On the Limitations of Scalarisation for Multi-objective Reinforcement Learning of Pareto Fronts

Peter Vamplew, John Yearwood, Richard Dazeley, Adam Berry

Abstract

Multiobjective reinforcement learning (MORL) extends RL to problems with multiple conflicting objectives. This paper argues for designing MORL systems to produce a set of solutions approximating the Pareto front, and shows that the common MORL technique of scalarisation has fundamental limitations when used to find Pareto-optimal policies. The work is supported by the presentation of three new MORL benchmarks with known Pareto fronts.
BibTeX
@inproceedings{vamplew2008limitations,
  author = {Peter Vamplew and John Yearwood and Richard Dazeley and Adam Berry},
  title = {{On the Limitations of Scalarisation for Multi-objective Reinforcement Learning of Pareto Fronts}},
  booktitle = {Australasian Conference on Artificial Intelligence},
  year = {2008},
  doi = {10.1007/978-3-540-89378-3\_37},
}

Loading related content...

Figure