An integrated MPC decision-making method based on MDP for autonomous driving in urban traffic

The decision-making module plays a critical role in autonomous vehicles (AVs). There are two main challenges in decision-making for autonomous driving: accurately predicting and reliably reacting to evolving environments. This work ad- dresses these challenges by implementing an integrated decision- making method that merges the Markov decision process (MDP) with model predictive control (MPC) structure. This method ensures that optimal and safe decision actions can be generated in real-time by solving the MPC optimization problem, subject to conditions such as environmental evolution, dynamics of continuous systems, MDP state transitions, and safety constraints. To validate the decision-making method, an information-rich urban crossroad scenario, including traffic signals, other vehicles, pedestrians, cyclists, has been considered for performance testing. The effectiveness and reliability of the decision-making method have been demonstrated through these highly variable urban environments.