Intelligence Capture

This paper presents intelligence capture, a novel technique for programming behavioral animation by demonstration. This technique operates in two modes: training and autonomous behavior. In training mode, the human user has direct control over the virtual character through an input device such as a joystick. The human user demonstrates the desired behavior for the character by dictating its actions during an interactive animation. This demonstration is recorded as a set of state-action pairs; each pair represents the specific action the human chose for the character when in the associated state. This data is then processed to ensure that any conflicts in the pairs are eliminated. Later, when the character is to behave autonomously, the recorded state-action pairs are generalized to form a continuous state-to-action mapping. This mapping dictates the action the character is to perform for any given state. Thus a behavioral model can be automatically constructed by an animator-programmer team in an intuitive manner, eliminating a programming bottleneck and making this process simpler and quicker. Moreover, the animator has greater control over the creation of the behavioral model, and stylized behavior can easily be realized. While not a panacea, intelligence capture can effectively produce many types of behavioral animation.