A Logical Approach to High-Level Agent Control

Recent work in animated human-like agent has made impressive progress toward generating agents with believable appearances and realistic motions for the interactive applications of inhabited virtual worlds. It remains difficult, however, to instruct animatedagents to perform specific tasks or take initiatives. This paper addresses the challenge of instructability by introducing cognitive modelling - a novel logical approach based on a highly developed logical theory of actions, i. e. Event Calculus. Cognitive models go beyond behavioural models in that they govern an agent’s behaviour by reasoning about its knowledge, actions and events. To facilitate the construction of the cognitive models, we propose a high-level behaviour specification language (BSL) from the event calculus formalism. Using BSL, we can specify an agent’s domain knowledge, design behaviour controllers and then control the agent’s behaviour in terms of goals and/or user’s instructions. This approach allows agent’s behaviours to be specified and controlled more naturally and intuitively, more succinctly and at a much higher level of abstraction than would otherwise be possible. It also provides a logical characterisation of planning viaabductive reasoning process. Furthermore, we integrate sensing capability into our underlying theoretical framework, thus enabling animated agents to generate appropriate behaviour even in complex, dynamic virtual worlds. An animated human-like interface agent for virtual environments is used to demonstrate the approach. The architecture for implementing the approach is also described.