MAVLab: A Phone-First Drone Digital-Twin Simulator for Flattening the Drone-Systems Learning Curve

Learning how drones sense, stabilize, communicate, navigate, and fail is gated by a steep infrastructure barrier: a beginner is typically pushed into ROS, Gazebo, ArduPilot or PX4 SITL, Docker, MAVProxy, and Linux networking before they can observe a single concept the drone itself embodies. This tooling cliff excludes learners who lack hardware, reliable connectivity, or systems-engineering background — a barrier felt acutely in low-resource and African educational contexts. We present MAVLab, a phone-first drone digital-twin simulator that runs its physics, autopilot, and a MAVLink server entirely on a commodity Android device, with no cloud, Docker, or bridge process required. MAVLab presents itself to unmodified ground-control software (QGroundControl) as an ArduPilot-like copter, so learners reach authentic ground-station workflows — arming, mode changes, mission upload, telemetry, and failure response — from the first session. The phone's own inertial sensors map device orientation to aircraft attitude, making the sense-to-state relationship physically tangible, while a state-driven 3D twin and a failure lab turn "drones are cool" into "drones are safety-critical systems." We describe a protocol-first engineering methodology in which GCS/MAVLink compatibility, rather than physics fidelity or UI, is treated as the riskiest assumption and proven before depth is built. We report build and unit-test verification of the open-source (Apache 2.0) implementation and its MAVLink/mission capability coverage, and outline a planned classroom evaluation with an IEEE drone bootcamp cohort. In live testing, MAVLab passed the full QGroundControl acceptance set — vehicle discovery, arm/disarm, command acknowledgement, mission upload, autonomous waypoint execution, link reconnection, and a sustained-stability run — on real Android devices in both same-phone split-screen and desktop-over-Wi-Fi configurations, and was exercised hands-on by a small group of external testers. MAVLab is positioned not as a replacement for professional tools but as the missing first learning layer that inverts the conventional learning order. A controlled classroom study of learning outcomes remains future work. The open-source (Apache 2.0) reference implementation is available at github.com (evaluated build: release v1.5.0). This report is also published at labs.fly-ascend.com.