Digest

Reset

1167 papers

2026-10-09 · arXiv:2610.12464

What 30,000 Hours of Ego-centric Video Does Not Teach

Jiahua Dong, Anurag Bagchi, Yash Jangir et al.

The authors scale Cosmos 3 ego-centric video world models from 300 to 30,000 hours and directly measure interaction fidelity (agent hands/body and manipulated objects) on an out-of-distribution benchmark. Data scaling improves both but unevenly: agent fidelity saturates early and conditioning can reach its ceiling with far less data, while object fidelity remains far lower and improves slowly. They introduce object-centric adaptive noise scheduling to improve object dynamics, with a remaining gap.

2026-10-09 · arXiv:2610.12468

DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-Training

Junyan Li, Ruizhi Li, Yu Liu et al.

DreamTrue is a multi-view, cross-embodiment robot world model for action-faithful and physically plausible video prediction. It improves action following via offline geometric calibration that aligns rendered robot actions with recorded video frames, and expands training coverage via counterfactual post-training that uses an embodied video reward model trained from human-annotated interaction defects. On AgiBot, it reports state-of-the-art action following and a large reduction in human-assessed interaction defect rate.

DrivingAI summary
2026-10-08 · arXiv:2610.08792

A Bayesian Mirror Architecture for Emergent Consciousness: Circular Hierarchies, Self-Manifolds, and Hybrid Event-Self Binding

Eduardo Righi Capanema de Almeida

The authors present the Bayesian Mirror Architecture (BMA), a generative framework with circular self-recursion S_t ← H_{t-1}, where a hybrid event–self latent H_t binds world abstractions to a self-latent and reinjects the coupling into the self-state. In belief space (posterior probability measures with Wasserstein-2 geometry), stability/coherence/integration are defined as diagnostics for causal learning (CLR).

AI summary
2026-10-08 · arXiv:2610.08800

A Review Of Robotic World Models For Dynamic Environments Based On Factor And Scene Graphs

Marco Giberna, Miguel Fernandez-Cortizas, Jose Luis Sanchez Lopez et al.

The authors review graph-based robotic world models for dynamic environments, focusing on factor graphs (uncertainty-aware geometric estimation), scene graphs (semantic relational abstraction), and hybrid combinations. They structure the survey around (I) representations, (II) pipelines to construct/update them, and (III) how they are exploited for downstream tasks, highlighting dynamics modeling, uncertainty propagation, observability under minimal sensing, lifelong maintenance, and evaluation gaps.

RoboticsAI summary
2026-10-08 · arXiv:2610.08825

Autonomous Driving Research Requires a Community-Driven Data Paradigm

Jinsu Yoo, Zanming Huang, Katie Z Luo et al.

The authors argue autonomous driving research is hindered less by modeling and more by a fragmented, benchmark-centric data paradigm. They report 600+ real-world autonomous driving datasets across nearly 50 countries, yet most are underused due to visibility, incompatible protocols, and incentives. They propose a collaborative community-driven data paradigm with structured metadata, shared interfaces, decomposed contributions, and long-term maintenance.

DrivingAI summary
2026-10-08 · arXiv:2610.08852

Autonomous Droplet Navigation via Model-Based Reinforcement Learning: Zero-Shot Transfer and Emergent Dynamics

Rajneesh Anand, Mayuresh V. Kothare

The authors introduce DropletRunner, a robotic platform for closed-loop autonomous liquid droplet navigation on an open, unconfined surface using model-based reinforcement learning. A DreamerV3-based policy is trained offline on experimental data (no simulator or analytical model), transfers zero-shot to unseen geometries, and autonomously discovers an oscillatory depinning strategy when the droplet sticks. The full training pipeline takes under 90 minutes.

DrivingAI summary
2026-10-08 · arXiv:2610.08941

SPW-Nav: A Streaming Panoramic World Model for Language-Guided Navigation

Yunheng Liu, Ziqi Cai, Siqi Yang et al.

