LatRec+: Learning-Based Goal Recognition in Latent Space

Existing approaches to goal recognition are able to infer domain knowledge by combining goal recognition techniques from automated planning, and deep autoencoders to learn domain theories from data streams. However, most recent approaches to goal recognition in these learned domains struggle with high spread during recognition process. LATREC+ leverages from the usage of learning approaches to recognize goals directly in real-world data (images), without relying on domain theories. The learned model is given a set of observations and returns the probability of each predicate being true. We demonstrate this approach in an online simulation of simple games, such as the n-puzzle game.