Andrés Occhipinti Liberman

Andrés Occhipinti Liberman, Blai Bonet, Hector Geffner

Two main approaches have been developed for learning first-order planning (action) models from unstructured data: combinatorial approaches that yield crisp action schemas from the structure of the state space, and deep learning approaches that produce action schemas from states represented by images. A benefit of the former approach is that the learned action schemas are similar to those that can be written by hand; a benefit of the latter is that the learned representations (predicates) are grounded on the images, and as a result, new instances can be given in terms of images. In this work, we develop a new formulation for learning crisp first-order planning models that are grounded on parsed images, a step to combine the benefits of the two approaches. Parsed images are assumed to be given in a simple O2D language (objects in 2D) that involves a small number of unary and binary predicates like "left", "above", "shape", etc. After learning, new planning instances can be given in terms of pairs of parsed images, one for the initial situation and the other for the goal. Learning and planning experiments are reported for several domains including Blocks, Sokoban, IPC Grid, and Hanoi.

Thomas Bolander, Nina Gierasimczuk, Andrés Occhipinti Liberman

We consider a learning agent in a partially observable environment, with which the agent has never interacted before, and about which it learns both what it can observe and how its actions affect the environment. The agent can learn about this domain from experience gathered by taking actions in the domain and observing their results. We present learning algorithms capable of learning as much as possible (in a well-defined sense) both about what is directly observable and about what actions do in the domain, given the learner's observational constraints. We differentiate the level of domain knowledge attained by each algorithm, and characterize the type of observations required to reach it. The algorithms use dynamic epistemic logic (DEL) to represent the learned domain information symbolically. Our work continues that of Bolander and Gierasimczuk (2015), which developed DEL-based learning algorithms based to learn domain information in fully observable domains.

Andrés Occhipinti Liberman, Andreas Achen, Rasmus Kræmmer Rendsvig

Many classical planning frameworks are built on first-order languages. The first-order expressive power is desirable for compactly representing actions via schemas, and for specifying quantified conditions such as $\neg\exists x\mathsf{blocks\_door}(x)$. In contrast, several recent epistemic planning frameworks are built on propositional epistemic logic. The epistemic language is useful to describe planning problems involving higher-order reasoning or epistemic goals such as $K_{a}\neg\mathsf{problem}$. This paper develops a first-order version of Dynamic Epistemic Logic (DEL). In this framework, for example, $\exists xK_{x}\exists y\mathsf{blocks\_door}(y)$ is a formula. The formalism combines the strengths of DEL (higher-order reasoning) with those of first-order logic (lifted representation) to model multi-agent epistemic planning. The paper introduces an epistemic language with a possible-worlds semantics, followed by novel dynamics given by first-order action models and their execution via product updates. Taking advantage of the first-order machinery, epistemic action schemas are defined to provide compact, problem-independent domain descriptions, in the spirit of PDDL. Concerning metatheory, the paper defines axiomatic normal term-modal logics, shows a Canonical Model Theorem-like result which allows establishing completeness through frame characterization formulas, shows decidability for the finite agent case, and shows a general completeness result for the dynamic extension by reduction axioms.