Digest

Reset

1167 papers

2026-10-15 · arXiv:2610.19364

Dynamic Graph World Modelling: Learning Future Dynamics from Temporal Context

Bo Li, Yue Yang, Yujin Huang et al.

DyG-WM is a dynamic graph world modelling framework that jointly predicts future latent node and relation representations from temporal context. It uses a historical node-relation context encoder and a node-relation path-signature encoder to capture multi-level evidence and path-dependent evolution. A dual-level joint-embedding predictive objective learns future dynamics in representation space without reconstructing the raw graph. Experiments on 12 link-prediction and two node-classification datasets show effectiveness across both tasks.

Latent DynamicsCode availableAI summary
2026-10-15 · arXiv:2610.19375

One-Step Prediction Fidelity Is a Poor Proxy for Planning Performance in JEPA World Models

Joshua Cheung, Zhang He

The authors test whether lower one-step latent prediction error predicts better long-horizon goal-conditioned planning in JEPA world models. Using a frozen-decoder protocol and within-environment checkpoint ladders, they find fidelity improves monotonically while planning saturates after the first epoch, with near-zero correlation (PushT: r = −0.010 over 15 post-epoch-1 checkpoints). Across environments, induced fidelity changes show unstable planning effects.

JEPAAI summary
2026-10-15 · arXiv:2610.19411

W-JEPA: Wavelet Joint-Embedding Predictive Architecture for Wearable Activity Recognition

Yogesh Prabhu, Loki Natarajan, Jingjing Zou

W-JEPA is a self-supervised wearable activity recognition method that tokenizes inertial motion into fixed wavelet tokens indexed by scale, sensor axis, and time, then learns by masked latent prediction using a JEPA-style EMA target encoder. The authors report higher performance than prior methods on six wearable benchmarks, with the strongest subject-held-out results improving average macro-F1 by 8.7 points.

AI summary
2026-10-15 · arXiv:2610.19479

Active Generative Simulation: Meta-Learning What to Simulate for Sim-and-Real Policy Co-Training

Liqian Ma, Vaibhav Saxena, Utkarsh Aashu Mishra et al.

The authors propose Active Generative Simulation and MetaGenSim, which meta-learns a rater network to score synthetic data by utility for real-world validation performance, then constructs a targeted data-generation distribution over simulation parameters. Across real and sim-and-sim experiments with limited real demos, MetaGenSim improves sim-and-real co-training over uniform generation and retrieval, yields a 150% relative improvement in real-world full-task success over uniform generation, and matches expert manual alignment.

DrivingAI summary
2026-10-15 · arXiv:2610.19495

Robota-Embed: Unified VLA Embeddings for Robot In-Context Learning and Evolution

Yunze Liu, Jing Wang, Shenrui Wu et al.

Robota-Embed is a vision–language–action (VLA) embedding model that jointly encodes current-state queries and recorded future continuations in a predictive retrieval setting. The authors train it on 1,695,684 filtered current–future pairs and evaluate on Robota-Embed-Bench (4,000 queries, 8,000-clip gallery), reporting higher retrieval accuracy with native action than action-free baselines. The same retrieval space is used for robot in-context learning (Embed2ICL) and task-local context evolution (Embed2Evolve).

DrivingAI summary
2026-10-15 · arXiv:2610.19557

TEX-MEX: Similarity-Guided Temporal Contrastive Learning for Time Series

Aditya Radhakrishnan, Maxwell A. Xu, Wanting Mao et al.

TEX-MEX is a time-series SSL contrastive objective that unifies structural (morphology) and temporal (time proximity) similarity. For each anchor window, it grades batch members as structurally similar/dissimilar in encoder representation space, then pulls/pushes the anchor toward/away from temporally shifted versions of those graded similar/dissimilar windows. It also contrasts temporal shifts against unshifted frames to separate temporal content.

AI summary
2026-10-15 · arXiv:2610.19579

Back on Track: Deployment-time Policy Improvement via World Model Planning

Colton Crosby, Hao Wang, Mac Schwager et al.

