A virtual environment differs from a classical frame based animation system mainly in its non-deterministic nature. To address these dynamic conditions, actors must respond to events within the environment as they occur and not simply follow pre-specified scripts.
We are developing an adaptive control technique to improve the creation and runtime control of reactive actors. A reactive actor is defined as a control entity that autonomously chooses its behavior based on the information it receives from the environment and its own internal state.
RAVE (Reactive Actors in Virtual Environments) uses a reinforcement learning model to automatically generate controllers for typical 2D navigational tasks. Collective Learning Systems (CLS) theory is integrated within a hierarchical control model to create controllers which quickly converge on optimal navigational strategies and also adapt to changing environment conditions during runtime.