UAV$^2$: A Unified and Adaptive Scheduling Framework for UAV Autopilot System with Reinforcement Learning

Unmanned aerial vehicle (UAV) autopilot systems typically comprise navigation and flight-control modules, and their effective scheduling is critical to achieving high flight performance. However, most existing UAV platforms adopt a split architecture in which navigation and flight control are deployed on separate hardware devices. This separation restricts system-wide observability and prevents holistic scheduling and optimization across the entire autopilot pipeline. Moreover, autonomous flight performance emerges from implicit, cross-coupled, and accumulated interactions among multiple factors, rendering traditional model-based or heuristic scheduling approaches ineffective. To address these challenges, we propose UAV$^2$, a unified and adaptive scheduling framework for UAV autopilot systems with reinforcement learning, targeting flight performance optimization. UAV$^2$ integrates navigation and flight control onto a single onboard computing platform and operating system, formulates the scheduling problem as a partially observable Markov decision process, and learns scheduling policies from runtime execution feedback. The proposed approach is trained and evaluated in a hardware-in-the-loop simulation environment. Experimental results demonstrate that the learned scheduling policy consistently outperforms fixed-rate scheduling strategies in terms of flight robustness and tracking performance.

Autonomous drones depend on two tightly linked software components: navigation, which decides where to go, and flight control, which executes those decisions. In most drones today, these two components run on separate computer chips. This split design means the system cannot see the full picture when allocating computing resources, making it difficult to optimize overall flight performance. Moreover, the impact of scheduling decisions — which task gets how much computing time — only becomes apparent after a delay, making it nearly impossible to design effective scheduling rules by hand. This paper introduces a framework that unifies all drone software onto a single onboard computer and treats scheduling as a learning problem. Using a technique called reinforcement learning, the system learns from actual flight experience — much like a pilot improving through practice — to dynamically adjust how computing time is distributed among tasks. The goal is straightforward: fly more accurately and crash less often. The approach was trained and tested in a hardware-in-the-loop simulator, where the drone's software runs on a real embedded computer while its physical motion is simulated with high fidelity. Across challenging obstacle-filled missions, the learned scheduler consistently outperformed traditional fixed-rate strategies, achieving both better trajectory tracking and a higher mission success rate. These results suggest that allowing drones to learn their own scheduling strategies from real execution feedback is a practical and promising direction for making autonomous flight safer and more reliable.