This video captures the final day of the ArduPilot Developers Conference 2025, featuring a series of in-depth technical presentations and live demonstrations from key developers. The sessions cover a wide range of advanced topics and recent breakthroughs within the ArduPilot ecosystem. Leonard Hall presents novel algorithms for cooperative lift, demonstrating how multiple drones can be tightly synchronized to carry heavy payloads. The challenges of operating large-scale drone light shows are detailed by Ben Pickard and Tamás Nepusz, who also provide a "wish list" for future MAVLink protocol enhancements. Sid Purohit introduces a new VS Code extension designed to streamline the developer environment, while Andy Piper showcases an innovative "acro-fencing" system that uses AI-assisted Lua scripting to enable safe acrobatic flight. Additional talks cover updates to the MAVLink protocol, including a move to 32-bit system IDs, advanced path planning for fixed-wing aircraft using ROS, real-time mapping with ArduRover and ROS2, and a corporate perspective on contributing to and leveraging the ArduPilot platform for specialized applications.
ArduPilot Developers Conference 2025: Day 3 Talks and Demos
1:11 Welcome & Admin: The final day's proceedings begin with administrative notes, including a shuffled schedule to accommodate weather for outdoor demos and a heads-up about a UK-wide emergency alert test scheduled for 3 PM.
5:41 Cooperative Lift Algorithms (Leonard Hall): Leonard Hall details recent work to enable multiple aircraft to cooperatively lift a single heavy payload.
Key Takeaway: A new "Follow Mode" solution allows a following aircraft to track the lead aircraft's intended kinematic path, not just its noisy position, enabling extremely tight synchronization. A demonstration shows two drones lifting a 54kg payload.
An "Auto Mode" solution is proposed where all drones run an identical mission with unique offsets, synchronized by a shared time reference to minimize data link traffic and allow for load balancing based on battery levels.
"Crane Mode" is introduced as a new manual control mode that provides the smooth, kinematically constrained inputs required for safely maneuvering a swarm.
2:00:38 The Challenges of a Drone Light Show (Ben Pickard & Tamás Nepusz): This talk explores the hardware and software complexities behind large-scale drone light shows.
Hardware (Light Dynamics): The system uses a custom motherboard with RTK GPS, Wi-Fi, and a unique remote on/off feature for managing hundreds of drones. An independent flight termination system is crucial for safety.
Software (Skybrush): The show runs on a custom ArduPilot flight mode based on Guided Mode, with position and velocity setpoints updated at 10Hz.
Key Takeaway: Major challenges include compass drift (the most common cause of failure), network management for large fleets, and protocol limitations. They request future MAVLink support for more than 255 system IDs and broadcast file transfers for efficient show/firmware uploads.
2:37:57 VS Code ArduPilot Developer Environment (Sid Purohit): A presentation on a new VS Code extension designed to dramatically simplify the setup and development workflow for ArduPilot.
Key Features: The extension provides environment validation to ensure all tools are correctly installed, a user-friendly UI for build configurations, visual selection for enabling/disabling features, and integrated step-through hardware and SITL debugging.
2:58:31 Acro-fencing, AI, and Acrobatics (Andy Piper): Andy Piper demonstrates a system for flying acro safely within a virtual fence, where the drone "bounces" off the boundary instead of executing a standard failsafe.
Key Takeaway: The entire system is built using Lua scripting and leverages Google Gemini (an LLM) with custom "playbooks" to write complex code for acrobatic maneuvers (like "flips on a switch") and the fence-bounce logic. The drone temporarily enters Guided Mode to perform the automated correction before returning control to the pilot.
5:07:14 MAVLink Report (Hamish Willee & James Patterson): An update on the MAVLink protocol, emphasizing a deliberately slow pace of change to ensure stability.
Key Takeaway: A major upcoming change is the move to 32-bit system IDs to support large drone swarms and fleets. The implementation will be backward-compatible, using a special flag so that only systems with IDs greater than 255 will use the new, slightly larger packet format.
5:59:07 ROS, Plane, and Advanced Path Planning (Ryan Friedman & Rhys Mainwaring): This talk follows their earlier demo, detailing efforts to create advanced, onboard path planning for fixed-wing aircraft.
The framework aims to allow planes to autonomously obey fences, avoid terrain, and respect operational constraints, even during RTL or GCS failsafe.
They showcase integration with two planners: Sheerwater.ai's commercial AI planner (for optimizing routes based on energy, time, or distance) and a Python-based port of ETH Zurich's terrain-aware planner.
6:39:30 ArduRover & ROS2 Autonomy Demo (Stephen Dade): A live indoor demonstration of a rover performing Simultaneous Localization and Mapping (SLAM).
Key Takeaway: The rover, running ArduPilot, uses a Raspberry Pi 4 with ROS2. ROS2 fuses IMU and wheel odometry data from ArduPilot with scans from a 2D LiDAR to generate a map of the room in real-time using the Cartographer SLAM algorithm.
6:56:01 Kraus Hamdani Aerospace: Enhancements to ArduPilot (Tom Pittenger): A presentation on how ArduPilot has been the core of Kraus Hamdani's success with their long-endurance V-TOL glider, which holds a 76-hour flight record.
Key Takeaway: The talk highlights the challenges commercial entities face when working with an open-source project, including the difficulty of upstreaming code, managing private forks across releases, the steep learning curve for new engineers, and overall software configuration management.