RORL: Robust Offline Reinforcement Learning via Conservative Smoothing

Explore this paper's citation graph

Summary

This work proposes Robust Offline Reinforcement Learning (RORL), a novel conservative smoothing technique that can achieve state-of-the-art performance on the general offline RL benchmark and is considerably robust to adversarial observation perturbations.

Type
preprint
Published
2022-06-06
Cited by
119
References
82
Access
Open access

Keywords

Reinforcement learning, Robustness (evolution), Computer science, Smoothing, Conservatism

References

Cited by

Related papers