Formalizing Human Machine Communication in the Context of Autonomous Vehicles

US Transportation Collection2020PDFTech ReportGopalswamy, SwaminathanSaripalli, SrikanthShell, DylanHickman, JeffHsu, Ya-ChuanSafety through Disruption (Safe-D) University Transportation Center (UTC)Virginia Tech Transportation InstituteUnited States. Department of Transportation. Office of the Assistant Secretary for Research and TechnologyUnited States. Department of Transportation. University Transportation Centers (UTC) ProgramTexas. Dept. of TransportationSafety through Disruption (Safe-D) University Transportation Center (UTC)United StatesAutonomous vehiclesBehaviorCommunicationDriversField studiesHuman machine systemsPedestrian vehicle interfacePedestriansProbabilityAutomatic data collection systemsDriver Pedestrian CommunicationVehicle Pedestrian CommunicationHuman-Machine-InteractionHuman-Robot-InteractionMDPPOMDPDSL02-01469A3551747115/Project 02-014Final Research ReportThere are many situations where tacit communication between drivers and pedestrians governs and enhances safety. The goal of this study was to formalize this communication and apply it to the driving strategy of an autonomous vehicle. Toward this, we performed a field study of the interaction between drivers and pedestrians. Vehicles were instrumented to capture behavioral information on a driver as well as passengers and the traffic scenario in general. The data captured were reduced by data analysts to provide insights into the communication and driving patterns. The categorical reduction on driver, pedestrian, and environmental variables was captured. A domain specific language (DSL) was developed to precisely describe the driver-pedestrian behavior, toward the development of a behavioral model for generating autonomous vehicle controls. Specifically, interaction was formalized through a probabilistic model, namely a partially observable Markov decision process (POMDP). This enabled study of what-if scenarios with different risk averseness characteristics. One particular strategy was implemented on an autonomous vehicle and experimental observations were made. Future work will consider (i) richer DSLs to better quantify the driver-human communication, (ii) faster POMDP solvers for real-time operation, and (iii) further applications.109