Project: #12 Prediction and Behaviors for Driver Assistance and Socially Cooperative Autonomous Driving Progress Report - Reporting Period Ending: March 30, 2018 Principal Investigator: John Dolan Status: Active Start Date: Jan. 1, 2017 End Date: Aug. 31, 2018 Research Type: None Grant Type: Technology Transfer Grant Program: FAST Act Grant Cycle: 2017 Mobility21 UTC Progress Report (Last Updated: March 30, 2018, 1:19 p.m.) % Project Completed to Date: 50 % Grant Award Expended: 41 % Match Expended & Document: 0 USDOT Requirements Accomplishments A Probabilistic Graphical Model (PGM)-based method for estimating the yield/not yield intention of a car merging onto a main road from an entrance ramp was developed. This method was extended to include the ability of the egovehicle to handle a leading vehicle in front of it in the main-road lane and more than one merging vehicle on the entrance ramp. The methods were validated by extracting merge-car entrance ramp behavior models from the NGSIM dataset, then replaying the NGSIM data and governing the egovehicle behavior by a combination of the intention estimation PGM with a standard ACC lane-keeping controller for the egovehicle. The ability of the egovehicle to safely navigate the merge point was compared to that of multiple other methods, and our method achieved the lowest collision rate and greatest similarity to human drivers based on multiple measures, including Kullback-Leibler distance. Impacts The primary impact of the project's work is improved ability of an autonomous car to deal safely with merging traffic in highway or urban situations. The use of existing ACC systems would lead to a 17% collision rate on the NGSIM dataset, whereas our method reduces this to around 2%. We are currently investigating improvements to our method that can further reduce this number, whether by increased sophistication of the algorithm, or by the introduction of heuristics dealing with the pathological cases, if these can be categorized in an exhaustive manner. We are also planning to add in the next reporting period the consideration of vehicle behaviors when navigating traffic circles, or roundabouts. Other We have created a human driving testbed based on a Logitech steering wheel and pedal system coupled with the VIRES driving simulator in order to collect human driving data in various scenarios. New Partners None Issues None