It is commonly agreed that a well-balanced mix of collaboration, training and simulation eventually produce a superior learner. Today's collaborative design and learning environments integrate variety of interactive objects as well as many technological aspects to achieve such balance. Unfortunately, the actual profit of the resulting learning systems is largely reduced by poorly represented interactive objects as well as poor interlinking between such objects. In particular, such objects appear isolated: they neither can be modified sufficiently (e.g., by choosing parameters or enhancing functionality) nor be interlinked properly with their context (e.g., by synchronizing with a guided tour or metadata). We are presenting in this article a model for representing virtual and 3D scenes as learning objects. The model utilizes notions and techniques based on Scene Graphs, X3D, Java3D, and SceneBeans. The prototype accompanied with a simple client-server protocol for exchanging and viewing the 3D SceneBeans. This research aims to extend this protocol by utilizing Sun JXTA primitives to link to the POOL of other learning objects repositories.