Digest

Reset

1167 papers

2026-10-26 · arXiv:2610.31722

Autoregressive Multimodal Anomaly Detection with Large Language Models for Microservice Systems

Yixiang Tang, Fagui Liu, Guoxiang Zhong et al.

LLMMAD is an unsupervised autoregressive multimodal anomaly detection framework for microservice systems. It interleaves metrics/logs/traces into a unified sequence for LoRA-adapted, frozen LLM inference, then applies modality-wise reconstruction heads. It uses a prediction-driven autoregressive objective to predict future states (not copy inputs) to avoid the “identity shortcut,” and aggregates modality decisions with OR-logic.

DrivingAI summary
2026-10-26 · arXiv:2610.31762

See What I Need, Dream What I Want: A Navigation Foundation Model on a Self-Constructed Representation Space

Jiawei Chen, Yinxi Wang, Zihan Yang et al.

RepresentNav is a navigation foundation model that constructs a self-constructed representation space from frozen vision foundation models using Representation Residuals, and learns a latent chain-of-thought “dream” in that same space via JEPA training (next-representation prediction plus VISReg). The authors report state-of-the-art results on VLNVerse, VLN-CE R2R, EVT-Bench, and two point-goal benchmarks, with gains transferring to a physical robot.

DrivingCode availableAI summary
2026-10-26 · arXiv:2610.31834

WorldMirror: Agentic Real-to-Sim Generation of Interactive Worlds from In-the-Wild Robot Demonstrations

Tianling Xu, Weijie Wan, Honghao Su et al.

WorldMirror is an agentic real-to-sim framework that turns in-the-wild robot demonstrations (RGB video + trajectories) into simulation-ready interactive digital twins. It formulates construction as iterative refinement of an executable world program, using a camera-pose-anchored scene representation and verification-driven repair, enabling direct trajectory replay without motion replanning or trajectory retargeting.

DrivingAI summary
2026-10-26 · arXiv:2610.31983

Interpreting Physics in Video Encoders using Multi-Object Interaction Videos

Ayushi Mehrotra, Sai Vidyaranya Nuthalapati

The authors probe whether two frozen self-supervised video encoders (V-JEPA2 and VideoMAE) encode relational physics in two-object interactions (momentum conservation, elasticity, kinetic-energy changes) and in cube–floor friction. They find elastic vs inelastic collisions are reliably separated in both, but momentum-violation and kinetic-energy transfer across environments are mixed, and friction-coefficient readouts do not survive geometry changes.

DrivingAI summary
2026-10-26 · arXiv:2610.32067

Controlled Evaluation of EEG Foundation Models: Random Initialisation, Band-Power and Validation-Selected Probes

Mingyu Jeon, Suwan Cho, Jae Young Suh

The authors evaluate six EEG foundation models (five public + one in-house JEPA-style) under a single controlled protocol with three controls: random-initialized same encoder, a five-band log-power baseline, and validation-selected probe layer (not test-selected). They report that these controls shrink or remove many downstream gains, with layer selection on test inflating peaks and fine-tuning gains mostly within seed noise. They release per-cell results for re-analysis.

Latent DynamicsAI summary
2026-10-26 · arXiv:2610.32069

DriveMind: Mind-Evolving Belief Tracking for Closed-Loop Autonomous Driving

Yan Shi, Kehua Chen, Jiarui Qi et al.

DriveMind is an MLLM-based closed-loop autonomous driving framework that maintains multiple natural-language scene hypotheses with time-evolving belief weights. Inspired by particle filtering, it propagates hypotheses through a VLM, scores them against the current observation, and resamples to preserve diversity. On Bench2Drive, belief tracking improves Driving Score (DS) and Success Rate (SR) over the same single-hypothesis pipeline.

DrivingCode availableAI summary
2026-10-26 · arXiv:2610.32177

$\Psi$-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model

Khai Loong Aw, Klemen Kotar, Baihan Zhang et al.

