Incorporating Social Information Into An Autonomous Vehicle's Decision-Making Process and Control

Download or read book Incorporating Social Information Into An Autonomous Vehicle's Decision-Making Process and Control written by Kasra Mokhtari and published by . This book was released on 2021 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: How can autonomous vehicles offer safer behavior by accounting for social information? Social information includes not only information about the number of pedestrians, but also pedestrians' behavior, age, course of action, etc. While driving, the interaction of a vehicle and the other road users is complicated because each operator acts dynamically and according to their own will, thus creating additional uncertainties for an autonomous vehicle to consider. To address some of these uncertainties and to avoid collisions human drivers use a variety of tricks and heuristics learned during their time driving. However, substituting human drivers with autonomous control systems comes at the price of eliminating the underlying social intelligence of human drivers that makes these predictions possible. Steps should, therefore, be taken to imbue autonomous vehicles with the ability to use social information to increase safety since information about the social environment may provide autonomous vehicles with valuable data influencing how these systems select and moderate their actions. This dissertation develops well-defined methods that will enable an autonomous vehicle to use social information to adjust the vehicle's course of action with the hope of providing a much safer environment for pedestrians, other car drivers, and AV passengers. We first generate our social information dataset by repeatedly driving in State College, PA along the different paths. We then present an initial examination of how social information (i.e. pedestrian density) could be used first for path recognition and then for predicting the number of pedestrians that the vehicle will encounter in the future which is intuitively related to the risk of traveling down a path for autonomous vehicles. Moreover, we develop a method for an AV operating near a college campus to evaluate the risk associated with different options and to select the minimal risk option in the hope of improving safety. We then design a decision-making framework for controlling an autonomous vehicle as it navigates through an unsignalized intersection crowded with pedestrians in both cases where it receives true state of the environment and noisy observations. We hope that the research presented in this dissertation will inspire future researchers to develop autonomous vehicles that more intelligently and efficiently account for pedestrian information in their decision-making framework to make a collision-free world.