A ROS2 framework for decentralized multi-agent coordination

This project implements decentralized multi-agent formation control and leader-follower coordination in ROS2. Our goal in this project was to bridge well-known multi-agent control laws from literature with a working robotic software architecture. Our architecture is decentralized, which means that each robot only needs to have the information about its neighbors to perform the multi-agent task. The framework supports positon-based formation control, distance-based formation control, and leader-follower coordination. We validated the system in simulation using five-agent formation control and four-agent leader-follower experiments. We also tested the framework on hardware, where we performed formation control with three TurtleBots.

Decentralized control: Each agent runs an independent controller, filters neighbor information according to a pre-defined communication graph, computes a single-integrator control input, and maps it to unicycle commands for differential-drive robots.

Motivation: Multi-agent formation control is often presented mathematically using idealized single-integrator dynamics. However, deploying these algorithms on robot platforms requires several additional system components: communication between agents, coordinate-frame transformations, task management, and real-time feedback. This project focused on building that implementation layer in ROS2. Specifically, we developed this architecture to support human-robot team performance, such as dance, in our university's robotics lab . This framework will be used by the lab in the future once we incorporate safety, such as robot-robot and robot-human collision avoidance, in our framework. For our work/progress on safety in multi-agent formation control, please visit this Safety-Forecasting page.