Ψ-0.6 is a unified vision-language world model that adaptively composes perception, language, geometry, and counterfactual simulation in a single shared token interface using streams of thought. Using reinforcement learning on task feedback, it learns to select and use intermediate geometric and imagined representations, improving zero-shot multi-view spatial reasoning and counterfactual visual world modeling results.

2026-10-26 · arXiv:2610.32208

Do VLAs Know They Will Fail? Interpretable Success and Failure Patterns in Physical AI Systems

Christian A. Rogers, Qian Lin, Daniel S. Brown

The authors show that sparse autoencoder (SAE) representations of OpenVLA activations contain outcome-predictive information and can predict task failure before an estimated point of no return. They identify individual SAE features whose strong activations correlate with success or failure and localize critical temporal intervals where features rise. A VLM-based pipeline generates behavioral descriptions of these patterns, validated via detection scoring on held-out clips.

RoboticsAI summary
2026-10-26 · arXiv:2610.32222

ACWM-RM: A Reference-Free Reward Model from Complementary Mismatch Preferences

Shibo Hong, Siyu Ren, FengJiao Chen et al.

The authors introduce AVM-Eval (67.2k pairs) to test whether inverse dynamics model (IDM) action-following rewards transfer from real to generated videos. They find real-video IDM accuracy does not reliably evaluate generated videos. They propose ACWM-RM, a reference-free reward model trained on AVM-Pref (506.3k pairs) to jointly assess visual quality (VQ) and action-following fidelity (AF) without ground-truth videos during reward computation.

Latent DynamicsAI summary
2026-10-26 · arXiv:2610.32381

UniWorld: A Unified World Model for Real-Time Visual World Simulation

Xin Zhang, Yabo Chen, Shuaiwei Wang et al.

UniWorld is a unified continuous visual world simulation framework that renders evolving scene geometry (image-aligned depth trajectories) together with temporally persistent appearance (reference RGB) using one shared video generator. It uses noise-aligned overlapping continuation and rolling depth latent state to preserve committed latent history exactly while bounding active depth state by window size, enabling real-time steady-state throughput.

2026-10-26 · arXiv:2610.32414

When Does Dynamics Randomization Test Context Inference?

Oleg Smirnov, Sofiane Ennadir, John Pertoft et al.

The authors argue that dynamics randomization may not actually test context inference if the current state and action already contain most predictive evidence or if kernel differences vanish on visited support. They define “predictive headroom” (one-step log-loss difference) and compare kernels under fixed vs visited state–action distributions to separate context coverage from context reweighting effects, across discrete and continuous synthetic environments.

Latent DynamicsAI summary
2026-10-26 · arXiv:2610.32554

HGWM: Hierarchical Graph-Guided World Model for Zero-Shot Object Navigation via Graph Matching

I-Tak Ieong, Hao Tang

HGWM is a hierarchical graph-guided world model for zero-shot object navigation. It uses an LLM to generate a category-conditioned room–anchor goal prior and maintains a persistent spatial-semantic graph from RGB-D observations. A coarse stage scores candidate headings with a vision-language model, and a fine stage matches observed objects/relations to the goal prior, fused with phase-dependent arbitration. The authors report improved SR/SPL versus WMNav on HM3D and MP3D.

DrivingAI summary
2026-10-26 · arXiv:2610.32562

Shared Is Not Enough: Predictor Width and Factor Readability Under Masked Latent Prediction

Nilotpal Kakati

The authors study masked latent prediction (I-JEPA) by training small ViT encoders in a simulator with known image factors. They probe how readable a factor shared by two views remains after training. With the default (smaller) predictor, shared-factor readability is about 40% of SimCLR’s. Widening the predictor to match the encoder reversed the ordering against SimCLR and BYOL. They release the simulator and trained models.

JEPAAI summary
2026-10-26 · arXiv:2610.32657

DT-JEPA: Learning to Predict Future Representations for Decision-Making

Changxu Ni, Yiyang Ni, Jun Li et al.

