Multi Objective Markov Decision Process

Multi-objective Markov Decision Processes (MOMDPs) address sequential decision-making problems with multiple, often conflicting, objectives. Current research focuses on developing efficient algorithms, such as those based on deep Q-networks and natural policy gradients, to find optimal or Pareto-optimal policies, often incorporating preference inference techniques from demonstrations or partially ordered preferences to handle unknown or complex human objectives. These advancements are significant for various applications, including autonomous driving, healthcare resource allocation, and robotics, where optimizing multiple criteria simultaneously is crucial for effective and safe system design.