RHC for Vision-Based Navigation of a WMR in an Urban Environment

This paper presents a control strategy for the navigation of an autonomous vehicle through an unknown environment with a focus on real-time implementation. The task at hand is to navigate from an initial position to a predetermined goal through an unknown environment using vision as the primary sensor. To solve this problem online the system must incorporate several component technologies: estimation of the environment, dynamic path planning, and a control strategy which incorporates this information into real-time implementation. In this paper, the obstacle estimation task is performed using a combination of Structure-From-Motion (SFM) fused with inertial measurements to provide 3D information from an onboard camera and an adaptive multiresolution-based learning algorithm to estimate the environment from 3D feature points. The learned obstacle map is used as a constraint in a Receding Horizon Control (RHC) path planner which calculates dynamically feasible and locally optimal trajectories. The feasibility of the real-time implementation of this system for vehicle navigation is investigated using a Wheeled Mobile Robot (WMR) testbed with a future goal of implementation on a Micro Air Vehicle (MAV).