Coddington, Alexandra Magrit; (2001) Self-Motivated Planning in Autonomous Agents. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Self-motivated planning in autonomous agents.pdf Download (5MB) | Preview

Abstract

This thesis describes research which is concerned principally with the design of a planning/execution architecture to be used within an autonomous motivated agent situated within a real world environment. To further its aims, such an agent is capable of generating its own goals and of planning and acting to achieve those goals in real or simulated time. Research in planning tends to assume that goals are externally generated by a human operator and that planning is complete once all outstanding goals have been achieved. Because our agent is able to continually generate its own goals, planning may never be complete which means planning and execution are ongoing activities. In addition, constraints upon time may mean it is not possible for the agent to achieve all goals. The agent must therefore be able to both reason about time (in particular the durations of actions) and prioritise its goals. A prototype planning/execution architecture designed to address these issues is described. The proposed architecture extends the classical planning framework to take into account both the context of the agent (where context may be interpreted as a function of both the perceived external environment and the internal state of the agent), and problems associated with planning and acting in real time. We argue that context, captured by modelling the motivations of an agent situated within an environment, plays an important role in the generation of goals and enables the agent to determine the importance of such goals. A crucial component of the planning/execution architecture is a temporal manager which enables the agent to reason about whether or not there is sufficient time available to execute all of the actions within a partial plan and to calculate deadlines for actions and outstanding goals. The importance and deadlines associated with goals enables the agent to prioritise those goals so that should there be insufficient time to achieve all goals, the agent can abandon some goals in favour of others. The architecture also demonstrates how modelling the motivations of an agent provides an effective means for evaluating and selecting partial plans. Finally, the architecture enables the agent to "execute" actions and to cope with unpredictable changes that may occur within its environment.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item