Deep reinforcement learning as multiobjective optimization benchmarks: Problem formulation and performance assessment

Open Webpage

Oladayo S. Ajani, Dzeuban Fenyom Ivan, Daison Darlan, Ponnuthurai N. Suganthan, Kaizhou Gao, Rammohan Mallipeddi

Published: 2024, Last Modified: 28 Sept 2025Swarm Evol. Comput. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Deep Reinforcement Learning is formulated as Multi-objective Optimization Problem (MOP).•An Evolutionary Multi-objective Optimization (EMO) benchmark suite is proposed.•The test suite contains 12 Multi-objective DRL tasks with different characteristics.•The proposed benchmark suite is incorporated into an open-source EMO platform.•Baseline results with 7 EMO algorithms from different classes are presented..