State Estimation for Autonomous Flight in Cluttered Environments

Abstract

Safe, autonomous operation in complex, cluttered environments is a critical challenge facing autonomous mobile systems. The research described in this paper was motivated by a particularly difficult example of autonomous mobility: flightofasmallunmannedaerialvehiclethroughaforest.Thefocuswasonenablingthethreecriticaltasks that comprise flight: 1) maintaining controlled flight while avoiding collisions (aviate); 2) flying from a known start location to a known goal location (navigate); and 3) providing information about the environment—a map—to a humanoperatororotherrobotsin the team(communicate).Presentedhere isasolution tothe problem ofestimating vehicle state (its position, orientation, and velocity) as well as the positions of obstacles or landmarks in the environment using only inertial measurements and bearings to landmarks. This is a highly nonlinear estimation problem, and standard estimation techniques such as the extended Kalman filter are prone to divergence in this application. In this paper asigma-point Kalman filter is implemented, resulting in anestimator which is ableto cope with the significant nonlinearities in the system equations and uncertainty in state estimates while remaining tractable for real-time operation. In addition, the issues of data association and landmark initialization are addressed. Estimator performance is examined through Monte Carlo simulations in two dimensions for scenarios involving unmanned aerial vehicle flight in cluttered environments. Simulations show that convergent, consistent estimates of vehicle state and obstacle positions can be obtained and that the estimates can be used by a trajectory planner to generate a path through a cluttered environment.