Section: New Results

Motion planning techniques

Participants : David González Bautista, Fernando Garrido Carpio, Vicente Milanés, Fawzi Nashashibi, Myriam Vaca Recalde, Jose Emilio Traver Becerra.

The latest developments in the Intelligent Transportation Systems (ITS) field allow emerging technologies to show promising results at increasing passengers comfort and safety, while decreasing energy consumption, emissions and travel time. Despite of great efforts, fully automated driving still remains unsolved, where research challenges such as navigation in urban dynamic environments with obstacle avoidance capabilities–i.e. Vulnerable Road Users (VRU) and vehicles–and cooperative maneuvers among automated and semi-automated vehicles, still need further efforts for a real environment implementation. A deep state-of-the-art review has been conducted to find the gaps in this important topic into the autonomous vehicle field, with special attention to overtaking and obstacle avoidance maneuvers [21].

Having this in mind, a novel local path planning algorithm combining both off-line and real-time generation has been proposed in [32], providing a significant reduction on the computational time with respect to prior implementations from RITS team. This new local planning architecture for urban environments benefits from a priori knowledge of the geometry of the road layout, vehicle's kinematics and dynamics, among others, to produce local smooth path for the vehicle to navigate. The planner relies on several databases containing optimized trajectories for a $G^2$ continuous path generation. Four different type of databases have been generated to provide our system with a naturalistic driving style, allowing the car to maintain smooth trajectories according to the characteristics of the road [33].

Based on the accuracy of current digital maps, it is possible to know before-hand the way-points that define the route by which the vehicle will pass to reach a predefined destination. Furthermore, the original route can be generated in real-time and modified on-demand according to the user needs through the use of Automatic Global Planners (AGP) [42]. That way, since urban scenarios can present several consecutive curves in a short period of time, a smoother and more comfortable path generation can be done by extending the planning horizon up to two curves. There, a set of paths are analyzed by considering the angles of the curves and the distances to them in order to find the best joint point for the consecutive curves.

In this sense, a speed planning algorithm has also been designed to increase passenger comfort and set continuous speed profiles [35]. The approach permits to improve the comfort in automated vehicles by integrating the speed profile with the previously computed path, constraining the global acceleration in the whole ride (longitudinal and lateral accelerations according to ISO 2631-1). It also minimizes distance error problems by associating the speed profile w.r.t. distance in the path instead of the time. The planner has been tested against other techniques in the state-of-the-art, providing better results.

The proposed architecture has been validated both on simulation (with Pro-Sivic and RTMaps) and on the Inria Rocquencourt terrain. The results showed a smoother tracking of the curves, reduction on the execution times and reduced global accelerations increasing comfort. Future works will improve the capacity to deal with dynamic obstacles, conducting avoidance maneuvers if possible, or returning to the original lane if not. The maneuver will be decided by building an occupancy grid with the information given by the perception system. It will provide the best point near the obstacle to carry out the avoidance trajectory by loading the pre-computed curves.