Digest

Reset

1167 papers

2026-11-03 · arXiv:2611.00734

MOST: Memory of Object States and Task Progress for Long-Horizon Mobile Manipulation

Diqi He, Jinbang Guo, Yijun Shen et al.

MOST is an instruction-conditioned, end-to-end policy for long-horizon mobile manipulation that uses policy-internal memory structured around task progress and object-state changes. It compresses multi-view history into temporal tokens, supervises historical and predictive future tokens with subtask and object states during training, and conditions action generation without an external planner or retrieval stage. The authors introduce RoboCasa-Long and report consistent gains over XR-1 and a visual-history-only variant.

DrivingCode availableAI summary
2026-11-03 · arXiv:2611.00758

Old Is Not False: Valid-Time Memory for Interactive Video World Models

Ziming Wang, Chuyue Gong, Hongliang Cao et al.

The authors identify an undo failure mode in interactive video world models: when an agent restores a region to an earlier state, memories that rely on capture-time ordering can make generators draw outdated content. ReDate keeps all frames but assigns tokens a valid time derived from the agent’s action log, so a reverted region is re-dated as having “just reached” its state. The authors report large gains on FAR and HY-WorldPlay reverted steps, and provide theory and a calibrated state metric.

2026-11-03 · arXiv:2611.00951

Why Koopman Structure Does Not Help Differentiable World Models for Humanoid Whole-Body Tracking

Haizhou Ge, Mengshuo Wang, Yufei Jia et al.

The authors test whether Koopman structure improves a differentiable world model used to train a simulated Unitree G1 whole-body tracking policy via BPTT. They find no gain: return is at best equal (0.823 vs 0.830), Koopman has 1.3–1.6× higher one-step error, and training is ~15% slower. Lifted rollouts diverge unless re-lifting at every imagined step; then Koopman effectively becomes action-affine, which is not better than an MLP.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.01001

GURU: A Hierarchical Neuro-Symbolic Multi-Agent System for Verification-Aware Automation of Mission-Critical Digital Engineering and Operations

Allan Grosvenor, Anton Zemlyansky, Abdul Wahab et al.

GURU is a hierarchical neuro-symbolic multi-agent system for verification-aware automation of mission-critical digital engineering and operations. Learned agents propose, rank, and accelerate actions, while a symbolic rules engine checks consequential actions against hard limits before execution; validated physics confirms and decides in a closed control loop using a shared internal language. Reported results include CFD speedups and higher ParaView control success, and improved conjunction-risk performance for a frozen self-supervised world model.

DrivingAI summary
2026-11-03 · arXiv:2611.01039

pico-WAM: Evaluating Counterfactual Prediction and Planning Error in a Compact World Action Model

Keivalya Pandya

pico-WAM is a compact world action model with a shared latent and explicit physical state readout to isolate whether control failures come from proposal generation or trajectory scoring. Using Push-T (206 demos, 15 seeds), counterfactual branch training improves cost ranking and increases policy-restarted vs warm-started coverage by 4.8 points. Using exact simulator costs for scoring boosts coverage by 12.2 points while keeping learned proposals; simulator-based proposal generation alone shows no detected gain.

DrivingAI summary
2026-11-03 · arXiv:2611.01139

World Models Need Not Be Right to Rank Policies Right: Relative Bounds and Audit Limits

Ahanaf Hasan Ariq, Hasaan Ahmad

The authors derive a relative simulation identity/bound showing when shared world-model errors can preserve correct pairwise policy rankings even if absolute value estimates are wrong. They propose RAVEL, a selective policy tournament that orders candidates via model rollouts but certifies eliminations only with held-out real audits. In DMC cheetah-run and walker-walk at horizon 500, rankings stay strong (Kendall τ 0.962/0.800) while matched audit errors are weakly correlated (0.007/0.090).

DrivingAI summary
2026-11-03 · arXiv:2611.01283

DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving

Ziming Liu, Yizhan Qu, Leichen Wang et al.

