I read the dyadic conversations paper. Here’s how I think it ties into the work that Omar and I are doing.
The goal is to make a conversation engine that is efficient while still being realistic. These conversations focus on gestures rather than words. They are to be used with background characters.
The conversation is controlled by a finite state machine. At each tick, the FSM will determine whether the current action is done. If it is, it will determine the next action through a mathematical calculation that depends on the current stage of the conversation and the agent’s relationship with the person he’s talking to. The FSM will then grab the correct action from a set of “base” actions (Nicole’s animations). Depending on the agent’s emotional state, this base action may be modified (through control curves, for example; Omar’s work). A conversation ends whenever a variable, which can be thought of as an “end conversation” variable for simplicity, is over a certain range.
Is that correct?
-Nicole
I think you got the main ideas. I'll invite Libo to read it through so that she can endorse it.
ReplyDeleteIn terms of the mapping from state to base actions and emotional state to control curves, we should have a solid and complete list on which we all agree.
I think you got the main ideas. I will discuss with Norm to determine which factors influence the variation of the actions actually.
ReplyDelete