Digest

Reset

1167 papers

2026-11-03 · arXiv:2611.02489

Cross-Domain Joint-Embedding Pretraining For Heterogeneous Industrial Time Series

Jonathan Adam Rico, Nagarajan Raghavan, Senthilnath Jayavelu

The authors propose JEPA-HITS, a channel-independent joint-embedding predictive pretraining method for heterogeneous industrial time series. They learn one shared JEPA encoder across battery, CMP, CNC, and bearing data while preserving dataset-specific temporal scales, using leave-one-domain-out (LODO) pretraining to test transfer to unseen domains. Compared to the same architecture with random initialization, LODO pretraining reduces RMSE on NASA/CMP/CNC and improves bearing fault macro-F1.

JEPAAI summary
2026-11-03 · arXiv:2611.02526

INTERACT: Learning Interaction-Aware World Dynamics from Driving Videos for Autonomous Driving

Zhan Chen, Songyan Zhang, Wenhui Huang et al.

INTERACT learns an interaction-aware, action-conditioned world model from unlabeled front-view driving videos. It uses recurrent object-centric slots (Stage I), infers latent ego actions from visual transitions and slot context while predicting future slots (Stage II), then distills the learned dynamics into a trajectory planner with multimodal hypotheses (Stage III). Experiments on NAVSIM and nuScenes show state-of-the-art NAVSIM performance, with improvements on safety-critical collision metrics.

DrivingAI summary
2026-11-03 · arXiv:2611.02600

W2A: Distilling World Priors into Action Models for Real-Time Robotic Control

Xiaopeng Zhang, Yongjin Mu, Yueyang Weng et al.

W2A is a two-stage distillation method that transfers world priors from a language-conditioned video diffusion transformer into an action model for real-time robotic control. During training, intermediate denoising video latents are compressed with PCA and distilled into an action DiT using a Q-Former. At test time, the video model is discarded, yielding video-free deployment with higher inference efficiency.

DrivingAI summary
2026-11-03 · arXiv:2611.02690

Human-Oriented World Model

Saihui Hou, Zengyun Mou, Chunshui Cao et al.

HOWM (Human-Oriented World Model) is a conceptual framework for repeated human support. It links current observations and individual history with environmental/social context to maintain continuing state estimation, make personalized dynamics predictions under action conditions, and choose support via an external policy tied to the person’s goals, preferences, and constraints. Interaction outcomes are used to revise state estimates and personal dynamics patterns over time.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.02699

Right Ranking, Wrong Scale: Reproducing the Robotic World Model's Dynamics Model and Testing the Uncertainty of Its Follow-up

Joyjeet Singh

The authors reimplement the Robotic World Model (RWM) dynamics model and test RWM-U’s ensemble-based uncertainty used to penalize reward in offline policy optimization. Their rebuild matches the released model’s gradients. On held-out data, RWM’s rollout training beats teacher forcing at a preregistered long horizon for all ten episodes, but loses at one step. RWM-U ranks rollout errors correctly (Spearman ρ=0.90 at 100 steps) yet its spread is badly miscalibrated and too small; they also find defects in the released evaluation pipeline.

RoboticsAI summary
2026-11-03 · arXiv:2611.02782

History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models

Hangyi Kuang, Chengzhi Wang, Yefei He et al.

History Forcing is an on-policy distillation framework for few-step autoregressive video diffusion. It computes a history-dependent residual by comparing an AR teacher’s predictions with vs. without informative historical context, then adds a weighted corrected term to the distillation score/velocity target. This strengthens long-range historical constraints for long rollouts and out-of-sight world modeling.

Video World ModelsCode availableAI summary
2026-11-03 · arXiv:2611.02807

Creating Hybrid Interactive Narrative Systems with LLM: From Formal Description to Evaluation

Jules Clerc (Heudiasyc), Domitile Lourdeaux (Heudiasyc), Mohamed Sallak (Heudiasyc) et al.