DT-JEPA is an offline pretraining and online fine-tuning framework for decision-making that predicts future latent representations from reward-free offline trajectories using a probabilistic JEPA objective. The predicted future latents condition a Decision Transformer via AdaLN for action generation, and critic-guided fine-tuning adapts to unseen reward objectives with limited online interactions.

JEPAAI summary
2026-10-26 · arXiv:2610.32742

Adaptive World-Models for Anomaly Recognition (AWARE): A Framework for Robot Dynamics Prediction and Introspection

Luke Beddow, George Mavroghenis, Rodrigo Chacon Quesada et al.

AWARE is a world-model framework for anomaly detection and introspection using only external camera monitoring. It estimates latent dynamics parameters and predicts future robot states; anomalies are detected via Mahalanobis distances on latent deviation and prediction error. On 240 real robotic crane trajectories, AWARE reports 0.93 AP (low noise) and 0.77 AP (high noise), and 88% accuracy in localizing the degrading motor.

RoboticsCode availableAI summary
2026-10-26 · arXiv:2610.32776

Predictive Long-Horizon Progress Modeling with Action-Aware Quantification

Tianchi Xu, Yalin Zheng, Yangxin Zhu et al.

The authors propose action-aware quantification for dense, absolute long-horizon progress feedback in robotic manipulation by assigning weighted action credits within object-level subtasks and handling temporary attempts and explicit credit revocation. They train NimCritic, a dual-diffusion-transformer predictive progress critic via Joint Video–Progress Learning with Historical Progress Perturbation and Video–Progress Alignment. It matches or outperforms prior process reward models on VPB/LOOM and retrieves successful rollouts on RoboArena.

RoboticsAI summary
2026-10-26 · arXiv:2610.32965

RynnUnify: Towards Unified Embodied Brain

Minghao Zhu, Bohan Hou, Kun Xiang et al.

RynnUnify is a 16B unified embodied model (MoT-based) that combines an autoregressive reasoner with a diffusion rectified-flow video generator via interleaved video–language reasoning. It is evaluated on long-horizon visual planning (Rynn-VP, Rynn-GEU) and extended to agentic reasoning (RL) and robot control (WAM).

RoboticsAI summary
2026-10-26 · arXiv:2610.32977

UniAction: Unified Action Conditioning for World Models

Bowen Xue, Zizhang Li, Hong-Xing Yu et al.

UniAction unifies action conditioning for world models by representing any action as a sparse set of action elements with (1) discrete semantic identities and (2) continuous spatiotemporal grounding coordinates. These coordinates are mapped into the pretrained video model’s native positional geometry, and cardinality-calibrated attention compensates for varying numbers of action elements. On four action forms (force, camera control, hand pose, robot pose), UniAction outperforms conditioning alternatives and a jointly trained single model surpasses action-specific experts.

2026-10-26 · arXiv:2610.32978

SceneBloom: Agentic Indoor Environment Generation with Realistic and Diverse Layouts for Embodied Learning

Yuhan Liu, Lianhui Qin, Shenji Wan

SceneBloom generates simulation-ready 3D indoor environments for embodied learning. It samples furniture layouts per functional zone using diffusion priors, then compiles proposals into 3D with contract-guided fitting: relation contracts extracted from the proposals preserve qualitative relations while geometric fitting repairs poses. Across five room types, the authors report highest visual quality and physical stability, with no detected mesh collisions and no out-of-bounds objects.

DrivingCode availableAI summary
2026-10-23 · arXiv:2610.30160

V-JEPA 2 Meets AQA: Phase-Based Scoring and Reference-Based Feedback for Mountain Biking

Santhiya Rajan, Savio Cardoz

The authors study mountain-bike action quality assessment (AQA) with weak binary technique labels and candidate corrective feedback. They use frozen V-JEPA 2 video features plus image-relative 2D pose/motion features aggregated over four heuristic phases. A small transformer predicts a binary technique score; a reference comparison produces phase/body/direction cues. In five-fold pooled out-of-fold evaluation, AUROC=0.768 and AP=0.812.