Zero-Shot Reinforcement Learning Under Partial Observability

Scott Jeen, Tom Bewley, Jonathan M. Cullen
6/18/2025
cs.LGcs.AI

No official code found. Best implementation: enjeeneer/bfms-with-memory (8 stars, MIT, last commit 2025-06-29).

Abstract

Recent work has shown that, under certain assumptions, zero-shot reinforcement learning (RL) methods can generalise to any unseen task in an environment after reward-free pre-training. Access to Markov states is one such assumption, yet, in many real-world applications, the Markov state is only partially observable. Here, we explore how the performance of standard zero-shot RL methods degrades when subjected to partially observability, and show that, as in single-task RL, memory-based architectures are an effective remedy. We evaluate our memory-based zero-shot RL methods in domains where the states, rewards and a change in dynamics are partially observed, and show improved performance over memory-free baselines. Our code is open-sourced via: https://enjeeneer.io/projects/bfms-with-memory/.

View on arXivView PDF

Code Implementations(1)

enjeeneer/bfms-with-memoryCommunity61%
MIT

Official implementation of FB/HILP with memory from "Zero-Shot Reinforcement Learning Under Partial Observability" (RLC 2025)

80Shell, PythonJun 8, 20251 years agoMIT

Similar Papers

On Zero-Shot Reinforcement Learning

87% match2025

Zero-Shot Reinforcement Learning from Low Quality Data

87% match1 repo292026

Benchmarking Partial Observability in Reinforcement Learning with a Suite of Memory-Improvable Domains

85% match2025

Provable Representation with Efficient Planning for Partially Observable Reinforcement Learning

83% match2026

Rethinking State Disentanglement in Causal Reinforcement Learning

81% match2024

Partially Observable Reinforcement Learning with Memory Traces

81% match2025

Cite this paper

@article{jeen2025zeroshot,
  title  = {Zero-Shot Reinforcement Learning Under Partial Observability},
  author = {Scott Jeen and Tom Bewley and Jonathan M. Cullen},
  year   = {2025},
  eprint = {2506.15446},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2506.15446v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2506.15446.svg)](https://codeofpaper.com/paper/2506.15446)

Discussion