A homotopy-based hierarchical framework for semi-autonomous/autonomous vehicle navigation

Abstract

Semi-autonomous and autonomous vehicles have been of interest for reasons such as safety, efficiency, and convenience. The thesis proposes a homotopy-based hierarchical motion planning and control framework for vehicle navigation. A homotopy is, roughly speaking, a set of trajectories with the same high-level navigation decision. The motivation of the proposed hierarchical framework based on homotopy class is twofold: compatibility with humans decision and computational benefits. The approach explicitly identifies and enumerates feasible homotopy classes corresponding to different navigation decisions allowing for interaction with a human operator/ supervisor. Also, the approach has computational benefits, specifically enabling a divide-and-conquer strategy. In a collision-free trajectory generation problem, the presence of obstacles serves to creating discontinuities in the set of feasible trajectories. However, the complexity can be reduced significantly if we independently consider multiple distinct continuous sets of feasible trajectories, where no discontinuity is created. The thesis first presents a method for enumeration and representation of the navigation decisions by cell sequences to divide a collision-free vehicle navigation problem using cell decomposition. Then, it proposes a sampling-based method to evaluate the desirability of each navigation decisions in terms of control input safety margin. In order to make a vehicle navigate safely within a chosen navigation decision, a model predictive control framework is utilized with a corresponding navigation decision constraint. The constraint is non-convex, but a sequence of convex cells is prescribed in advance. An efficient formulation of the problem into mixed integer programming is proposed and validated in the thesis. Finally, a user study in a driving simulator shows that users accept semi-autonomous/ autonomous vehicles based on the proposed framework on highways as much as regular vehicles.