Frankenstein: To Be Hurt or Not to Be Hurt \\Entity Tagging-Based Decision Transformer for Robot Safety Decision-Making

Keywords: Decision Making, tagging, ANN

TL;DR: Propose a hypothesis based on psychology that the execution process of thinking is dichotomous, suggesting that the thing with the highest probability of execution is proposed first.

Abstract: The conflict between human instructions and safety requirements constitutes a core dilemma in autonomous robotic systems. Inspired by Mary Shelley's "Frankenstein," this study addresses the fundamental question of whether robots should blindly obey their creator's commands or choose safer alternatives based on learned experiences. We propose an entity tagging-based Decision Transformer architecture that chains past action-result pairs by entity type and implements a safety decision-making system integrating neuropsychological conflict monitoring mechanisms.

Submission Number: 41