Adapting Behaviors to New Environments, Characters, and Tasks

Abstract

Animated characters and robots are exhibiting increasing levels of competence in complex tasks ranging from locomotion to juggling. One continuing challenge, however, is to develop behaviors that are robust with respect to variation in the environment, the character, and the task. Examples of a behavior (e.g. an individual performing a motion) are a rich source of information, but reliable techniques are needed to adapt examples to new situations. This paper describes how task information can be used to adapt existing examples to changes in the environment or to changes in the character's physical characteristics. Sets of rules for adapting or scaling examples to new
situations allow a behavior to be described in a compact way, while also capturing some of the benefits of the information that may be contained in the examples. Results are shown for adapting example grasps to new object geometries and adapting dynamic behaviors to characters with differing physical characteristics.