The authors propose an offline-trained monitoring-and-intervention wrapper for black-box visuomotor policies. They learn a policy-conditioned liveness value to monitor progress during deployment and train a compact latent world model from the same rollouts plus play data. When the monitor detects loss of progress, a short-horizon planner searches the world model for corrective action chunks using the same value as objective, then returns control to the nominal policy.

DrivingAI summary
2026-10-15 · arXiv:2610.19642

Model What Matters: Decision-Guided Selective World Modeling for Language Agents

Maolin Ran, Minda Chen, Di Huang et al.

DGS-WM (Decision-Guided Selective World Modeling) is an RL + world-modeling co-training method for language agents. It selects decision-relevant parts of each environmental observation as world-modeling targets (Decision-Guided Observation Selection) and down-weights targets the model already predicts confidently (Confidence-Adaptive Weighting). The auxiliary world-modeling loss is jointly optimized with the RL loss, improving over full-observation world modeling (ECHO).

DrivingAI summary
2026-10-15 · arXiv:2610.19683

MemWM: Associative Memory-Enhanced Latent World Models for Long-Horizon Planning

Yuwei Sun, Siyu Zhu

MemWM is an associative memory-enhanced latent world model for long-horizon planning. It augments a JEPA-style latent predictor with a content-addressable memory bank that retrieves historical context via cross-attention, then conditions predictions using sequential AdaLN (memory first, then action). Two memory losses encourage structural diversity across slots and balanced slot usage across timesteps to prevent collapse.

Latent DynamicsAI summary
2026-10-15 · arXiv:2610.19708

SceneAE: Scene Generation and Reconstruction in Queryable 3D-Anchored Latent Space

Shuofeng Sun, Wenzhao Zheng, Yunpeng Zhang et al.

SceneAE represents a scene as a sparse 3D-anchored latent point cloud where each latent has explicit 3D position plus implicit content. The authors decode RGB-D from projected latents for novel camera views without regenerating a video, and generate scenes with a two-stage “geometry-content decoupled” approach: first generate target 3D anchor geometry from camera trajectories, then synthesize content conditioned on it.

2026-10-15 · arXiv:2610.19726

Generative World Simulation with DAO and JING

XGEN Team, Guohao Cai, Haoxuan Che et al.

The authors present XGEN, a Generative World Simulation framework that couples DAO, a computable persistent shared world engine, with JING, an egocentric interactive audiovisual generator. DAO maintains characters/objects/spaces and updates them from actions and world time; it then derives observer-specific conditions for JING. Together, they form a continuous loop where action consequences persist and shape future first-person experiences.

2026-10-15 · arXiv:2610.19746

One-shot Human Video-Guided Adaptation for Vision-Language-Action Models

Qi Chen, Yuwei Wu, Quan Chen et al.

CapPro is a one-shot adaptation framework for vision-language-action (VLA) models that uses a single human demonstration video to guide a pretrained VLA policy to previously unseen manipulation tasks. It learns phase-aware human-video representations (compact video tokens + phase-caption supervised query tokens) and injects them into the action-conditioning context via a gated adaptation module, without target-task robot demonstrations or fine-tuning.

DrivingAI summary
2026-10-15 · arXiv:2610.19795

StateCraft: Augmenting Video World Models with Abstract-State Reasoning

Lorenzo Venturoli, Nedko Savov, Xi Wang et al.

StateCraft decouples a persistent, updatable abstract state from the visual state in video world models. It pairs a language model that updates an explicit graph-based abstract state with a video diffusion model that generates the corresponding observations. Evaluated on generative software with a new benchmark and dataset pipeline, StateCraft improves visual memory metrics and supports state edits that guide generation beyond observed history, plus a proof-of-concept 3D case.

2026-10-15 · arXiv:2610.19825

What Matters for World Modeling in Vision-Language-Action Models?

Qisen Ma, Yixiang Chen, Yuan Xu et al.

