Learning to Fly Smarter: Reinforcement Learning for Adaptive UAV Data Collection with AERPAW

Ahmed Ibrahim

Associate Professor, Florida International University

Ahmed Ibrahim, an Associate Professor at Florida International University, and his student, Joarder Jafor Sadique, are rethinking how UAVs collect data in dynamic wireless environments. Traditional “fixed-path” routing approaches, where UAVs follow pre-determined trajectories, fail to adapt to changing signal conditions, often resulting in longer mission times and inefficient data collection. This limitation becomes especially critical in real-world deployments where wireless conditions vary due to terrain, interference, and mobility.

To address this challenge, the team developed a reinforcement learning (RL)-based approach that enables UAVs to learn optimal flight paths by continuously balancing energy consumption with data collection performance. Rather than relying on static planning, the RL agent adapts its decisions based on the evolving environment, leading to more efficient and responsive mission execution.

AERPAW played a central role in enabling this innovation through its integrated digital twin and real-world experimentation capabilities. The researchers used the AERPAW development environment as a high-fidelity digital twin of the Lake Wheeler testbed, accurately capturing both physical terrain and RF characteristics. This allowed them to train and validate their machine learning models under realistic conditions before deployment. Leveraging the “develop-to-deploy” workflow, the team was able to transition seamlessly from the emulator to the outdoor testbed with minimal software modifications, significantly accelerating the experimentation cycle. In addition, by utilizing AERPAW’s standardized portable nodes and UAV configurations, the researchers ensured that their experiments were repeatable and directly comparable with other studies on the platform. Through this work, the team demonstrates how intelligent, adaptive routing can dramatically improve UAV data collection efficiency, paving the way for more autonomous and resilient aerial sensing systems.

Their findings are presented in “UAV-aided fast data collection via machine learning using AERPAW’s digital twin,” published at the ACM WiNTECH 2025 workshop, and in the forthcoming article “Aerial Link Scheduling over Riemannian Manifolds Using AERPAW’s Digital Twin,” submitted to IEEE Communications Magazine in March 2026.