Abstract
When the navigational environment is known, it can be represented as a graphwhere landmarks are nodes, the robot behaviors that move from node to node areedges, and the route is a set of behavioral instructions. The route path fromsource to destination can be viewed as a class of combinatorial optimizationproblems where the path is a sequential subset from a set of discrete items.The pointer network is an attention-based recurrent network that is suitablefor such a task. In this paper, we utilize a modified R-NET with gatedattention and self-matching attention translating natural language instructionsto a high-level plan for behavioral robot navigation by developing anunderstanding of the behavioral navigational graph to enable the pointernetwork to produce a sequence of behaviors representing the path. Tests on thenavigation graph dataset show that our model outperforms the state-of-the-artapproach for both known and unknown environments.