MAV Belief Space Planning In 3D Environments With Visual Bearing Observations

This paper examines the problem of path planning for a MAV in complex 3D environments without the use of GPS. Instead the MAV must rely on using sensor measurement of its surroundings for localisation. The quality of localisation and uncertainty in the MAV's estimated state will be determined by the path it takes through the environment. In order to minimize the probability of failure this state uncertainty must be incorporated into the planning process determining the path of the MAV. We present a path planner that can be used to produce paths which attempt to minimize the MAV's state uncertainty when operating in complex 3D environments. Simulation results are presented for a MAV equipped with a limited field of view camera sensor in environments featuring fixed localisation beacons from which, bearing measurements can be obtained and used for localisation.