We present an extended framework for modelling agent-object interactions in virtual environments. Our framework is based on the concept of Smart Objects and provides agents with pre-programmed interaction information for the automatic generation of animations. The ability to generate such animations without human intervention is vital when constructing plausible, real-time agents. Unlike previous approaches, our model also contains information for directing the attention of agents when interacting with objects. Such information is useful for driving gaze behaviours, for example when grasping objects. Our framework supports both bottom-up (attention capture) and top-down, task driven, simulation of behavioural animation on a per-object basis. It also provides support for the management of the interactions of multiple agents with a single object. We show how objects are designed and provide a concrete example of using the modelling approach with a gaze controller in an animation system....