is somewhere in be- tween those focusing on a limited set of goal types and those in which arbitrary LTL formulae can be taken as goals. We propose an approach in which goals that are represented by relatively com- plex LTL formulae are operationalized by translating these LTL formulae to more basic achieve and maintain goals. The advantage of this approach is that the goal types can be integrated in existing
Rich Goal Types in Agent Programming
Mehdi Dastani Utrecht University The Netherlands [email protected]
M. Birna van Riemsdijk Delft University of Technology The Netherlands [email protected]
Michael Winikoff University of Otago
New Zealand [email protected]
Goals are central to the design and implementation of intelligent software agents. Much of the literature on goals and reasoning about goals in agent programming frameworks only deals with a limited set of goal types, typically achievement goals, and some- times maintenance goals. In this paper we extend a previously proposed unifying framework for goals with additional richer goal types that are explicitly represented as Linear Temporal Logic (LTL) formulae. We show that these goal types can be modelled as a com- bination of achieve and maintain goals. This is done by providing an operationalization of these new goal types, and showing that the operationalization generates computation traces that satisfy the temporal formula.