A Brief Guide to Multi-Objective Reinforcement Learning and Planning: JAAMAS Track ■
Conor F. Hayes, Conor F. Hayes, Conor F. Hayes, Roxana Radulescu, Roxana Radulescu, Roxana Radulescu, Eugenio Bargiacchi, Eugenio Bargiacchi, Eugenio Bargiacchi, Johan Källström, Johan Källström, Johan Källström, Matthew Macfarlane, Matthew Macfarlane, Matthew Macfarlane, Mathieu Reymond, Mathieu Reymond, Mathieu Reymond, Timothy Verstraeten, Timothy Verstraeten, Timothy Verstraeten, Luisa Zintgraf, Luisa Zintgraf, Luisa Zintgraf, Richard Dazeley, Richard Dazeley, Richard Dazeley, Fredrik Heintz, Fredrik Heintz, Fredrik Heintz, Enda Howley, Enda Howley, Enda Howley, Athirai A. Irissappane, Athirai A. Irissappane, Athirai A. Irissappane, Patrick Mannion, Patrick Mannion, Patrick Mannion, Ann Nowe, Ann Nowe, Ann Nowe, Gabriel De Oliveira Ramos, Gabriel De Oliveira Ramos, Gabriel De Oliveira Ramos, Marcello Restelli, Marcello Restelli, Marcello Restelli, Peter Vamplew, Peter Vamplew, Peter Vamplew, Diederik M. Roijers, Diederik M. Roijers, Diederik M. Roijers
Real-world sequential decision-making tasks are usually complex, and require trade-offs between multiple -- often conflicting -- objectives. However, the majority of research in reinforcement learning (RL) and decision-theoretic planning assumes a single objective, or that multiple objectives can be handled via a predefined weighted sum over the objectives. Such approaches may oversimplify the underlying problem, and produce suboptimal results. This extended abstract outlines the limitations of using a semi-blind iterative process to solve multi-objective decision making problems. Our extended paper serves as a guide for the application of explicitly multi-objective methods to difficult problems.