Academic Publications
Research papers published in peer-reviewed journals and conferences
A Physics-Constrained Learning Approach for Active Tracking of Dynamic Objects from a UAV
This paper proposes a hybrid perception and control framework for real-time vision-based active tracking of flying targets using unmanned aerial vehicles (UAVs). The perception module integrates a YOLO-based detector with a KCF tracker to ensure consistent 3D pose estimation of fast-moving aerial targets, while minimizing computation overhead. The estimated target state is then fed into a Proximal Policy Optimization (PPO)-based reinforcement learning controller trained in simulation, which computes optimal chaser UAV actions that account for system dynamics and physics-informed constraints such as velocity and field-of-view limitations. The framework is first trained and evaluated in the AirSim simulation platform and then successfully deployed in a physical lab setting using a Crazyflie nano drone, demonstrating robust sim-to-real transfer, improved tracking stability, and energy-aware maneuvering in GPS-denied environments.
Scalable and Load-Balanced Coverage Path Planning for Multiple UAVs Surveying Non-Convex Areas
This paper introduces SCoPP, a novel coverage path planning framework for multi-UAV teams operating in complex, non-convex environments with discontinuities like no-fly zones. It ensures workload balancing and fast path generation, with extensions for prioritizing critical waypoints and deadlines. Extensive experiments, including a 3-UAV field deployment and large-scale simulations up to 150 UAVs, demonstrate its scalability, efficiency, and applicability to time-sensitive missions like post-flood surveys.
An Open-Source Hardware/Software Architecture and Supporting Simulation Environment to Perform Human FPV Flight Demonstrations for UAV Autonomy
This paper presents an open-source architecture for conducting human-guided FPV flight experiments using a low-cost F450 quadcopter with Pixhawk. The system supports dual-view video, synchronized telemetry logging, and full control access via a Python-based GUI. A digital twin simulation environment, built in AirSim and Unreal Engine, enables hardware-in-the-loop (HITL) testing via Pixhawk, supporting scalable training and validation of autonomous UAV agents.
Interested in collaborating on robotics research?
Let's discuss potential research collaborations, projects, or applications of my work in robotics and AI.