# StarWM: Self-Supervised Trained Attention Routing for Robust World Models

arXiv: https://arxiv.org/abs/2609.30667 · HTML: https://mlsanity.com/paper/2609.30667/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Video-generation world models and simulators, one of 11 directions in World models.
- Builds on 33 of its 40 references that are in the map.
- First posted on arXiv 25 Sep 2026.

## Placed in

- World models / [Video-generation world models and simulators](https://mlsanity.com/wm/video-generation-world-models-and-simulators/)

## Builds on (most cited first, 30 of 33)

- [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018)
- [End-to-End Object Detection with Transformers](https://mlsanity.com/paper/2005.12872/) (2020)
- [Dream to Control: Learning Behaviors by Latent Imagination](https://mlsanity.com/paper/1912.01603/) (2019)
- [World Models](https://mlsanity.com/paper/1803.10122/) (2018)
- [Curiosity-driven Exploration by Self-supervised Prediction](https://mlsanity.com/paper/1705.05363/) (2017)
- [Mastering Atari with Discrete World Models](https://mlsanity.com/paper/2010.02193/) (2020)
- [Object-Centric Learning with Slot Attention](https://mlsanity.com/paper/2006.15055/) (2020)
- [Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://mlsanity.com/paper/2108.13264/) (2021)
- [CURL: Contrastive Unsupervised Representations for Reinforcement Learning](https://mlsanity.com/paper/2004.04136/) (2020)
- [TD-MPC2: Scalable, Robust World Models for Continuous Control](https://mlsanity.com/paper/2310.16828/) (2023)
- [Learning Invariant Representations for Reinforcement Learning without Reconstruction](https://mlsanity.com/paper/2006.10742/) (2020)
- [Temporal Difference Learning for Model Predictive Control](https://mlsanity.com/paper/2203.04955/) (2022)
- [dm_control: Software and Tasks for Continuous Control](https://mlsanity.com/paper/2006.12983/) (2020)
- [Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning](https://mlsanity.com/paper/2107.09645/) (2021)
- [Data-Efficient Reinforcement Learning with Self-Predictive Representations](https://mlsanity.com/paper/2007.05929/) (2020)
- [An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution](https://mlsanity.com/paper/1807.03247/) (2018)
- [Illiterate DALL-E Learns to Compose](https://mlsanity.com/paper/2110.11405/) (2021)
- [SlotFormer: Unsupervised Visual Dynamics Simulation with Object-Centric Models](https://mlsanity.com/paper/2210.05861/) (2022)
- [Generalization in Reinforcement Learning by Soft Data Augmentation](https://mlsanity.com/paper/2011.13389/) (2020)
- [DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations](https://mlsanity.com/paper/2110.14565/) (2021)
- [Denoised MDPs: Learning World Models Better Than the World Itself](https://mlsanity.com/paper/2206.15477/) (2022)
- [Learning Task Informed Abstractions](https://mlsanity.com/paper/2106.15612/) (2021)
- [Iso-Dream: Isolating and Leveraging Noncontrollable Visual Dynamics in World Models](https://mlsanity.com/paper/2205.13817/) (2022)
- [Unsupervised Visual Attention and Invariance for Reinforcement Learning](https://mlsanity.com/paper/2104.02921/) (2021)
- [Which Mutual-Information Representation Learning Objectives are Sufficient for Control?](https://mlsanity.com/paper/2106.07278/) (2021)
- [Sample-Efficient Reinforcement Learning in the Presence of Exogenous Information](https://mlsanity.com/paper/2206.04282/) (2022)
- [Learning Latent Dynamic Robust Representations for World Models](https://mlsanity.com/paper/2405.06263/) (2024)
- [A Recipe for Unbounded Data Augmentation in Visual Reinforcement Learning](https://mlsanity.com/paper/2405.17416/) (2024)
- [Generalizable Visual Reinforcement Learning with Segment Anything Model](https://mlsanity.com/paper/2312.17116/) (2023)
- [R2-Dreamer: Redundancy-Reduced World Models without Decoders or Augmentation](https://mlsanity.com/paper/2603.18202/) (2026)

## Abstract (arXiv)

A robust world model must strike the balance between faithfully capturing environmental dynamics and abstracting away from irrelevant content. While reconstruction-based world models ensure faithful supervision, they misallocate representational capacity by pixel area rather than dynamics relevance for visual tasks, which can cause task-irrelevant content to dominate the learned representation. Alternatively, reconstruction-free methods avoid this bias but risk discarding possibly relevant information. We propose StarWM, which uses a cross-attention module trained on self-supervised dynamics to decide where reconstruction applies. A dual-stream decoder then restricts reconstruction to the attended regions, with stop-gradient barriers preventing interference between the two objectives. These components allows reconstruction to supervise the visual content of attended regions without contaminating the latent with non-predictive information. On DeepMind Control with dynamic video backgrounds, default (reward-free) StarWM achieves the strongest performance under random-frame distractors and substantially outperforms reconstruction-based baselines under sequential video. In addition, its reward-augmented variant matches or exceeds reconstruction-free methods on sequential video, achieving the highest overall return across all distractor regimes. Mechanistic probing confirms StarWM preserves state attributes with near-perfect fidelity through long-horizon imagination while systematically discarding distractors.