Reports value function interference and overestimation sensitivity as issues hindering value-based MORL with non-linear scalarisation, shown via tabular Q-learning on simple multi-objective MDPs.
Multi-criteria reinforcement learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Issues with Value-Based Multi-objective Reinforcement Learning: Value Function Interference and Overestimation Sensitivity
Reports value function interference and overestimation sensitivity as issues hindering value-based MORL with non-linear scalarisation, shown via tabular Q-learning on simple multi-objective MDPs.