DriftingPlanner trains an end-to-end trajectory decoder with the drifting loss alone, using a one-step generator and a scene-isolated attraction–repulsion field (one ground-truth positive per scene). At inference it removes the “repulsion shell” via antithetic pairing of noise latents in one batched forward pass. On NAVSIM navtest it reports 93.75 PDMS state-of-the-art and 0.13 PDMS from the best iterative diffusion/flow matching setting (10 steps).

DrivingAI summary
2026-11-03 · arXiv:2611.01600

SCOUT: Planning under Occlusion via Object-Centric World Model Rollouts

Amir Hossain Raj, Xuesu Xiao

SCOUT is an occlusion-aware planning framework for cluttered manipulation. From sparse frontal RGB-D views, it reconstructs an object-centric hidden scene (GHOST) and derives a relational planning state with blocker relationships and accessibility corridors. An LLM proposes structured object-direction action candidates, and MCTS evaluates them using a learned action-conditioned world model with physical rollout predictions. The authors validate in simulation and on a Spot-based real shelf setup.

RoboticsAI summary
2026-11-03 · arXiv:2611.01666

Stand-In Effects and Pooling Amplification: Lessons from Biological World Models for Testing Action Effects

Bowen Xu, Boyu Chen, Jiaying Li

The authors show that common action-conditioning diagnostics for world models can report nonexistent “action effects” due to autocorrelated designs (future actions acting as stand-ins for history) and due to pooling amplification when many sites share one action sequence. They propose a design-based conditional randomization test using e-values (valid in finite samples under a known design law) and show tempering is a statistic choice that can fail, while fixed grids/mixtures can recover power.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.01677

Xiaomi Auto World Model 2.0: Causal World Models for Real-World Driving and Embodied Intelligence

Zhenxin Zhu, Zhijian Huang, Cheng Chi et al.

Xiaomi Auto World Model 2.0 is a streaming causal world model for Physical AI. It uses View-Temporal Attention (VTA) with multi-view rotary position encoding and a WorldRec 2.0 module that reconstructs static and dynamic 3D Gaussians in a single forward pass, enabling long-horizon view-consistent generation. It incorporates Gated Linear Attention (GLA) for efficient streaming inference and trains with a physical-interaction data pipeline covering long-tailed events.

DrivingAI summary
2026-11-03 · arXiv:2611.01729

ODE Policy: Robot Learning via Action Dynamics, Not Actions

Jindou Jia, Jingliang Li, Kuangji Zuo et al.

ODE policy learns the dynamics of robot actions as an ODE and integrates a learned action vector field over physical time, so executed trajectories are continuous by construction. Trained by matching integrated actions to targets in IL (over action chunks) and in RL (single control steps), with a diversity head for multi-modal behavior. It is evaluated across IL (including VLA and world action model heads) and model-based RL on multiple real and simulated robot tasks.

RoboticsAI summary
2026-11-03 · arXiv:2611.01731

Robust EEG-based Emotion Recognition from Coarse Trial-level Labels under Intra-trial Affective Variability

Xiaocong Zeng, Buxuan Wang, Xin Yang et al.

VARY is a variability-aware robust learning framework for EEG-based emotion recognition when only coarse trial-level labels are available. The authors use Two-stage Sample Selection (TSS) to select windows with reliable propagated labels, via a Label-Prediction Consistency Criterion (LPCC) and confidence filtering, and introduce an EEG-tailored time-series augmentation (MixAug). On label-noise learning on DREAMER and SEED, VARY reports improved test accuracy over best baselines.

AI summary
2026-11-03 · arXiv:2611.01815

When Does Latent Prediction Error Support Physical Surprise Claims in Video World Models?

Anton Hristov, Aleksandar Slavov, Vasil Vasilev

The authors argue that video world-model prediction error is not a reliable measure of physical surprise: it can rise or fall due to appearance, static content, motion, or ordinary prediction difficulty. Using IntPhys2, Physion++, and Kubric, they show appearance is easy to decode from frozen features, action sensitivity exists, but native prediction error can order physical-possibility outcomes inconsistently and can be lower for physically impossible events.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.01844

OpenNWM: A Generalist Navigation World Model with Latent Action Pretraining

Dujun Nie, Ruiqi Song, Zixuan Wang et al.