SPW-Nav is a streaming panoramic world model for language-guided navigation. It interprets movement instructions using the previously generated panorama as camera motion, then streams continuous one-minute 2K 360° video in real time. Key ideas include spherical rotation decoupling (rotation by spherical resampling) and pose-aligned conditioning to keep translation inputs bounded, with streaming supported by multi-term memory and few-step generation.

2026-10-08 · arXiv:2610.08960

Directed Temporal Representations for Offline Visual Control

Chenyang Yuan, Haoyu Wang, Zhuo Sun et al.

DTRC learns a directed temporal quasimetric for offline visual goal-conditioned control on top of frozen LeWorldModel features. It calibrates local distance with short temporal offsets, bootstraps long-horizon reachability, and enforces action-conditioned consistency. The resulting distance change defines goal-relative temporal progress, used as a temporal critic for direct goal-conditioned policy learning (no test-time trajectory search).

Latent DynamicsAI summary
2026-10-08 · arXiv:2610.09048

Towards Financial World Modeling

Humzah Merchant, Alec Guthrie, Simon Mahns et al.

The authors study state representations for “world models” in financial markets. They introduce Market-1T, a publicly available dataset with nearly one trillion one-second aggregates across U.S. equities (2008–2025), and propose a 32-month evaluation protocol. Using 18 encoder-training strategies, they evaluate predictive utility (rank IC) and latent organization (retrieval/probing of structure), showing that similar predictive performance can correspond to different latent organization.

DrivingCode availableAI summary
2026-10-08 · arXiv:2610.09090

Tucker Bottleneck Attention for Multi-Dimensional Sequence Modeling

Ryan Solgi, Parsa Madinei, Zheng Zhang

Tucker bottleneck attention (TuBA) replaces quadratic self-attention over multidimensional token grids with attention over a compact low-rank Tucker core. TuBA reads a hidden tensor into Tucker cores using learned mode-wise factors, applies multi-head self-attention on the cores, then writes the update back to the original space. An autoregressive variant uses block-causal masking on core slots. The authors evaluate on video prediction and global weather forecasting.

2026-10-08 · arXiv:2610.09134

World Models Dream of Success: Diagnosing and Repairing Failure Insensitivity in Robot World Models

Jiuyi Xu, Xiao Hu, Meida Chen et al.

Robot world models can be failure-insensitive, producing success-like predictions on verified failures. The authors introduce CureWM: construct execution-verified counterfactual failures by perturbing successful demonstrations across a severity grid, then fine-tune released models on verified failing and surviving successful replays without changing architecture or training objective. On 484 held-out LIBERO failure counterfactuals, optimism drops to 30–43% (38% mean); on a physical Franka arm, failure diagnostic optimism drops from 90% to 33%.

RoboticsCode availableAI summary
2026-10-08 · arXiv:2610.09194

Patient, Place, Prior (P$^3$): What Counts as Personalization in Medical World Models?

Xingrui Gu, Hanxue Gu, Yuxiang Zhang et al.

The authors introduce Patient, Place, Prior (P^3), an audit framework to separate (1) whether a longitudinal medical forecast benefits from a patient's own history (Patient), (2) whether it benefits from patient-matched externally supplied spatial support (Place), and (3) whether it adds value beyond a population-average under matched support and context (Prior). They instantiate it with Cancer JEPA, a lesion-constrained one-step latent forecasting model.

JEPAAI summary
2026-10-08 · arXiv:2610.09254

RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer

Huang Huang, Wensi Ai, Ziyu Chen et al.

RoboRender converts simulation trajectories into photorealistic, robot-oriented multi-view RGB videos for zero-shot visual sim-to-real policy transfer. The model is conditioned on simulated depth, language instructions, and robot RGB masks, preserving geometry, robot motion, and simulator action labels while synthesizing realistic textures, backgrounds, and distractors. Policies trained on RoboRender videos achieve 71% average real-world success, outperforming raw simulation and visual domain randomization.

DrivingAI summary
2026-10-08 · arXiv:2610.09285

LeCuration: A Tiny World Model as a Data Curation Multi-Tool

Mayank Sengupta, Nirmit Desai, Eric Song et al.

