상세 보기
Vacuous QoS Constraints in Constrained DRL-URLLC Simulations: Root Causes, Detection, and Remedies
- Tran, Hao Hoang;
- Nguyen, The Vi;
- Nguyen, Thi My Tuyen;
- Won, Dongwook;
- Oh, Junsuk;
- ... Cho, Sungrae
SCOPUS
0초록
Deep Reinforcement Learning (DRL) has become a dominant approach for resource allocation in Ultra-Reliable Low-Latency Communication (URLLC) systems. However, an unaddressed methodological failure silently undermines many of these studies: the use of vacuous Quality of Service (QoS) constraints. We reveal that predefined QoS thresholds (Tk) are frequently set so high that they are never violated under typical simulation geometries. Consequently, constraint enforcement mechanisms, such as Lagrangian multipliers or Proportional-Integral-Derivative (PID) controllers, receive zero training signal. Thus, supposedly 'constrained' DRL agents are functionally trained without constraints, rendering baseline comparisons invalid. In this letter, we introduce the concept of vacuous constraints in DRL-URLLC simulations, supported by empirical evidence from a representative UAV-URLLC environment. We provide a systematic taxonomy of constraint regimes, unveil the root cause via the capacity gap condition, and propose a comprehensive pre-training audit protocol and checklist for researchers and reviewers to detect and remedy this widespread, silent failure mode.
키워드
- 제목
- Vacuous QoS Constraints in Constrained DRL-URLLC Simulations: Root Causes, Detection, and Remedies
- 저자
- Tran, Hao Hoang; Nguyen, The Vi; Nguyen, Thi My Tuyen; Won, Dongwook; Oh, Junsuk; Cho, Sungrae
- 발행일
- 2026-07
- 유형
- Conference Paper
- 페이지
- 238 ~ 243
- 언어
- ENG
- 출판사
- IEEE Computer Society
- 발행국가
- 미국
- 분량
- 6 페이지
- ISSN
- E 2165-8536
P 2165-8528