Digest

Reset

1167 papers

2026-10-07 · arXiv:2610.08464

Federated Bayesian Surveillance of Mechanical Thrombectomy Adverse Events: A Population Risk Layer for Surgical Digital Twins

Damini Rijhwani

The authors propose a population risk layer for surgical digital twins that models adverse-event rates with per-class Gamma–Poisson posteriors learned via a federated Bayesian protocol under Rényi differential privacy. They benchmark on the complete FDA MAUDE cohort for thrombus-retrieval catheters (NRY) across K=8 manufacturer sites, reporting held-out Poisson log-likelihood improvements for their conjugate federated method at matched privacy budgets.

DrivingAI summary
2026-10-07 · arXiv:2610.08469

A Belief-State World Model for Catheter Navigation under Sparse Fluoroscopy: A Planar Proof of Concept

Damini Rijhwani

The authors present a POMDP “belief-state world model” for catheter navigation under sparse fluoroscopy, using a Cosserat rod simulator for transitions, sparse noisy projections for observations, and a particle filter to maintain belief and calibrated uncertainty. In a synthetic planar vessel phantom, the belief model tracks tip position with RMSE 1.30 mm at one-fifteenth frame rate and maintains near-nominal 90% credible coverage (0.92 empirical coverage).

DrivingAI summary
2026-10-07 · arXiv:2610.08475

Learning PDE solution operators with variable initial conditions via Latent Dynamics Networks

Stefano Maria Pizzamiglio, Stefano Pagani, Francesco Regazzoni

The authors extend LDNet to handle PDEs with variable initial conditions while staying encoder-free and resolution-independent. They infer the initial latent state from early-time observations using either an auto-decoding latent-code optimization (AD-IC-LDNet) or a CAVIA-style meta-learning strategy (Meta-IC-LDNet). The coordinate-based decoder supports training from spatially subsampled data and recovering high-resolution fields at inference across multiple PDE phenomena.

Latent DynamicsAI summary
2026-10-07 · arXiv:2610.08526

WareFly-VLA: A Vision-Language-Action Framework for UAV Navigation and Human Tracking in Smart Warehouses

Thinh D. Le, Son T. Nguyen, Duong Q. Nguyen et al.

WareFly-VLA is a photorealistic UAV Vision–Language–Action (VLA) framework and dataset for language-guided human search, localization, and tracking in warehouse environments. It includes 507 human-teleoperated flight episodes and 8,504 non-terminal RGB transitions with appearance-based language and synchronized 4-DoF body-frame control commands, plus a leakage-free benchmark over four open-source VLA architectures.

DrivingAI summary
2026-10-07 · arXiv:2610.08640

RIWANav: Recursive World-Action Models with Self-Improvement for Urban Navigation

Jing Xie, Shouwei Ruan, Yubin Wang et al.

RIWANav is a post-training framework that recursively improves a navigation policy and a visual action-conditioned world model for long-horizon urban navigation. Each recursion alternates (1) world-model-based imagined outcome comparison for GRPO policy optimization and (2) policy-guided construction of a grounded self-curriculum of expert-consistent action–video pairs (by behavioral novelty and prediction error) to refine the world model. Experiments on UrbanNav show consistent gains over baselines; real-world trials demonstrate feasibility.

DrivingAI summary
2026-10-07 · arXiv:2610.08760

WorldSonus: Bringing Sound to Worlds

Pengjun Fang, Jingyi Fa, Kam Man Wu et al.

WorldSonus is an interactive video-to-audio framework for real-time spatial stereo in world models. It uses streaming causal autoregressive diffusion to synthesize 100ms audio chunks at RTF 0.41, supports mid-stream text prompt updates via chunk-indexed prompt scheduling, and trains spatially aligned stereo using clear-stereo and panoramic FOA supervision. Experiments show it matches or outperforms bidirectional baselines on acoustic quality and stereo balance.

2026-10-07 · arXiv:2610.08773

AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model

Sarim Hashmi, Mukul Ranjan, Kshitij Mishra et al.

