Probabilistic Game Theory and Stochastic Model Predictive Control-Based Decision Making and Motion Planning in Uncontrolled Intersections for Autonomous Driving

This paper presents a motion planner of autonomous vehicles for uncontrolled intersection driving. Uncertainty about the surrounding target vehicles is an important consideration for decision-making and motion planning for autonomous driving. An interactive Multiple Model (IMM) filter is used to estimate the maneuver probability of target vehicles by using the measurement of environment sensors. An intelligent driver model is utilized to reflect the asymmetry of the longitudinal acceleration to a local model of the IMM filter. Estimates of the drivers' intentions are used to estimate the probability of the possible driving scenes. A game theory-based supervisor is configured based on the probability of the driving scene and uncontrolled intersection theory to determine the driving mode. A payoff for each driving scene is designed to consider the risk and traffic efficiency based on the prediction of the targets. Local filters of IMM are used to predict the future motion of the targets for evaluation of the payoff function. A stochastic model predictive controller is utilized to determine the desired longitudinal acceleration. Chance constraints are defined to reflect the uncertainty of the target prediction in motion planning. The proposed algorithm is evaluated via case studies and Monte Carlo simulation. The simulation results show the probabilistic game theory-based motion planner achieves the safety and efficiency of autonomous intersection passing.