Preliminary study on drone navigation in urban environments using visual odometry and partially observable Monte Carlo planning

Abstract

Due to the recent technological development in drone technology, a drone is used in many applications like delivery, search and rescue, and safety inspection especially in low altitude airspace. However, the mass deployment of drones for commercial purposes is yet to be matured. Therefore, normally drone is used in time-critical applications like the delivery of essential medical supplies, these applications often require high reliability. Nowadays, drone normally relies on Global Positioning System (GPS) alone for outdoor navigation, but there is also the possibility that the GPS signal is lost due to GPS jamming in the area. This paper provides a solution for drone navigation in an unknown outdoor environment with no GPS signal. The drone’s surrounding environment is perceived via a camera and is constructed into a 3D occupancy grid map, followed by localization of its position. The navigation is formulated as a sequential decision-making problem and modeled using a partially observable Markov decision process (POMDP). The simulation shows the drone can navigate towards the goal by taking a local optimum decision iteratively based on its perceived surrounding environment at each step.