Exploration Methods in Sparse Reward Environments
2021
Reinforcement learning (RL) usually performs best in domains with dense reward signal. When the reward signal is sparse, RL algorithms may perform poorly, especially if naive exploration methods such as \(\epsilon \)-greedy are employed. With this paper, we outline the problems caused by sparse reward signals and survey different exploration strategies which are able to address them.
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
54
References
0
Citations
NaN
KQI