The authors propose an LLM-based narrative platform to compare hybrid experience managers (EMs). Managers are interchangeable modules over a shared symbolic world model (facts, operations, goals), so each session can be represented as a common state-transition plan and evaluated via indicators of player agency, EM controllability, author intent, and robustness. They instantiate three managers (open generation, MCTS repair, multi-agent NPCs) and report trade-offs on an interactive Little Red Riding Hood scenario.

DrivingAI summary
2026-11-03 · arXiv:2611.02815

Predictive Safety Guardrails Do Not Always Make LLM Agents Safer

Weixiang Zhao, Zhihao Liu, Guancheng Li et al.

The authors diagnose predictive safety guardrails as a prediction-to-action pipeline. Across seven policy models and two safety environments, stronger world-model prediction involvement is not monotonic for safety, can reduce task completion, and increases inference cost. Predictions are also insufficiently reliable, especially at longer horizons, and safety improvements can persist even with corrupted predictions—suggesting a prediction-to-action gap.

DrivingAI summary
2026-11-03 · arXiv:2611.02853

Scaling Action-Free Video Latent Prediction for Transferable Representations in End-to-End Autonomous Driving

Xingtai Gui, Xubin Zhou, Yucheng Zhou et al.

ScaleDrive is a scalable self-supervised framework for end-to-end autonomous driving that pretrains on large-scale internet open-world driving videos using action-free video latent prediction. The pretrained video encoder is kept frozen during downstream planning to test transferability. Scaling up action-free pretraining improves downstream closed-loop planning on NAVSIM/HUGSIM/Bench2Drive, and probes indicate increasing temporal-motion and spatial-geometric information in the frozen representation.

DrivingAI summary
2026-11-03 · arXiv:2611.02876

Identifiable Signed Causal Learning via Latent Stochastic Differential Equations for Inferring Brain Dynamic Effective Connectivity

Yiding Wang, Chen Qiao

ISCL is a continuous-time latent stochastic differential equation framework for dynamic effective connectivity. It derives a signed, observation-space causal influence operator from learned latent dynamics, with a rationality theorem (directed interventional response per brain state) and an identifiability result under invariance assumptions. On Lorenz96 and NetSim it recovers causal structure and reports lowest mean SHD; on HCP Social Cognition it yields a signed, seed-stable architecture and task reconfiguration.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.02922

An intelligent satellite must do nothing well

Fabrizio Pilone, Giordana Bucchioni

The authors argue that algorithm choice is not the main design axis for autonomous space inspection; fuel efficiency depends on representation and planning choices that let an agent exploit natural drift (no rest state) and plan long enough. A control barrier function (CBF) filter eliminates keep-out-zone (KOZ) violations in evaluation episodes where learned policies alone can fail.

DrivingAI summary
2026-11-03 · arXiv:2611.03005

BLPO: Beyond-Local Policy Optimization for Long-Horizon LLM Agents

Xuanren Chen, Chongyang Tao, Wenpeng Hu et al.

BLPO (Beyond-Local Policy Optimization) estimates step-level credit for long-horizon LLM agents by combining trajectory-relative advantages, local advantages over normalized actions, and task-progress advantages pooled across tasks and training updates. It summarizes trajectory prefixes into task-progress states and abstracts actions by function, using a global memory with context-relative residuals plus support/uncertainty reliability weighting. The method is critic-free and uses no additional rollouts, showing improved success on ALFWorld and WebShop.

DrivingAI summary
2026-11-03 · arXiv:2611.03294

ForceLDM: A Force-aware Latent Dynamics Model for Contact-Rich Manipulation

Peiyuan Tang, Yao Li, Xiaodong Zhang et al.

ForceLDM is a force-aware latent dynamics model for contact-rich manipulation. It uses a privileged teacher with future force/torque and optical-flow signals to extract control-relevant future dynamics, then distills these latent representations into a student trained only on current observations. At inference, only the student is deployed. The method also uses curriculum-based progressive noise injection for robustness and generalization.

