Multi-objective Reinforcement Learning

Most real decisions involve trade-offs: speed vs. safety, cost vs. quality, efficiency vs. fairness. Multi-objective reinforcement learning (MORL) studies how to train agents that learn to navigate these trade-offs rather than collapse them into a single score. Our work develops algorithmic frameworks that combine evolutionary computation with reinforcement learning to produce sets of policies representing different trade-off solutions, with applications in logistics, finance, and autonomous systems.

Funded by PAPIIT IA102025, UNAM.