Efficient RL via Disentangled Environment and Agent Representations

Kevin Gmelin, Shikhar Bahl, R. Mendonca, Deepak Pathak
2/12/2026

No official code found. Best implementation: sear-rl/sear-rl (9 stars, MIT, last commit 2024-01-29).

Abstract

Agents that are aware of the separation between themselves and their environments can leverage this understanding to form effective representations of visual input. We propose an approach for learning such structured representations for RL algorithms, using visual knowledge of the agent, such as its shape or mask, which is often inexpensive to obtain. This is incorporated into the RL objective using a simple auxiliary loss. We show that our method, Structured Environment-Agent Representations, outperforms state-of-the-art model-free approaches over 18 different challenging visual simulation environments spanning 5 different robots. Website at https://sear-rl.github.io/

View on arXivView PDF

Code Implementations(1)

sear-rl/sear-rlCommunity61%
MIT

Efficient RL via Disentangled Environment and Agent Representations

90PythonJan 29, 20242 years agoMIT

Similar Papers

DEAR: Disentangled Environment and Agent Representations for Reinforcement Learning without Reconstruction

88% match1 repo12024

DeGuV: Depth-Guided Visual Reinforcement Learning for Generalization and Interpretability in Manipulation

82% match2025

Render and Diffuse: Aligning Image and Action Spaces for Diffusion-based Behaviour Cloning

82% match2024

Make the Pertinent Salient: Task-Relevant Reconstruction for Visual Control with Distractions

82% match2024

RePo: Resilient Model-Based Reinforcement Learning by Regularizing Posterior Predictability

81% match2026

Lift3D Policy: Lifting 2D Foundation Models for Robust 3D Robotic Manipulation

81% match1 repo1852026

Cite this paper

@article{gmelin2026efficient,
  title  = {Efficient RL via Disentangled Environment and Agent Representations},
  author = {Kevin Gmelin and Shikhar Bahl and R. Mendonca and Deepak Pathak},
  year   = {2026},
  eprint = {2309.02435},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/pdf/2309.02435},
  journal = {ICML 2023 2023}
}

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/2309.02435.svg)](https://codeofpaper.com/paper/2309.02435)

Discussion