The authors study how to integrate world modeling into vision-language-action models (VLAs) by organizing prior work into four paradigms (Implicit Future Alignment, Explicit Future Query, MoT Future Expert, External World Predictor) and testing them under a unified framework. They find 2D semantic representations (DINOv3/SigLIP) give robust cross-paradigm gains, decoupled MoT Future Expert is competitive esp. for OOD, world modeling helps on long-horizon/dynamic tasks, and it often improves over action-only policies.

DrivingAI summary
2026-10-15 · arXiv:2610.19831

CTIA-ODE: Continuous-Time Inter-Variable Attention for Long-Term Time Series Forecasting

Chengyi Pan, Huangqimei Zheng, Hao Xue

CTIA-ODE is a continuous-time multivariate forecasting framework that embeds Continuous-Time Inter-Variable Attention (CTIA) inside an ODE solver. The continuously evolving endogenous trajectory dynamically queries a static exogenous context along the integration path, avoiding explicit temporal alignment. The authors add Linear Residual Augmentation (LRA) and an Outlier-Robust Hybrid Loss (ORHL) for stable, robust continuous latent dynamics.

Latent DynamicsAI summary
2026-10-15 · arXiv:2610.19957

Complementary Future Token Pruning for World Action Models

Baekseung Kim, Junhoo Lee, Sohyeon Hong et al.

CF-Prune is a training-free method to prune future-frame tokens in pretrained world action models by selecting task-relevant future tokens from the initial denoising forward pass using instruction and action attention. It further disables pruning during “precision phases” detected via attention entropy contrast. Experiments on RoboLab, RoboCasa, and real-world manipulation report inference speedups with maintained or improved success rates.

DrivingAI summary
2026-10-15 · arXiv:2610.20134

Tactile-Guided Robot Learning: From Representation Learning to Foundation Models

Kiyanoush Nazari, Luchen Li, Haoshu Fang et al.

This paper is a learning-centric review of tactile-guided robot learning for dexterous manipulation. It organizes research by touch’s functional role across representation learning, tactile-conditioned policies, tactile foundation/world models, predictive/adaptive tactile control, and generalization (objects, tasks, sensors, embodiments, and sim-to-real). It argues touch is shifting from task-specific feedback to a core modality for general-purpose embodied intelligence.

RoboticsAI summary
2026-10-15 · arXiv:2610.20157

Atomic-WAM: Execution-Grounded Atomic-Motion Reasoning for World Action Models

Shijun Shi, Zhe Liu, Zhi Hou et al.

Atomic-WAM is an understand-then-execute world action model for robot control. A vision-language reasoner converts task instructions plus visual context into state-dependent atomic-motion descriptions, which condition a WAM to predict future video and an action chunk. The authors build temporally aligned, contact-enriched atomic-motion supervision from multi-view DROID demonstrations and post-train the reasoner with execution-grounded GRPO (EG-GRPO).

DrivingAI summary
2026-10-15 · arXiv:2610.20160

OneLA: Learning Many Worlds from Latent Actions

Ziyu Zhu, Zhaoxu Sun, Fan Jiang et al.

OneLA learns a shared continuous latent action space from unlabeled videos using a latent action model with a shared orthogonal motion dictionary and a common predictive objective (no cross-domain alignment losses during latent action pretraining). Lightweight adapters map native controls into this space; a diffusion world model predicts context-dependent transitions for controllable rollouts. The authors report improved action readout across domains and lower action cycle-transfer rotation error.

Latent DynamicsAI summary
2026-10-15 · arXiv:2610.20220

Glance Forcing: Making Bidirectional Diffusion Autoregressive at a Glance

Zhuobai Dong, Junchao Yi, Hu Jian Guo et al.

The authors propose Glance Forcing, a lightweight distillation method that makes bidirectional diffusion video models causal and few-step with much less PF-ODE trajectory supervision. When supervision is reduced to one PF-ODE trajectory, prior methods collapse due to conflicting high-noise vs low-noise optimization regimes. Glance Forcing decouples the denoising trajectory into two phase experts (Slow/Fast) using dual-LoRA with phase-specific objectives, while freezing the pretrained teacher.