An Online Spatial-Temporal Graph Trajectory Planner for Autonomous Vehicles

Abstract

The autonomous driving industry is expected to grow by over 20 times in the\ncoming decade and, thus, motivate researchers to delve into it. The primary\nfocus of their research is to ensure safety, comfort, and efficiency. An\nautonomous vehicle has several modules responsible for one or more of the\naforementioned items. Among these modules, the trajectory planner plays a\npivotal role in the safety of the vehicle and the comfort of its passengers.\nThe module is also responsible for respecting kinematic constraints and any\napplicable road constraints. In this paper, a novel online spatial-temporal\ngraph trajectory planner is introduced to generate safe and comfortable\ntrajectories. First, a spatial-temporal graph is constructed using the\nautonomous vehicle, its surrounding vehicles, and virtual nodes along the road\nwith respect to the vehicle itself. Next, the graph is forwarded into a\nsequential network to obtain the desired states. To support the planner, a\nsimple behavioral layer is also presented that determines kinematic constraints\nfor the planner. Furthermore, a novel potential function is also proposed to\ntrain the network. Finally, the proposed planner is tested on three different\ncomplex driving tasks, and the performance is compared with two frequently used\nmethods. The results show that the proposed planner generates safe and feasible\ntrajectories while achieving similar or longer distances in the forward\ndirection and comparable comfort ride.\n