Experimental Assessment of Chance-Constrained Motion Planning for Small Uncrewed Aircraft

Abstract

This work extends the experimental evaluation of chance-constrained motion planning algorithms to fielded fixed-wing small uncrewed aircraft systems (sUAS). Despite advances in planning algorithms, certain challenges remain to producing trajectories for nonholonomic mobile robotic systems such as sUAS. These challenges include nonlinear dynamics which create a complex mapping from inputs to outputs, initialization uncertainty in online motion planning due to compute time of planners and latency in data transfer, environmental uncertainty that has non-Gaussian impact on the robot’s trajectory, and system uncertainty that arises from incomplete models of complex systems. Small UAS often have proprietary components such as commercial, off-the-shelf autopilots which prevent motion planning models from accurately capturing system behavior in all regions of the state space. These challenges can be addressed by leveraging probabilistic motion planners that accurately represent and reason over dynamics and uncertainty. Chance-constrained motion planning offers a method of reasoning over uncertainty as feasibility constraints, and Monte Carlo sampling within a motion planning algorithm offers a method of representing complex uncertainty that may not have a closed form representation. This work extends the chance-constrained motion planning problem to reason over constraint on the trajectory and constraints on the state which ensure that the system model remains valid. Dynamical and systems models are formulated to extend to a broad class of fixed-wing UAS through the experimental derivation of input distributions and system parameters. Uncertainty is quantified using a data-driven approach to modeling input distributions, and Monte Carlo uncertainty sampling deployed within a Rapidly-exploring Random Tree motion planning algorithm is used to plan a trajectory containing a representation of uncertainty. The motion planning algorithm’s ability to accurately reason over layered chance constraints is evaluated experimentally in 61 fielded missions. The results show that when the flight conditions fall within the domain of the uncertainty distributions, the motion planning system is able to accurately reason over the chance constraints.