RoboticsAI summary
2026-11-03 · arXiv:2611.03428

Toward Uncertainty-Aware, Physics-Grounded Human World Models

Yufei Zhang, Zijun Cui

The authors connect two published components to form an uncertainty-aware, physics-grounded human world model: KNOWN estimates SMPL body state along with aleatoric and epistemic uncertainty in real time (11.6 ms/frame), and PhysMoP predicts future body configuration by integrating Euler–Lagrange dynamics with learned terms. They define an interface to couple per-frame perception beliefs as the input history, propose three uncertainty-injection mechanisms, and specify an evaluation protocol on in-the-wild video.

DrivingAI summary
2026-11-03 · arXiv:2611.03585

Seeing Through the Illusion: Test-Time Scaling for World Action Models

Cheikh Ahmed, Mahdi Mostajabdaveh, Milad Yazdani et al.

WISE is a test-time pre-execution selector for frozen world action models. For each decision it samples multiple action–future rollouts, scores each candidate on action validity/temporal regularity, inverse-dynamics consistency between the proposed action and imagined motion, and instruction-conditioned task progress, then executes the highest-scoring action chunk. With four candidates per decision, it improves closed-loop success on RoboLab (Cosmos 3 Edge) and LIBERO (DreamZero).

DrivingAI summary
2026-11-03 · arXiv:2611.03596

Spacetime Emergence in the Memory of a Learning Agent

Philip A. LeMaitre, T. Rick Perche

The authors train a model-based reinforcement-learning agent on a quantum information task involving two Unruh–DeWitt qubits coupled to a massless scalar field in Minkowski spacetime. The only reward is the quantum mutual information between the detectors, with no metric/causal structure provided. The agent’s internal planner develops null/lightcone structure, enabling reconstruction of the Minkowski causal structure with high accuracy from its memory.

DrivingAI summary
2026-11-03 · arXiv:2611.03622

D&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward Models

Marius Memmel, Shawn Yin, Urvi Rutia et al.

The authors propose D&A (Describe & Assess), a purely prompt-based zero-shot reward model for robotic video task-progress estimation. It decouples a VLM-based trajectory description (without the task) from an assessment stage that uses an explicit, text-based success criterion/rubric to score per-frame progress in [0,1]. This mitigates confirmation bias and yields dense partial progress rewards and rich failure feedback for downstream RL.

DrivingAI summary
2026-11-03 · arXiv:2611.03626

Isekai3D SDG: A Platform for Diverse Synthetic Gameplay Visual Data

Rafal Karp, Gene Liu, Shih-Hsin Li et al.

Isekai3D SDG is a two-stage synthetic data generation platform that separates world simulation/recording from capture. It records world state, character actions, and camera tracks, then replays the same recorded gameplay to produce multiple aligned visual outputs (e.g., RGB, depth, normals) with frame-aligned metadata at chosen resolutions and frame rates. It provides reusable assets and Unreal Engine plugins and a packaged application for data generation without installing the Unreal Engine editor.

2026-11-02 · arXiv:2610.40662

A small HER2 reader for routine H&E slides: real-world deployment

Thang Tran (CloudKites AI Lab, New South Wales, Australia) et al.

The authors build an end-to-end HER2 reader for routine H&E slides: a 22M-parameter ViT-S/16 tile encoder trained on public TCGA slides via self-supervision, distillation from H-optimus-0, and two JEPA refinement rounds, with five cross-validated attention-based slide classifiers. They report strong external AUCs on Yale-HER2 and other hospitals and provide a deployed CPU research-use pre-read with public weights and per-slide predictions.

DrivingAI summary
2026-11-02 · arXiv:2610.40748

WorldBridge: Bridging Video and Action for Task Generalization in World Action Models

Jun Cen, Yubo Guo, Hao Li et al.

WorldBridge is a World Action Model for task generalization that bridges video prediction and action generation via separate video and action experts connected through structured attention. The authors study how adding out-of-distribution task videos and sparse action labels improves out-of-distribution success on RoboTwin 2.0, and report state-of-the-art results across multiple manipulation benchmarks.

DrivingAI summary