FST Graduate Research Topic

Research Topic: Multiagent Reinforcement Learning (MARL) for Situation Awareness
Suggested By: Oscar Lin
Last Update: May 14th, 2025

Multiagent Reinforcement Learning (MARL) is an advanced extension of reinforcement learning in which multiple autonomous agents interact within a shared environment, each learning to make decisions based on their own observations and experiences. In the context of 'situation awareness', MARL is particularly promising: it enables distributed agents-such as robots, vehicles, drones, or sensors to collaboratively perceive, interpret, and respond to complex, dynamic scenarios where no single agent has complete information. Situation awareness refers to the ability to perceive environmental elements, comprehend their meaning, and predict their future status. In multiagent systems, this often requires agents to share information, coordinate actions, and adapt to the behaviors of others in real time. MARL provides a framework for agents to learn these behaviors through trial and error, optimizing their collective or individual objectives in environments that are often only partially observable and highly dynamic. For your MSc IS essay/project/thesis or BSc CIS final year project (note: Financial support for this research may be available), you can choose one of the key research topics in MARL for Situation Awareness: - Cooperative Perception and Information Sharing:   Developing MARL algorithms that enable agents to share observations and build a more complete, global picture of the environment, improving group-level situation awareness. - Coordination and Communication Protocols:    Designing learning-based methods for agents to coordinate actions and communicate efficiently, especially under bandwidth or privacy constraints. - Partial Observability and Uncertainty:   Addressing challenges where agents have limited or noisy observations, and must infer hidden aspects of the environment or other agents’ intentions. - Scalability and Sample Efficiency:  Improving the scalability of MARL algorithms to handle large numbers of agents and making learning more efficient so agents can operate effectively with less data-critical for real-world deployment. - Explainable MARL for Situation Awareness:   Creating explainable MARL systems so that human operators can understand and trust the decisions made by autonomous agents in safety-critical applications. - Adversarial and Competitive Scenarios:  Exploring how MARL can be used in environments with both cooperative and adversarial agents, such as security, gaming, or autonomous driving. - Real-World Applications:  Applying MARL for situation awareness in domains like autonomous vehicles, multi-robot systems for disaster response, smart grids, and industrial automation, where agents must adapt to rapidly changing conditions and work together to achieve shared goals. MARL for situation awareness is a rapidly growing field, with research focusing on enabling teams of agents to perceive, interpret, and act in complex environments more effectively than any single agent could alone. References:  [1] Dominick J. Perini,Braeden P. Muller,Justin Kopacz, and Alan J. Michaels, An Application of Explainable Multi-Agent Reinforcement Learning for Spectrum Situational Awareness, Electronics 2025, 14(8), 1533; https://doi.org/10.3390/electronics14081533 [2] Renato D. Costa, and Celso M. Hirata, Reinforcement learning applied to a situation awareness decision-making model,  Information SciencesVolume 704, June 2025, 121928  https://www.sciencedirect.com/science/article/abs/pii/S002002552500060X [3] Stefano V. Albrecht,  Filippos Christianos,  Lukas Schäfer, Multi-Agent Reinforcement Learning: Foundations and Modern Approaches, https://www.marl-book.com/ For more information, see https://oscar.athabascau.ca