Deep Reinforcement Learning via Object-Centric Attention
Not on the map. Why?
Abstract (arXiv)
Deep reinforcement learning agents, trained on raw pixel inputs, often fail to generalize beyond their training environments, relying on spurious correlations and irrelevant background details. To address this issue, object-centric agents have recently emerged. However, they require different representations tailored to the task specifications. Contrary to deep agents, no single object-centric architecture can be applied to any environment. Inspired by principles of cognitive science and Occam's Razor, we introduce Object-Centric Attention via Masking (OCCAM), which selectively preserves task-relevant entities while filtering out irrelevant visual information. Specifically, OCCAM takes advantage of the object-centric inductive bias. Empirical evaluations on Atari benchmarks demonstrate that OCCAM significantly improves robustness to novel perturbations and reduces sample complexity while showing similar or improved performance compared to conventional pixel-based RL. These results suggest that structured abstraction can enhance generalization without requiring explicit symbolic representations or domain-specific object extraction pipelines.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 41 of its 57 references that are in the map. First posted on arXiv 3 Apr 2025.
Builds on
41 of 57 references.
References · 41
- 2017Proximal Policy Optimization Algorithmscited by 14,441
- 2017Mask R-CNNcited by 7,259
- 2015You Only Look Once: Unified, Real-Time Object Detectioncited by 4,829
- 2013Playing Atari with Deep Reinforcement Learningcited by 3,211
- 2015Spatial Transformer Networkscited by 1,423
- 2015Show, Attend and Tell: Neural Image Caption Generation with Visual Attentioncited by 1,421
- 2017Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real Worldcited by 1,401
- 2019Mastering Atari, Go, Chess and Shogi by Planning with a Learned Modelcited by 1,327
- 2022DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detectioncited by 1,306
- 2020Object-Centric Learning with Slot Attentioncited by 865
- 2021Deep Reinforcement Learning at the Edge of the Statistical Precipicecited by 781
- 2017A Distributional Perspective on Reinforcement Learningcited by 743
29 more
- 2020Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixelscited by 573
- 2020Learning Invariant Representations for Reinforcement Learning without Reconstructioncited by 424
- 2018Graph R-CNN for Scene Graph Generationcited by 296
- 2020Inductive Biases for Deep Learning of Higher-Level Cognitioncited by 273
- 2020SPACE: Unsupervised Object-Oriented Scene Representation via Spatial Attention and Decompositioncited by 190
- 2024Stop Regressing: Training Value Functions via Classification for Scalable Deep RLcited by 136
- 2018Generalization and Regularization in DQNcited by 117
- 2017The Consciousness Priorcited by 105
- 2019Towards Interpretable Reinforcement Learning Using Attention Augmented Agentscited by 82
- 2021Generalization and Robustness Implications in Object-Centric Learningcited by 81
- 2023On the Foundations of Shortcut Learningcited by 64
- 2023An Investigation into Pre-Training Object-Centric Representations for Reinforcement Learningcited by 49
- 2022Look where you look! Saliency-guided Q-networks for generalization in visual Reinforcement Learningcited by 32
- 2023OCAtari: Object-Centric Atari 2600 Reinforcement Learning Environmentscited by 30
- 2020Discriminative Particle Filter Reinforcement Learning for Complex Partial Observationscited by 27
- 2021A Consciousness-Inspired Planning Agent for Model-Based Reinforcement Learningcited by 27
- 2024Interpretable Concept Bottlenecks to Align Reinforcement Learning Agentscited by 23
- 2024Interpretable and Editable Programmatic Tree Policies for Reinforcement Learningcited by 18
- 2019Reinforcement Learning with Attention that Works: A Self-Supervised Approachcited by 17
- 2022Boosting Object Representation Learning via Motion and Object Continuitycited by 17
- 2024End-to-End Neuro-Symbolic Reinforcement Learning with Textual Explanationscited by 17
- 2021Graphhopper: Multi-Hop Scene Graph Reasoning for Visual Question Answeringcited by 13
- 2021A Review for Deep Reinforcement Learning in Atari:Benchmarks, Challenges, and Solutionscited by 12
- 2024HackAtari: Atari Learning Environments for Robust and Continual Reinforcement Learningcited by 11
- 2023Representation Matters for Mastering Chess: Improved Feature Representation in AlphaZero Outperforms Switching to Transformerscited by 8
- 2020Investigating Simple Object Representations in Model-Free Deep Reinforcement Learningcited by 5
- 2024Interpretable end-to-end Neurosymbolic Reinforcement Learning agentscited by 4
- 2020Relevance-Guided Modeling of Object Dynamics for Reinforcement Learningcited by 3
- 2024Contrastive Abstraction for Reinforcement Learningcited by 2
Built on by
Citing papers per year: 2025: 1; 2026: 1.
2 papers.