The authors train LeCuration, a small predictive world model for data curation on CS:GO gameplay. Using LeWorldModel-style latent encoder/predictor with a DiT decoder and TAESD latents, they show that CLS embeddings support content-based clustering and anomaly detection via embedding divergence during autoregressive rollouts. The paper is a qualitative proof-of-concept and does not report quantitative curation metrics or downstream results.

Latent DynamicsAI summary
2026-10-08 · arXiv:2610.09305

Kuration SDK: Addressing the Virtual2Real Gap via Data Curation

Nirmit Desai, Eric Song, Mayank Sengupta et al.

The authors report a “Virtual2Real gap” for action-conditioned diffusion world models: qualitative playability on CounterStrike gameplay can diverge sharply from metrics like FVD, LPIPS, and JEDi. They attribute the gap to data curation issues and introduce Kuration SDK, an open-source toolkit for curating gameplay/sequential data and measuring diagnostic properties.

DrivingAI summary
2026-10-08 · arXiv:2610.09309

Predicted Futures Are Not Enough: Learning Executable Goals for Robot Manipulation

Tzu-Yu Chuang, Ching-Hsiang Chang, Yi-Hsiu Lee et al.

The authors argue that generative world-model futures don’t directly optimize the compact task variables needed for control. They introduce Entity-Level Goal Readout, which converts a 3D trace world-model representation into a compact SE(3) terminal goal (object-relative pose with depth-grounded translation). A Pose-Native Executor uses this fixed goal with online pose feedback. On five ManiSkill3 manipulation tasks, the pipeline reaches 79.69% mean success.

RoboticsAI summary
2026-10-08 · arXiv:2610.09335

SearchWorld: Spatial Value-Grounded Imagination for UAV Object Search via World Models

Yatai Ji, Zhengqiu Zhu, Yong Zhao et al.

SearchWorld is a recurrent state-space world model for UAV object search that links explicit BEV exploration/obstacle memory with a decoded task-aware spatial value layer. A cognition–action network refines expert-imitation movement policies using value-guided imagination rollouts without training a separate scalar critic. On UAV-ON, the authors report 23.80% success rate (SR), 35.52% oracle success rate (OSR), and 19.92% SR on unseen environments.

DrivingAI summary
2026-10-08 · arXiv:2610.09427

Event-Aligned Visual Action Reasoning for World Action Models

Xiaomeng Yang, Yushu Wu, Yi Gao et al.

The authors introduce an event-aligned visual action reasoning framework for World-Action Models (WAMs). Instead of using fixed temporal horizons, they build visual-action chunks from interaction events and their evidence/support phases. They also add an execution validity head to skip padded/repeated action slots during chunked inference. On DOMINO, they report a 10.26 percentage-point DOMINO success-rate gain over LingBot-VA.

Latent DynamicsAI summary
2026-10-08 · arXiv:2610.09438

Controllable Crowd Generation through World-Model Planning

JunGyu Lee, Jisu Shin, Seunghyun Shin et al.

Ctrl-CWM is a multi-agent controllable crowd world model that uses imagined futures for both crowd generation and run-time control. It learns human motion dynamics via trajectory prediction on real pedestrian videos (then freezes the encoder), trains an actor to propose one-step displacements and a critic to rank imagined trajectories, and uses CEM planning with a user cost to steer avoidance/attraction without retraining.

DrivingAI summary
2026-10-08 · arXiv:2610.09457

DSReg: Provably Recovering Individual World Latents without Reconstruction

Yujia Zheng, David Klindt, Randall Balestriero et al.

DSReg (Dependency-Sparsity Regularization) provably recovers individual world latent variables from a linearly identified JEPA representation without reconstruction, no decoder, and no labels. It resolves the rotation ambiguity left by linear identifiability by optimizing a dependency-support sparsity criterion under Structural Diversity, yielding signed-permutation identifiability. The authors report improvements in individual-latent recovery and downstream sparse control/editing while preserving dense recovery.

JEPAAI summary