Autonomous UAV Navigation for Target Detection in Visually Degraded and GPS Denied Environments

The use of autonomous Unmanned Aerial Vehicles (UAVs) continues to increase for a wide range of applications, such as wildlife monitoring, planetary exploration, and emergency Search and Rescue (SAR). Autonomous UAVs typically rely on Global Positioning System (GPS) usage and most operations require correct flight conditions with good visibility (no rain, fog nor smoke). However, some difficult environments are characterized by partial to no GPS signal, limiting the drone's ability to localise itself. Moreover, visual obscurants affect the UAV's performance and behaviour by restricting its visibility, adding complexity to some subsystems such as target detection. This paper presents the development of a framework enabling a drone to autonomously navigate and explore in GPS denied environments impacted by visual obscurants. The study illustrates this by formulating the navigation and target detection problem as an autonomous Sequential Decision Problem (SDP) with uncertainty caused by degraded visibility and lack of GPS. The SDP is modelled as a Partially Observable Markov Decision Process (POMDP) and tested using the Adaptive Belief Tree (ABT) algorithm. The navigation task was created following the model of a classic SAR operation in a cluttered indoor environment with visual obscurants. The framework includes target detection under uncertainty from a vision-based camera. The formulated SDP is tested in a simulated environment with partial observability and in presence of smoke. The framework integrates Gazebo, Robot Operating System (ROS), Software In The Loop (SITL) and PX4 firmware to recreate a SAR scenario in simulation. Experiments conducted in the simulated SAR scenario test the target finding task under different levels of visibility, with the target position being unknown, and a pose uncertainty modelled from real-life flight experiments. The experiments in normal visibility were a success, but the developed framework was limited in presence of visual obscurants, reducing the target detection rate. Allowing an UAV to autonomously navigate in challenging environments with low visibility conditions aims at increasing the range of UAV's utilization in critical applications such as SAR and mining, where human interventions are often dangerous and unfeasible.