AdvSim2Real trains a web agent against adaptive prompt injection inside a frozen web world model (WebWorld-14B) by co-evolving (1) a task curriculum, (2) an injection adversary, and (3) the executor. The curriculum is rewarded for tasks the executor solves about half the time (judge-graded), and the adversary for success flips only. On 150 web tasks, it increases completion under an unseen frontier adversary (Kimi-K3) by 33.6% relative to the base agent.

DrivingAI summary
2026-10-07 · arXiv:2610.08777

CtrlCache: Accelerating Interactive Video World Models with Control-Aware Caching

Shangye Song, Dong Gong, Hong Jia et al.

CtrlCache is a training-free, control-aware caching method for interactive video world models with chunk-wise autoregressive few-step denoising. Using incoming controls, it assigns each chunk a state (initial/transition/turning/steady) and selectively reuses a cached transformer residual at one interior denoising step. For steady chunks, it also applies a frequency-mixed history prior guidance blended into the first clean latent estimate.

2026-10-07 · arXiv:2610.08780

DepthWorld: 3D World Model for Robot Manipulation

Jai Bardhan, Josef Sivic, Vladimir Petrik

DepthWorld targets geometric consistency in robot manipulation world models. The authors build DROID-3D by combining learned stereo depth with a joint factor graph over episodes from the same robot to recover shared kinematic parameters and per-scene extrinsics. Using this calibrated 3D supervision, they train a Stable Video Diffusion model (Stable Video Diffusion backbone with unmodified VAE) that jointly predicts multi-view RGB and metric depth via spatial latent tiling and robot-frame point-map supervision.

RoboticsAI summary
2026-10-07 · arXiv:2610.08791

World Models' Last Exam in Physics

Mingju Gao, Qingle Liu, Yuzhao Peng et al.

The authors introduce World Models’ Last Exam in Physics, a measurement-based benchmark for physical consistency in video world models. It uses 40 controlled tasks across nine physical domains, with an evaluator that gates physics scoring by temporal consistency/observability and otherwise applies task-specific quantitative measurements from each video. Across eight video generation models on 1,280 videos, the best overall score is 57.76/100 and physical inconsistencies persist.

2026-10-06 · arXiv:2610.03797

WAMJET: A Harness for World Action Model Acceleration

Le Chen, Lixin Liu, Jan Schneider et al.

WAMJET is an agentic harness for accelerating World Action Model (WAM) inference. It guides coding agents through a bottleneck-driven workflow: profile inference latency, modify targeted code for lossless or approximate accelerations, and validate numerical/state correctness plus paired task performance. Experiments across six WAMs and two GPU architectures show up to 9.95× lossless speedup over upstream implementations, with comparable success rates under approximate acceleration.

DrivingCode availableAI summary
2026-10-06 · arXiv:2610.03959

LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization

Ziye Deng, Lufang Chen, Shuyu Feng et al.

LatentQuant targets NVFP4 W4A4 quantization for world action model VAEs whose encoder latents condition a frozen action policy. The authors report joint QAT can match reconstruction yet violate the “policy-facing latent contract,” collapsing control (e.g., LIBERO success 95.5%→10.5%). LatentQuant uses a two-stage NVFP4 QAT: align/fix the quantized encoder first, then freeze it while adapting only the decoder, preserving control and reconstruction while enabling NVFP4 speedups.

DrivingAI summary
2026-10-06 · arXiv:2610.04009

SUAVE: Unified Video-Action Models via Masked Diffusion

Rhythm Syed, Jean Mercat, Sedrick Keh et al.

SUAVE is a single masked-diffusion transformer that represents language, video, and robot actions as discrete tokens in one sequence. By choosing which token blocks to mask at inference, the same network can act as a world model (mask goal frames), a robot policy (mask actions), or a video-action model (mask both). The authors report simulation and real-robot results on manipulation benchmarks and closed-loop control.

DrivingAI summary
2026-10-06 · arXiv:2610.04034

LD-EnFF: Latent-Dynamics Ensemble Flow Filtering for Data Assimilation with Sparse Observations

Ziyu Tian, Kaichen Shen, Wenbo Hao et al.

