2020/04/27 by Christian Muench, Muench, Christian, Frans A. Oliehoek +3
Computer Science · Energy · Engineering · #Autonomous Vehicle Technology and Safety #Bayesian Modeling and Causal Inference #Energy, Environment, and Transportation Policies #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Traffic control and management
paper · pdf · doi:10.48550/arxiv.2004.12678
openalex publication_date 2020/04/27 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Traffic scenarios are inherently interactive. Multiple decision-makers\npredict the actions of others and choose strategies that maximize their\nrewards. We view these interactions from the perspective of game theory which\nintroduces various challenges. Humans are not entirely rational, their rewards\nneed to be inferred from real-world data, and any prediction algorithm needs to\nbe real-time capable so that we can use it in an autonomous vehicle (AV). In\nthis work, we present a game-theoretic method that addresses all of the points\nabove. Compared to many existing methods used for AVs, our approach does 1) not\nrequire perfect communication, and 2) allows for individual rewards per agent.\nOur experiments demonstrate that these more realistic assumptions lead to\nqualitatively and quantitatively different reward inference and prediction of\nfuture actions that match better with expected real-world behaviour.\n