# Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability

arXiv: https://arxiv.org/abs/2107.06277 · HTML: https://mlsanity.com/paper/2107.06277/

## Key facts

- Cited by 109 papers in the map (data through 14 Oct 2026).
- Builds on 41 of its 58 references that are in the map.
- First posted on arXiv 13 Jul 2021.

## Builds on (most cited first, 30 of 41)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms](https://mlsanity.com/paper/1708.07747/) (2017)
- [Addressing Function Approximation Error in Actor-Critic Methods](https://mlsanity.com/paper/1802.09477/) (2018)
- [Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems](https://mlsanity.com/paper/2005.01643/) (2020)
- [Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World](https://mlsanity.com/paper/1703.06907/) (2017)
- [Solving Rubik's Cube with a Robot Hand](https://mlsanity.com/paper/1910.07113/) (2019)
- [Sim-to-Real Transfer of Robotic Control with Dynamics Randomization](https://mlsanity.com/paper/1710.06537/) (2017)
- [Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels](https://mlsanity.com/paper/2004.13649/) (2020)
- [Deep Exploration via Bootstrapped DQN](https://mlsanity.com/paper/1602.04621/) (2016)
- [Leveraging Procedural Generation to Benchmark Reinforcement Learning](https://mlsanity.com/paper/1912.01588/) (2019)
- [Learning Invariant Representations for Reinforcement Learning without Reconstruction](https://mlsanity.com/paper/2006.10742/) (2020)
- [Reinforcement Learning with Unsupervised Auxiliary Tasks](https://mlsanity.com/paper/1611.05397/) (2016)
- [Quantifying Generalization in Reinforcement Learning](https://mlsanity.com/paper/1812.02341/) (2018)
- [Decoupling Representation Learning from Reinforcement Learning](https://mlsanity.com/paper/2009.08319/) (2020)
- [Learning to Optimize Via Posterior Sampling](https://mlsanity.com/paper/1301.2609/) (2013)
- [(More) Efficient Reinforcement Learning via Posterior Sampling](https://mlsanity.com/paper/1306.0940/) (2013)
- [CAD2RL: Real Single-Image Flight without a Single Real Image](https://mlsanity.com/paper/1611.04201/) (2016)
- [Distral: Robust Multitask Reinforcement Learning](https://mlsanity.com/paper/1707.04175/) (2017)
- [VariBAD: A Very Good Method for Bayes-Adaptive Deep RL via Meta-Learning](https://mlsanity.com/paper/1910.08348/) (2019)
- [The NetHack Learning Environment](https://mlsanity.com/paper/2006.13760/) (2020)
- [Prioritized Level Replay](https://mlsanity.com/paper/2010.03934/) (2020)
- [A Study on Overfitting in Deep Reinforcement Learning](https://mlsanity.com/paper/1804.06893/) (2018)
- [EPOpt: Learning Robust Neural Network Policies Using Model Ensembles](https://mlsanity.com/paper/1610.01283/) (2016)
- [Contrastive Behavioral Similarity Embeddings for Generalization in Reinforcement Learning](https://mlsanity.com/paper/2101.05265/) (2021)
- [Generalization and Regularization in DQN](https://mlsanity.com/paper/1810.00123/) (2018)
- [Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning](https://mlsanity.com/paper/1910.05396/) (2019)
- [Generalization in Reinforcement Learning with Selective Noise Injection and Information Bottleneck](https://mlsanity.com/paper/1910.12911/) (2019)
- [The Distracting Control Suite -- A Challenging Benchmark for Reinforcement Learning from Pixels](https://mlsanity.com/paper/2101.02722/) (2021)
- [Context-aware Dynamics Model for Generalization in Model-Based Reinforcement Learning](https://mlsanity.com/paper/2005.06800/) (2020)
- [Observational Overfitting in Reinforcement Learning](https://mlsanity.com/paper/1912.02975/) (2019)