LD-EnFF is a sequential Bayesian data-assimilation framework that performs forecast propagation and filtering updates in a compact augmented latent space. It uses LDNet for latent, parameter-conditioned dynamics and reconstruction, and combines a VAE-based latent observation encoder with a learned forward y⋆ operator to build a state-dependent latent likelihood. An ensemble flow-matching update (EnFF) generates posterior samples, jointly updating latent states and uncertain parameters, avoiding full-state simulation and repeated full-field likelihood evaluation.

Latent DynamicsAI summary
2026-10-06 · arXiv:2610.04152

Kepler4D: Controllable Future Video Generation via 4D Scene State Evolution

Feiran Wang, Bin Duan, Junyi Wu et al.

Kepler4D is a controllable future video generation framework that builds a shared 4D scene state from a monocular video (static background geometry, foreground object motion histories, coarse spatial supports, and semantic context). Chain-of-Motion selects structured speed/heading decisions (optionally a lower-bound height constraint) and deterministic rollout produces editable future 3D object-center trajectories, which are rendered as geometric controls for a pretrained video generator.

2026-10-06 · arXiv:2610.04168

Agentic Cognitive Depth: Operational Criteria for Evaluating LLM Agents

Nijesh Upreti, Chris Sypherd, Vaishak Belle

The authors define “agentic cognitive depth” as a trajectory-level profile across five operational criteria: context sensitivity (C), temporal continuity (T), multimodal coordination (M), adaptive interaction (A), and metacognitive monitoring (Mc). They provide operational proxies and a perturbation procedure for each criterion, and connect the profile to an agent’s world-model substrate to support per-criterion diagnostics beyond end-to-end task success.

AI summary
2026-10-06 · arXiv:2610.04301

EnvDreamer: Large-Scale Multimodal-to-Environment Generation for Embodied AI

Kabir Swain, Sijie Han, Antonio Torralba

EnvDreamer is a language-first framework that generates validator-passed, physics-enabled Unreal Engine 5 environments for embodied AI, world models, and robotics. It plans assets/layout/task logic, compiles to Unreal Python, applies edits via Remote Control, and validates playability, physics consistency, and goal completeness. The authors release EnvDreamer-20k (20,000 environments) and show competitive zero-shot/few-shot results after pretraining on multiple embodied benchmarks.

2026-10-06 · arXiv:2610.04336

A differentiable Lagrangian-coupled 3D Gaussian Splatting-SPH model for forward simulation and inverse analysis in solid mechanics

Tian Xu, Soroush Atashi, Tianju Xue

The authors propose a differentiable Lagrangian-coupled 3D Gaussian Splatting–SPH model for forward simulation and inverse analysis of deformable solids. They reconstruct a 3DGS model from multi-view calibrated images, generate an independent SPH support from the Gaussian geometry, transfer SPH deformation to Gaussian centers/covariances via a Lagrangian mapping, render differentiable deformation frames, and use the same pipeline for gradient-based parameter identification from images (including noisy cases).

DrivingAI summary
2026-10-06 · arXiv:2610.04432

Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos

Jinzhou Tang, Zijun Zhang, Jing Yang et al.

Video2World studies autonomous video-to-simulation as an end-to-end software engineering task: a coding agent observes an embodied RGB video, constructs a simulated environment and robot behavior, then iteratively refines using execution feedback. The benchmark contains 222 reconstruction instances from 189 robot/human demonstration videos, evaluated with geometric fidelity, dynamic fidelity, and functional correctness.

Video World ModelsCode availableAI summary
2026-10-06 · arXiv:2610.04475

VCLMU: Mechanism-Centric Virtual Cell World Modeling for Perturbation Response

Yuwei Miao, Azim Dehghani Amirabad, Scott Oloff et al.

VCLMU is a mechanism-centric virtual cell world model that predicts perturbation response via action-conditioned stochastic latent transitions. Cellular state is represented as a sparse set of Latent Mechanism Units (LMUs) with reusable, evidence-grounded identities and observation-specific states; perturbations act as actions that induce mechanism-specific transitions before decoding transcriptional responses. Trained with two-stage pretraining (pseudo-bulk then gene-aligned single-cell), it improves perturbation-specific recovery across six perturbation-disjoint benchmarks.

Latent DynamicsAI summary