Topological navigation graph framework

Abstract

In this paper, we focus on the utilisation of reactive trajectory imitation controllers for goal-directed visual navigation in mobile robotics. We propose topological navigation graph (TNG) framework. TNG is an imitation-learning-based topological navigation framework for navigating through environments with intersecting trajectories. It represents the environment as a directed graph composed of perception and action modules. Each vertex of the graph corresponds to a trajectory and is represented by a trajectory identification classifier and a trajectory imitation controller. The edges of TNG correspond to intersections between trajectories and are represented by trajectory intersection recognition classifiers. Having a visually specified goal state, TNG navigates by forming a sequential composition plan of trajectory imitation controllers. We also propose to apply neural object detection architectures for the task of trajectory following by detecting direction of movement. We provide empirical evaluation of the proposed navigation framework and its components both in simulated and real-world environments and demonstrate that TNG allows us to utilise non-goal-directed, imitation-learning methods for goal-directed autonomous navigation.