SCALE-COMM boosts robot coordination with stable, decoupled communication

SCALE-COMM (Shared, Contrastively-Aligned Latent Embeddings for COMMunication) tackles a core challenge in decentralized multi-agent reinforcement learning: emergent communication between autonomous mobile robots. Existing approaches often suffer from unstable protocols, ungrounded message semantics, and interference between communication learning and policy optimization. SCALE-COMM decouples these two processes by training low-dimensional latent messages that carry task-relevant planning and traffic information, enforced to be consistent across agents and time via a contrastive alignment objective.

Tested on standard MARL benchmarks and a realistic warehouse coordination task, SCALE-COMM consistently outperforms baselines in both representation quality and final task performance. It achieves better sample efficiency and stability during policy fine-tuning, with measurable improvements in throughput. The work, authored by Mahmoud Abouelyazid and Eman Hammad (submitted to IEEE IV 2026), shows that representation-driven communication is a scalable path for multi-agent coordination – no need to reinvent the protocol with each policy update.