Socially Compatible Control Design of Automated Vehicle in Mixed Traffic

A Cooperative Optimal Control Framework for Connected and Automated Vehicles in Mixed Traffic Using Social Value Orientation

“…The goal is to develop a cooperative optimal control framework for CAV-1 which considers the driving intention of HDV-2. In previous research efforts [16], [17], [19], CAV-HDV interaction was modeled as a non-cooperative Stackelberg game, in which a leader makes a decision, then a follower makes its optimal decision with respect to the leader's decision. Generally, computing a Nash equilibrium for the Stackelberg game in CAV-HDV interactions can be computationally expensive for real-time optimization [19].…”

“…In previous research efforts [16], [17], [19], CAV-HDV interaction was modeled as a non-cooperative Stackelberg game, in which a leader makes a decision, then a follower makes its optimal decision with respect to the leader's decision. Generally, computing a Nash equilibrium for the Stackelberg game in CAV-HDV interactions can be computationally expensive for real-time optimization [19]. In addition, the Stackelberg game might not ideally reflect CAV-HDV interaction since determining who should be the leader and the follower is not explicit in many traffic scenarios [17].…”

“…to perform socially-compliant navigation at an intersection given the social preferences of the vehicles. Ozkan and Ma [19] investigated the impacts of a socially compatible control design in a car-following scenario through four different SVO realizations for an automated vehicle. Larsson et al [20] developed a pro-social model predictive control algorithm in which SVO was used to derive weighting strategies.…”

“…Inspired by using SVO to understand the driving preferences of human drivers, in this paper, we develop a socially cooperative optimal control framework for a CAV while interacting with an HDV. Contrary to other efforts that have used the Stackelberg game with a leader-follower structure [16], [17], [19], which may require much computation for finding a Nash equilibrium, our approach models the interaction between the CAV and the HDV as a simultaneous game. In the imposed game, the agents simultaneously take control actions to minimize their objective functions that are given by a weighted sum of their individual (egoistic) objective and a shared (cooperative) one.…”

Inverse Reinforcement Learning Based Stochastic Driver Behavior Learning

“…The driver behavior model is necessary for AVs to understand the intentions of drivers to safely interact with human-operated vehicles in highly interactive human-dominated traffic scenarios. However, existing driver behavior models, such as inverse reinforcement learning (IRL) based models, assume that human drivers are perfect rational decision-makers during the operation of vehicles and consider that human drivers make actions with a fixed planning horizon setting in traffic (Kuderer et al, 2015) (Sadigh et al, 2016) (Gao et al, 2018) (Naumann et al, 2020) (Ozkan and Ma, 2022). However, this assumption is not realistic because of human drivers' behavioral stochasticity and limited cognitive resources to make a decision in a time-varying manner.…”