A Controlled Comparison of MARL Training-Execution Paradigms for Cooperative Multi-UAV Fire Detection
Abstract
This paper presents a controlled empirical comparison of three fundamental multi-agent reinforcement learning (MARL) paradigms, the Centralized Training and Execution (CTE), Decentralized Training and Execution (DTE), and Centralized Training with Decentralized Execution (CTDE), in a cooperative Unmanned Aerial Vehicle (UAV) fire-detection environment under identical task conditions. In MARL, the training and execution architecture shapes what the agents can see, what they can coordinate on, and how stable the learning problem is. Our study addressed a gap where paradigm-level comparison under controlled settings remains scarce. Results across four benchmark scenarios and a 6 to 15 agents scalability test show that CTDE may be the most reliable paradigm, winning Level-1 mission completion in three of four scenarios and Level-3 path quality in all four. CTE provides a measurable advantage over CTDE in Hard-Dynamic, where execution-time global state conditioning enables faster response to mid-episode fire spawning, but its joint action space of 4^n renders it infeasible beyond thirteen agents. DTE fails systematically due to non-stationarity, with lower completion rates across all configurations and apparent safety gains that are artefacts of coordination failure rather than genuine collision avoidance. These findings confirm that paradigm choice is a first-order design decision in cooperative MARL and that CTDE may be a promising architecture for coordination-intensive multi-agent deployments at moderate to large scales.
Keywords: Multi-Agent Reinforcement Learning, Wildfire Detection
DOI: 10.54941/ahfe1008489
Cite this paper
More from this volume
- AI-Enhanced Live-Streamed English Language Learning: Designing and Evaluating an Intelligent Platform for Chinese Learners
- "They Will Move": Confirmation Bias, Expectation Frames, and Premature Monitoring Drop Off in COLREG Encounters
- Automation transparency communication does not improve supervisory control performance, but shapes operator trust
- The influence of neuroticism on startle and surprise response: a flight simulator study
- EEG-Based Classification of Reward, Penalty, and Movement Events Across Four Video Games of Varying Complexity
- The Effects of Task Similarity and Time Pressure on Task Performance Following Interruptions in Complex Human-Computer Interaction Systems
- Trusting the Chain, Not the Black Box: A Provenance-Based Architecture for Human-Machine Teaming in Military Autonomy
- Design of a Simplified Humanoid Robot Face: Cultural Differences and Emotional Synchronization
- From Visual Observation to Auditory Interaction: A Human-Centered Tangible Music System Based on Physical Objects
- The Impact of Scarcity-Based Dark Patterns on User Performance, Perceived Darkness, and Workload in Subscription Services
- A Behavioral Observation Tool for Diagnosing Human–AI Interaction Profiles in Engineering Design Education
- The Mediated Moon: Semantic Mediation Between Gesture and Generative Image


AHFE Open Access