Dynamic Multi-Agent Communication via ROS 2 Zenoh Protocol for Collision Avoidance Applications

Distributed multi-robot systems rely on timely and reliable communication to coordinate motion and avoid collisions. In ROS~2, most middleware implementations are DDS-based and provide automatic discovery and rich QoS support, but they tend to build dense communication graphs and rely on broadcast discovery whose overhead grows with the number of agents. In the baseline architecture considered in this thesis, collision avoidance is centralized. This thesis instead proposes a distributed collision-avoidance communication layer in which decisions are taken locally on each agent, relying on on-board cameras and ArUco markers for neighbor identification, while the underlying collision-avoidance controller remains unchanged. The distributed scheme is implemented with three alternative middleware configurations that share the same per-agent control logic but differ in communication topology: a standard ROS~2 setup over Fast~DDS, Zenoh with a single central router, and Zenoh with a router-per-agent topology. The router-per-agent design removes broadcast discovery and encapsulates each agent in an isolated communication domain, exposing only a minimal set of inter-agent topics. All configurations are evaluated in Webots on a ten-agent formation and collision-avoidance task. The results indicate that the distributed, sensor-driven design is compatible with the computational budget of individual robots and that Zenoh-based topologies can support the required dynamic distributed communication pattern.