Better Slots, Better Worlds: Representation Quality & Robustness in Object-Centric World Models

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Shukrullo Nazirjonov, Sai Prasanna, Anna Manasyan, Georg Martius
• Venue: arXiv: cs.CV
• Published: 2026-08-12
• Collected: 2026-08-13
• URL: http://arxiv.org/abs/2608.12078v1
• PDF: https://arxiv.org/pdf/2608.12078v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The study investigates the performance of object-centric world models for visual model-predictive control, which enable agents to accomplish tasks through planning. The researchers find that these models perform better with high quality object-centric representations, and become robust under distribution shifts compared to scene-centric models when slots are well-bound.

What the Document Says

• The research focuses on learning world models from offline trajectories for accomplishing tasks through planning using object-centric (OC) representations.
• Prior OC world models (OCWMs) rely on a given slot encoder and are only evaluated in-distribution.
• The study conducts a controlled investigation of OCWMs along two axes: object-centric representation quality and generalization under distribution shift relative to scene-centric models.
• Planning success correlates positively with unsupervised slot-quality metrics (FG-ARI, mBO), though the gains saturate at high slot quality.
• With well-bound slots, auxiliary proprioception inputs and masking inductive bias become unnecessary.
• Under unseen distribution shifts, the OCWM with well-bound slots is more robust overall than the end-to-end trained scene-centric LeWM.
• DINO-WM, built on similar frozen pretrained features, remains competitive, pointing to pretrained features as a key contributor to robustness.

Key Concepts Mentioned

• world models
• offline trajectories
• object-centric representations
• slots
• scene-centric models
• visual model-predictive control
• distribution shift
• pretrained features

Claims Made by the Document

• Object-centric world models are more sample-efficient and generalize better.
• Planning success correlates positively with unsupervised slot-quality metrics.
• With well-bound slots, auxiliary proprioception inputs and masking inductive bias become unnecessary.
• Under unseen distribution shifts, the OCWM with well-bound slots is more robust overall than the scene-centric LeWM.
• Pretrained features contribute to robustness.

Evidence Used

• Results from a controlled study of OCWMs for visual model-predictive control.

Methods / Approach

• Controlled study of OCWMs
• Evaluation of planning success with unsupervised slot-quality metrics (FG-ARI, mBO)

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The study focuses only on visual model-predictive control and does not explore other types of tasks.

Open Questions

• What are the specific mechanisms behind the performance gains from high quality object-centric representations?
• How do OCWMs perform in other types of tasks beyond visual model-predictive control?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The study does not specify which exact slot-quality metrics were used for evaluation.

Raw Abstract / Excerpt

Learning world models from offline trajectories enables agents to accomplish different tasks through planning. Object-centric (OC) representations, which decompose a scene into a set of slots that bind to its objects, have been proposed as an inductive bias for world models that are more sample-efficient and generalize better. Yet prior object-centric world models (OCWMs) take the slot encoder as given and evaluate only in-distribution, leaving open whether the object-centric bias actually delivers for planning and what within the OCWM drives it. We conduct a controlled study of OCWMs for visual model-predictive control along two axes: object-centric representation quality and generalization under distribution shift relative to scene-centric models. We find that (i) planning success correlates positively with unsupervised slot-quality metrics (FG-ARI, mBO), though the gains saturate at high slot quality; (ii) with well-bound slots, the auxiliary proprioception inputs and masking inductive bias that prior methods relied on become unnecessary; and (iii) under unseen distribution shifts, the OCWM with well-bound slots is more robust overall than the end-to-end trained scene-centric LeWM, while DINO-WM, built on similar frozen pretrained features, remains competitive -- pointing to pretrained features as a key contributor to robustness.