Parametrized maneuvers for autonomous vehicles

Abstract

This thesis presents a method for creating continuously parametrized maneuver classes for autonomous vehicles. These classes provide useful tools for motion planners, bundling sets of related vehicle motions based on a low-dimensional parameter vector that describes the fundamental high-level variations within the trajectory set. The method follows from a relaxation of nonlinear parametric programming necessary conditions that discards the objective function, leaving a simple coordinatized feasible space including all dynamically admissible vehicle motions. A trajectory interpolation algorithm uses projection and integration methods to create the classes, starting from arbitrary user-provided maneuver examples, including those obtained from standard nonlinear optimization or motion capture of human-piloted vehicle flights. The interpolation process, which can be employed for real-time trajectory generation, efficiently creates entire maneuver sets satisfying nonlinear equations of motion and nonlinear state and control constraints without resorting to iterative optimization. Experimental application to a three degree-of-freedom rotorcraft testbed and the design of a stable feedforward control framework demonstrates the essential features of the method on actual hardware. Integration of the trajectory classes into an existing hybrid system motion planning framework illustrates the use of parametrized maneuvers for solving vehicle guidance problems. The earlier relaxation of strict optimality conditions makes possible the imposition of affine state transformation constraints, allowing maneuver sets to fit easily into a mixed integer-linear programming path planner.

(cont.) The combined scheme generalizes previous planning techniques based on fixed, invariant representations of vehicle equilibrium states and maneuver elements. The method therefore increases the richness of available guidance solutions while maintaining problem tractability associated with hierarchical system models. Application of the framework to one and two-dimensional path planning examples demonstrates its usefulness in practical autonomous vehicle guidance scenarios.