Pulkit Goyal

Understanding how humans effortlessly learn general concepts of the workings of the world through everyday interactions from early childhood is pivotal for advancing artificial intelligence in both physical and non-physical domains. This project seeks to develop a theoretical framework and its practical implementation for robots to extract high-level conceptual descriptions of their environment through embodied interpretation of autonomous sensorimotor interactions and by observing and interacting with humans.

Inspired by cognitive science, these high-level abstractions would be built on the logical formalization of such interactions based on image schemas (simple yet fundamental notions humans learn in early childhood that enable conceptual and metaphoric thinking) and affordances (actions an object offers an actor). The learned representations will empower a robot to reason about its interactions irrespective of their concrete sensory manifestations, transfer skills to new environments, solve novel tasks, and communicate with humans on the basis of shared conceptualizations.