OpenNWM is a generalist navigation world model that learns universal latent actions from large-scale action-free navigation videos, then aligns them with physical motions via action-conditioned post-training. It introduces NavAnywhere (17.5M frames from 15 sources) and evaluates improved visual prediction and downstream navigation performance on in-domain and out-of-domain benchmarks.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.01866

The (Butter)fly Dream: Linear Separability of World-Model Rollouts in a Connectome-Constrained Visual Model

Junichiro Niimi

The authors test whether a fixed connectome-constrained fly visual system (trained only for optic-flow estimation) can tell whether stimuli come from real video or from “dreams” generated by world models. Using a classifier two-sample test on temporal statistics of cell-type responses with sequence-level cross-validation, the authors report AUCs 0.85–0.99 for nine dream conditions in the trained observer, but 0.58–0.75 in the untrained observer.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.01986

RubricWorld: World Model-Guided Rubric Optimization for LLM Graders

Hongji Pu, Hang Li, Yucheng Chu et al.

RubricWorld is a framework for optimizing LLM grading rubrics without retraining the frozen grader. It trains a world model to predict, from the current rubric state and a proposed edit, correction and degradation rates at ordinal score boundaries using paired before/after grading results. It selects a candidate for evaluation and uses held-out confidence bounds to certify updates. The authors report best results on 20/27 metric comparisons across three benchmarks.

AI summary
2026-11-03 · arXiv:2611.02218

Beyond Ensemble Disagreement: Predictive STL Monitoring of RL Locomotion under Distribution Shift

Qurban Ali, Badaruddin Chachar, Maqsood Ali et al.

The authors evaluate predictive Signal Temporal Logic (STL) monitoring for frozen Ant-v5 and HalfCheetah-v5 PPO policies under distribution shift, using closed-loop world-model rollouts and a three-model ensemble. They report that predicted STL robustness ranks HalfCheetah’s future STL violations strongly, while ensemble disagreement does not reliably track observation prediction error or violations. A current-state robustness baseline recovers most ranking value; disagreement-penalized robustness does not consistently improve ranking.

DrivingAI summary
2026-11-03 · arXiv:2611.02341

AgiWorld: An Agentic Video World Model for Building Persistent Worlds

Shaoheng Fang, Yiqun Mei, Hanwen Jiang et al.

AgiWorld is an agentic video world model that builds and explores a persistent 3D world from a single image and a free-form instruction. AgiWorld-Agent plans a metric camera trajectory and structured prompts via a high-level world state. AgiWorld-WM generates video chunk-by-chunk with spatial memory (point cloud + geometry) in a closed loop, using occlusion-aware history selection and reliability-based memory maintenance for long rollouts.

2026-11-03 · arXiv:2611.02405

MirrorComm: A Communication-Efficient Cooperative Perception Framework via Mirrored World Models

Shenghan Luo, Yin Xu, Cixiao Zhang et al.

MirrorComm is a communication-efficient cooperative perception framework using mirrored world models. The receiver predicts the transmitter’s current BEV features from reconstructed history; the transmitter compares this prediction to its current observation and selectively transmits RVQ-encoded updates for positions with large discrepancies. Updated features are fused for downstream detection/segmentation, enabling prediction during communication outages.

DrivingAI summary
2026-11-03 · arXiv:2611.02474

Gnomon4D: Learning Where and When to Focus for Geometric Motion-Grounded World-Action Modeling

Lingwei Dang, Yan Zhou, Juntong Li et al.

Gnomon4D is a world-action model that jointly generates future video, robot actions, and GeoDelta: a video-aligned 4D geometric motion field encoding image-plane displacement and depth change for embodiment and object points. It uses Interaction-Salient Attention (InSAttn) to choose where to read interactions and Transition-Anchored Aggregation (TransAgg) to choose when (around grasp/release), and links action-free pretraining on EgoManip-5000H to robot mid-training on Robot-30H.

DrivingAI summary