Logical-Graph Interpretation of Scene in the Transformer Core of an Autonomous Drone

This paper presents the concept of an intelligent drone equipped with a semantic core designed to construct logical-graph representations of the environment based on sensory data. The main objective is to develop an adaptive system capable not only of object recognition but also of forming causal, spatial, and functional relationships between objects to enable logical reasoning. The proposed approach integrates reinforcement learning techniques with graph-based semantics within a transformer architecture, thus ensuring autonomous operation and behavioral flexibility in partially observable environments.