ICML 2022spotlight48 citations

Stabilizing Off-Policy Deep Reinforcement Learning from Pixels

Edoardo Cetin, Philip J Ball, Stephen Roberts, Oya Celiktutan

Abstract

Off-policy reinforcement learning (RL) from pixel observations is notoriously unstable. As a result, many successful algorithms must combine different domain-specific practices and auxiliary losses to learn meaningful behaviors in complex environments. In this work, we provide novel analysis demonstrating that these instabilities arise from performing temporal-difference learning with a convolutional encoder and low-magnitude rewards. We show that this new visual deadly triad causes unstable training and premature convergence to degenerate solutions, a phenomenon we name catastrophic self-overfitting. Based on our analysis, we propose A-LIX, a method providing adaptive regularization to the encoder’s gradients that explicitly prevents the occurrence of catastrophic self-overfitting using a dual objective. By applying A-LIX, we significantly outperform the prior state-of-the-art on the DeepMind Control and Atari benchmarks without any data augmentation or auxiliary losses.

BibTeX
@InProceedings{pmlr-v162-cetin22a,
  title = 	 {Stabilizing Off-Policy Deep Reinforcement Learning from Pixels},
  author =       {Cetin, Edoardo and Ball, Philip J and Roberts, Stephen and Celiktutan, Oya},
  booktitle = 	 {Proceedings of the 39th International Conference on Machine Learning},
  pages = 	 {2784--2810},
  year = 	 {2022},
  editor = 	 {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan},
  volume = 	 {162},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {17--23 Jul},
  publisher =    {PMLR},
  pdf = 	 {https://proceedings.mlr.press/v162/cetin22a/cetin22a.pdf},
  url = 	 {https://proceedings.mlr.press/v162/cetin22a.html},
  abstract = 	 {Off-policy reinforcement learning (RL) from pixel observations is notoriously unstable. As a result, many successful algorithms must combine different domain-specific practices and auxiliary losses to learn meaningful behaviors in complex environments. In this work, we provide novel analysis demonstrating that these instabilities arise from performing temporal-difference learning with a convolutional encoder and low-magnitude rewards. We show that this new visual deadly triad causes unstable training and premature convergence to degenerate solutions, a phenomenon we name catastrophic self-overfitting. Based on our analysis, we propose A-LIX, a method providing adaptive regularization to the encoder’s gradients that explicitly prevents the occurrence of catastrophic self-overfitting using a dual objective. By applying A-LIX, we significantly outperform the prior state-of-the-art on the DeepMind Control and Atari benchmarks without any data augmentation or auxiliary losses.}
}
Stabilizing Off-Policy Deep Reinforcement Learning from Pixels · ICML 2022