Multi-Objective Reinforcement Learning for Balancing Efficiency and Safety in Autonomous Decision Making
Afreen Kubra · 2025
This book chapter explores the integration of multi-objective reinforcement learning (MORL) for balancing efficiency and safety in autonomous decision-making. As autonomous systems are increasingly deployed in complex, real-world environments, ensuring both high-performance optimization and safety becomes paramount. The chapter delves into the fundamental concepts of MORL, emphasizing how multiple conflicting objectives can be addressed simultaneously, while maintaining robust safety standards. Various methods, such as risk-sensitive learning, safety constraints, and reward shaping, are examined to effectively guide autonomous agents in navigating trade-offs between exploration and safety. Additionally, the chapter discusses policy-based methods, verification, and validation techniques crucial for real-world implementation. With a focus on real-world testing for safety assurance, it presents strategies for integrating safety-critical components in decision-making processes. This work provides valuable insights into enhancing the reliability, performance, and safety of autonomous systems in dynamic environments.