Methods for Online UAV Path Planning for Tracking Multiple Objects

Abstract

Unmanned aerial vehicles (UAVs) or drones have rapidly evolved to enable carrying various sensors such as thermal sensors for vision or antennas for radio waves. Therefore, drones can be transformative for applications such as surveillance and monitoring because they have the capability to greatly reduce the time and cost associated with traditional tasking methods. Realising this potential necessitates equipping UAVs with the ability to perform missions autonomously. This dissertation considers the problems of online path planning for UAVs for the fundamental task of surveillance comprising of tracking and discovering multiple mobile objects in a scene. Tracking and discovering an unknown and time-varying number of objects is a challenging problem in itself. Objects such as people or wildlife tend to switch between various modes of movements. Measurements received by the UAV’s on-board sensors are often very noisy. In practice, the on-board sensors have a limited field of view (FoV), hence, the UAV needs to move within range of the mobile objects that are scattered throughout a scene. This is extremely challenging because neither the exact number nor locations of the objects of interest are available to the UAV. Planning the path for UAVs to effectively detect and track multi-objects in such environments poses additional challenges. Path planning techniques for tracking a single object are not applicable. Since there are multiple moving objects appearing and disappearing in the region, following only certain objects to localise them accurately implies that a UAV is likely to miss many other objects. Furthermore, online path planning for multi-UAVs remains challenging due to the exponential complexity of multi-agent coordination problems. In this dissertation, we consider the problem of online path planning for UAV-based localisation and tracking of multi-objects. First, we realised a low cost on-board radio receiver system on aUAV and demonstrated the capability of the drone-based platform for autonomously tracking and locating multiple mobile radio-tagged objects in field trials. Second, we devised a track-before-detect filter coupled with an online path planning algorithm for joint detection and tracking of radio-tagged objects to achieve better performance in noisy environments. Third, we developed a multi-objective planning algorithm for multi-agents to track and search multi-objects under the practical constraint of detection range limited on-board sensors (or FoV limited sensors). Our formulation leads to a multi-objective value function that is a monotone submodular set function. Consequently, it allows us to employ a greedy algorithm for effectively controlling multi-agents with a performance guarantee for tracking discovered objects while searching for undiscovered mobile objects under practical constraints of limited FoV sensors. Fourth, we devised a fast distributed tracking algorithm that can effectively track multi-objects for a network of stationary agents with different FoVs. This is the first such solution to this problem. The proposed method can significantly improve capabilities of a network of agents to track a large number of objects moving in and out of the limited FoV of the agents’ sensors compared to existing methods that do not consider the problem of unknown and limited FoV of sensors.