Preparing for GSoC 2026: Multi-Drone Mesh Networking - What Does the Community Actually Need?
Who am I
Hello everyone,
My name is Eduardo Gonzalez Silva, and I am a Computer Engineering student at POLI-USP (Brazil). I have been a member of the Skyrats autonomous drone team for the past two years and currently serve as the Lead Software Developer.
Skyrats is a student research group focused on developing autonomous UAV systems using ArduPilot. My work primarily involves companion computers, MAVLink-based communication, ROS2 integration, and swarm coordination architectures.
I have recently developed a multi vehicle project and experienced some of the hardships in multi-vehicle connection. That is why the Multi-Drone Mesh Networking project immediately caught my attention.
Before submitting my formal GSoC proposal, I would like to better understand what the community actually needs in this area.
Suggested Project
The Multi-Drone Mesh Networking (MAVLink-aware) project, proposed for GSoC 2026 and mentored by Nathaniel Mailhot (@NDev), aims to enable resilient communication between multiple ArduPilot vehicles operating as a swarm.
In current multi-vehicle setups, communication typically relies on direct links between each vehicle and a ground station, or between specific vehicle pairs. When one of these links drops due to range limitations, interference, or obstacles, telemetry and command exchange can fail, reducing system robustness and scalability.
This project proposes the development of a mesh networking layer that allows MAVLink messages to propagate across multiple vehicles using multi-hop routing. In such a system, vehicles can relay messages for one another, dynamically maintaining communication paths even when direct connections are unavailable.
Importantly, this would not be a generic mesh implementation, but a MAVLink-aware one. Routing decisions would consider the type and criticality of MAVLink messages, for example, prioritizing heartbeats and critical commands over lower-priority telemetry streams, ensuring efficient use of bandwidth in constrained, distributed environments.
The expected outcome is a practical and deployable communication layer that increases reliability, scalability, and resilience in multi-vehicle ArduPilot operations, enabling more robust swarm deployments in real-world scenarios.
Previous Experience
Over the past two years, I have worked on some interesting projects at Skyrats.
One of my main projects (available on my GitHub) is a Leader/Follower architecture built with ROS2 and DroneKit.
Leader/Follower System
The system consists of:
-
A Leader node that connects to an ArduPilot vehicle via DroneKit, reads GPS data, converts it to UTM/ENU, and publishes the position.
-
A Follower node that subscribes to the leader position, computes a desired offset, and runs a PID controller to generate follow commands.
Through this project, I experienced firsthand the limitations of single-link communication architectures in swarm systems. The coordination logic works well, but robustness quickly becomes an issue when communication drops occur.
In fact our solution depended on using my phone as a WiFi router, and so it could be closer to the swarm, my phone needed to be embedded on the leader drone 
.
This practical exposure strongly motivates my interest in building a resilient multi-hop MAVLink communication layer.
Help From the Community
To better align my proposal with real-world needs, I would greatly appreciate insight from developers and users who have also worked on multi-vehicle systems within ArduPilot.
In particular, I would like to better understand:
-
What communication limitations have you experienced in real multi-vehicle deployments?
-
In what scenarios have current MAVLink routing approaches failed or become unreliable?
-
What bottlenecks (bandwidth, latency, packet loss, scaling limits) have been most problematic?
-
Have there been previous attempts at mesh-style networking in ArduPilot? What were the main challenges or reasons they did not progress?
-
Are there specific use cases (e.g., search and rescue, long-range mapping, distributed sensing) where communication resilience is currently the biggest blocker?
My goal is to design this project around real operational constraints and lessons learned from the community, rather than building something theoretically interesting but practically misaligned.
Understanding the concrete problems encountered by experienced developers and integrators would help me scope the project realistically and prioritize the features that would provide the greatest impact.
I am fully committed to contributing during the GSoC and beyond, continuing development in the area of multi-vehicle autonomous systems.
Thank you for your time and feedback.
Eduardo Gonzalez Silva