Faculty of Aerospace Engineering

State-of-the-art sensing technologies provide an accurate estimation of the vehicle’s location and available routes in highway and urban scenarios. Lane markings, road signs, buildings, trees, traffic lights, and many other features enable sensing algorithms to perform localization and build a map of the area. These technologies become less accurate when the amount of visible features is small and when the vehicle is stationary/ driving slowly and performing sharp turns. These challenges are present in an underground parking scenario: the parking lot may have multiple floors, many exits, sharp turns and ramps, markings that are obscured, tight space for maneuvering, short view range and many types of obstacles. Therefore, autonomous valet parking is the next open issue within the rapidly extending performance envelope of autonomous operation. The idea is to leave the vehicle at the entrance to the (previously unknown) parking lot and it should autonomously drive inside it, find an empty slot, park and when called – find the exit and drive out, while the exit might be in a different place than the entrance. The solution to this challenge is completely unconstrained and may take conceptually different approaches, for example:

1. Mapping: the main effort may be directed to constructing a map of the parking lot from visible features and localizing the host vehicle. This might require relying on sensing such features as color of walls/markings, numbers/letters assigned to slots, and may be even requiring dedicated QR codes posted in strategic locations inside the parking lot.
2. Estimation of ego-motion: the main effort may be directed to accurately estimating ego motion from inertial sensors, wheel sensors, and detailed in- ternal model of vehicle dynamics.
3. Replicating Human Behavior: the main effort may be directed to designing an efficient decision making and/or short-term path planning algorithm that allows to navigate inside the parking lot relying only on the closest visible surroundings, similar to how a human would address this task (most likely through trial and error).

In any case, it can be assumed that the basic sensing information is available: the position and speed of other vehicles, pedestrians, and objects relative to the host vehicle, the semantic classification of all visible space into: empty/not empty, detection of some of the markings on the floor, such as slot boundaries and driving direction arrows, detection of exit sign. There is an option to test the developed algorithms in one the autonomous vehicles currently operating in Mobileye Vision Technologies, Jerusalem.