Emergent Communication Protocols with Hardware-Informed Constraints

Recent work (Smith et al., 2025; Wang et al., 2024) investigates communication in MARL, but often abstracts away from real hardware constraints. This idea flips the paradigm: agents learn to coordinate, but their communication actions are subject to realistic, learned constraints based on the underlying hardware (e.g., wireless spectrum, network congestion). The goal is to see what kind of “language” or protocol emerges when agents are forced to be efficient, robust, and adaptive under strict hardware-imposed bottlenecks. This could be evaluated using heterogeneous hardware platforms as in Wiggins et al. [2023]. The innovation is in closing the simulation-to-reality gap by letting real-world hardware limitations become an evolutionary pressure shaping emergent communication—yielding protocols that are practical, efficient, and robust in the wild, not just in simulation.

References:

1. Robust and Safe Multi-Agent Reinforcement Learning with Communication for Autonomous Vehicles: From Simulation to Hardware. Keshawn Smith, Zhili Zhang, H. Ahmad, Ehsan Sabouni, Maniak Mondal, Song Han, Wenchao Li, Fei Miao (2025).
2. Multiple Ships Cooperative Navigation and Collision Avoidance using Multi-agent Reinforcement Learning with Communication. Y. Wang, Y. Zhao (2024). arXiv.org.
3. Evaluating multi-agent reinforcement learning on heterogeneous platforms. Samuel Wiggins, Yuan Meng, R. Kannan, V. Prasanna (2023). Defense + Commercial Sensing.