Learning to Simulate the Behavior of a Dynamical Object

摘要

Since sensory processing takes time both for a human and for a robot, it is often necessary to base action not on the perceived state of the word, but on the anticipated future state of the world. Anticipations can either be in the form of predictions where a future state is directly estimated from a number of observations (Balkenius and Johansson, 2007), or in the form of a simulation where a predicted future state is used as input to a subsequent prediction step to simulate the behavior of the world in a number of discrete steps (Hesslow, 2002).