Our long-term goal is the development of an embodied cognitive agent that interacts intelligently with its environment, and communicates its beliefs, actions, and observations in natural language.
Two main outcomes are expected from this work: getting a deeper understanding of human cognition and language competence, and producing intelligent artifacts for various applications of robotics. We pursue this project making use of the SNePS knowledge representation and reasoning and acting system of Shapiro et al. Our research moves in three main, parallel directions:
- First, there are research problems, falling under the umbrella of cognitive robotics, of an agent's reasoning about its perception and action. These give rise to issues of reasoning about fallible perception, causal reasoning, counterfactual reasoning, belief, and belief change.
- Second, we address problems related to embodied linguistic competence and its link to acting. Chief among these are issues related to indexicals, event ontologies, and the appropriate linguistic generation of causal reports.
- Third, we continue to develop and enhance both the theory and the implementation of the SNePS system.
Research Topics & Projects
The Logic of Perception
Perception may broadly be described as a process whereby an agent acquires information about its environment. The study of perception can focus on different stages of such a process. We investigate the stages of higher-level perception, where raw sensor data is mapped onto higher-level symbolic representations of knowledge. The investigation is carried out using the SNePS knowledge representation and reasoning system and the GLAIR grounded, layered agent architecture.
Propositional attitudes include all the attitudes an agent can have towards a proposition. This includes, but is not restricted to, knowledge, belief, and desire. Representing and reasoning about propositional attitudes is crucial for effective communication (where we need to reason about the attitudes of other agents), default reasoning, and action (where we need to reason about our own beliefs.) The most common approach to propositional attitudes is the modal approach, dating back to Hintikka. A first-order approach, though extremely desirable, is not popoular since it is prone to paradoxes of self -reference. Getting the best of both worlds---the simplicity of first-order logic and the consistency of modal logic---is the main objective of this research.
The term “ontology” refers to the branch of metaphysics concerned with the study of the categories of, properties of, and interrelations between entities that exist. More recent uses of the term (mainly within the computer science community) refer to a logical theory of our conceptualization of a certain domain. Typically, an ontology is that part of a logical theory that represents terminological knowledge. Such knowledge is necessary in constraining the interpretation of the logical theory, thereby ensuring that statements in our logical theory mean what we intend them to mean. This explicit highlighting of semantics is particularly crucial in an age in which we strive to access information based on content, not on form. This is the main thrust behind the notion of the semantic web.