A sample discarding strategy for rapid adaptation to new situation based on Bayesian behavior learning

Abstract

Bayesian reasoning is used in many robotics applications when there is significant uncertainty accompanying perception and action. Generally for Bayesian belief changes in query nodes, we are more interested in evidence that may lead to a change in decision. If an observation has very little effect on decisions, it could be regarded as an insignificant observation for the learning process. This paper presents a method for discarding such insignificant observations so that we can concentrate on evidence that is more important and useful for learning. The main advantage of our method is that it can closely follow a user's preference or change in environment without requiring a huge amount of data.