## Built on by (most cited first, 30 of 109)

- [The Rise and Potential of Large Language Model Based Agents: A Survey](https://mlsanity.com/paper/2309.07864/) (2023)
- [Training Language Models to Self-Correct via Reinforcement Learning](https://mlsanity.com/paper/2409.12917/) (2024)
- [WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning](https://mlsanity.com/paper/2411.02337/) (2024)
- [DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning](https://mlsanity.com/paper/2406.11896/) (2024)
- [Evolving Curricula with Regret-Based Environment Design](https://mlsanity.com/paper/2203.01302/) (2022)
- [A Survey of Zero-shot Generalisation in Deep Reinforcement Learning](https://mlsanity.com/paper/2111.09794/) (2021)
- [Large Language Models for Robotics: A Survey](https://mlsanity.com/paper/2311.07226/) (2023)
- [Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning](https://mlsanity.com/paper/2503.07572/) (2025)
- [Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought](https://mlsanity.com/paper/2501.04682/) (2025)
- [POPGym: Benchmarking Partially Observable Reinforcement Learning](https://mlsanity.com/paper/2303.01859/) (2023)
- [e3: Learning to Explore Enables Extrapolation of Test-Time Compute for LLMs](https://mlsanity.com/paper/2506.09026/) (2025)
- [Offline RL Policies Should be Trained to be Adaptive](https://mlsanity.com/paper/2207.02200/) (2022)
- [Contextualize Me -- The Case for Context in Reinforcement Learning](https://mlsanity.com/paper/2202.04500/) (2022)
- [PartManip: Learning Cross-Category Generalizable Part Manipulation Policy from Point Cloud Observations](https://mlsanity.com/paper/2303.16958/) (2023)
- [AMAGO: Scalable In-Context Reinforcement Learning for Adaptive Agents](https://mlsanity.com/paper/2310.09971/) (2023)
- [RaC: Robot Learning for Long-Horizon Tasks by Scaling Recovery and Correction](https://mlsanity.com/paper/2509.07953/) (2025)
- [CORA: Benchmarks, Baselines, and Metrics as a Platform for Continual Reinforcement Learning Agents](https://mlsanity.com/paper/2110.10067/) (2021)
- [Tell me why! Explanations support learning relational and causal structure](https://mlsanity.com/paper/2112.03753/) (2021)
- [On the Importance of Exploration for Generalization in Reinforcement Learning](https://mlsanity.com/paper/2306.05483/) (2023)
- [Don't Start From Scratch: Leveraging Prior Data to Automate Robotic Reinforcement Learning](https://mlsanity.com/paper/2207.04703/) (2022)
- [Disentangling Epistemic and Aleatoric Uncertainty in Reinforcement Learning](https://mlsanity.com/paper/2206.01558/) (2022)
- [Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization](https://mlsanity.com/paper/2311.03351/) (2023)
- [The Generalization Gap in Offline Reinforcement Learning](https://mlsanity.com/paper/2312.05742/) (2023)
- [A Comprehensive Survey of Data Augmentation in Visual Reinforcement Learning](https://mlsanity.com/paper/2210.04561/) (2022)
- [Explore to Generalize in Zero-Shot RL](https://mlsanity.com/paper/2306.03072/) (2023)
- [Generalization in Cooperative Multi-Agent Systems](https://mlsanity.com/paper/2202.00104/) (2022)
- [Learning to Extrapolate: A Transductive Approach](https://mlsanity.com/paper/2304.14329/) (2023)
- [Incremental Learning of Retrievable Skills For Efficient Continual Task Adaptation](https://mlsanity.com/paper/2410.22658/) (2024)
- [Towards Interactive and Learnable Cooperative Driving Automation: a Large Language Model-Driven Decision-Making Framework](https://mlsanity.com/paper/2409.12812/) (2024)
- [AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers](https://mlsanity.com/paper/2411.11188/) (2024)

## Abstract (arXiv)

Generalization is a central challenge for the deployment of reinforcement learning (RL) systems in the real world. In this paper, we show that the sequential structure of the RL problem necessitates new approaches to generalization beyond the well-studied techniques used in supervised learning. While supervised learning methods can generalize effectively without explicitly accounting for epistemic uncertainty, we show that, perhaps surprisingly, this is not the case in RL. We show that generalization to unseen test conditions from a limited number of training conditions induces implicit partial observability, effectively turning even fully-observed MDPs into POMDPs. Informed by this observation, we recast the problem of generalization in RL as solving the induced partially observed Markov decision process, which we call the epistemic POMDP. We demonstrate the failure modes of algorithms that do not appropriately handle this partial observability, and suggest a simple ensemble-based technique for approximately solving the partially observed problem. Empirically, we demonstrate that our simple algorithm derived from the epistemic POMDP achieves significant gains in generalization over current methods on the Procgen benchmark suite.