Digest

Reset

1167 papers

2026-10-12 · arXiv:2610.13194

JANUS: Enhancing Long-Horizon Autonomous Research with Co-Evolving World Models

Mengdi Liu, Chenghao Jia, Hong Chang et al.

JANUS is a long-horizon autonomous research framework that replaces reactive Generate–Execute–Feedback with a Predict–Verify–Evolve loop. It maintains an evolving DAG, uses a co-evolving research world model to forecast candidate value improvement, information gain, and exploration utility, and scores/refines/prunes branches before execution. After ground-truth feedback, JANUS distills experience into hierarchical skills and reflection/coordination protocols for continual co-evolution.

DrivingAI summary
2026-10-12 · arXiv:2610.13212

Beyond Actions: Learning Future Operation Targets for Vision-Language-Action Models

Min Lin, Bokai Zhou, Jianqi Lin et al.

The authors propose Action–Future Coupled Effect Learning (AFCE) for vision-language-action models: instead of predicting action chunks directly, the policy predicts temporally ordered physical “Effect” tokens grounded in demonstrated executed actions, realized robot responses, and sparse visual transitions. A flow-matching model predicts Effects from current image+instruction+state, then a state-conditioned decoder outputs action chunks.

DrivingAI summary
2026-10-12 · arXiv:2610.13217

Golden Anchor for Video Generation

Jinyuan Xu, Shitong Shao, Lichen Bai et al.

The authors study how “anchor” images (produced by a text-to-image model) affect image-to-video (I2V) generation, and find that image aesthetic/quality scores weakly correlate with downstream video quality—especially structure/texture, which is negatively correlated. They propose Golden Anchor Learning (GAL), which fine-tunes a T2I generator using downstream video rewards via offline preference learning (DPO). GAL improves VBench and video/frame reward scores and yields consistent motion gains.

2026-10-12 · arXiv:2610.13237

The Predictability Bias of Joint-Embedding Predictive Learning in Medical Imaging

Fabian Drexel, Bernhard Kainz, Jakob N. Kather et al.

The authors identify a “predictability bias” in joint-embedding predictive learning: point prediction preserves only what hidden targets are predictable from visible context, not clinically relevant variation that the context cannot resolve. In TAP-JEPA (ViT-B/3D on 113,611 CT volumes), prediction reduces pathology discrimination much more than anatomy and shifts pathology representations toward matched controls.

JEPAAI summary
2026-10-12 · arXiv:2610.13315

JEPA-kMMVAE: CSI Prediction for Time-Varying Wireless Channels

Efstathios Chatziloizos, Konstantinos Vandikas, Aneta Vulgarakis Feljan et al.

JEPA-kMMVAE predicts the next clean multi-antenna CSI snapshot from a finite history of noisy CSI. It combines a joint-embedding predictive architecture with a k-memory Markov variational autoencoder for temporally structured latent representations. The authors provide direct and residual decoding variants (residual uses a denoising backbone and residual references). Simulations on 3GPP UMa channels show residual JEPA-kMMVAE is best at moderate/high input SNRs, while sequence baselines win under stronger noise.

JEPAAI summary
2026-10-12 · arXiv:2610.13409

ManeuverFlow: Maneuver Hypothesis Generation with Future Evidence for Trajectory Evaluation and Refinement in End-to-End Autonomous Driving

Yuhao Wei, Yong Chen, Chen Yang et al.

ManeuverFlow generates explicit maneuver-structured trajectory hypotheses in one step, then evaluates and refines them using current and future evidence from a shared world model. It selects maneuver components via scene-conditioned Gaussian-mixture priors, transports samples with a maneuver-routed MeanFlow field, extracts trajectory-specific PDM-aligned evidence from a candidate-independent future BEV field, refines selected candidates with Frenet residuals, and jointly re-scores retained+refined trajectories.

DrivingCode availableAI summary
2026-10-12 · arXiv:2610.13504

Beyond Temporal Proximity: Explanatory World Models for Credit Assignment in Reinforcement Learning

Oriol Corcoll, Qiyao Wei, Ciar\'{a}n Mark Gilligan-Lee

The authors introduce Explanatory World Models (EWM), an attribution framework that uses explanations of an observed outcome to redistribute credit to earlier events. They instantiate EWM with a frontier language model and a conservative hierarchical redistribution operator that recursively reallocates reward while preserving undiscounted episode return. Across Crafter, WordCraft, KuaiRand (offline/online), and an offline industry recommender evaluation, EWM improves credit assignment especially when rewards are highly delayed.

Latent DynamicsAI summary
2026-10-12 · arXiv:2610.13522

Beyond Successful Robot Demonstrations: Learning from Human Interactions and Modeling Failure Dynamics

Peiyan Tu, Cong Wang, Hanxin Zhu et al.

HF-WM is an action-conditioned world model that learns interaction dynamics from egocentric human videos and improves failure modeling via GRPO post-training on counterfactual action perturbations. It uses a dual-branch architecture with stabilized pointflow supervision to bridge the human–robot embodiment gap, and GRPO with action alignment plus physical plausibility rewards to model plausible failure outcomes without paired failure video supervision.

2026-10-12 · arXiv:2610.13732

LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos

Zhiyuan Xiang, Peidong Liu, Mingyang Li et al.

LAPT-Scale pretrains embodied VLA policies using human egocentric videos without robot action labels. It introduces LAPT-Scale-LAM, a continuous latent action model trained to explain frame-to-frame transitions using perceptual, semantic (DINOv2), and optical-flow (RAFT) alignment. LAPT-Scale-Policy uses these latent actions for Stage-1 pre-training, then performs target-robot embodiment alignment and task post-training. The authors report improved real-robot success and scaling gains with more annotated human video.

Latent DynamicsAI summary
2026-10-12 · arXiv:2610.13740

SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices

Timi Owolabi, Hikmah Olawore, Frances Adelakun et al.

SmolWAM is a compact latent world-action model for robotic manipulation on edge devices. It distills from the Cosmos Policy into a single-frame, one-step rectified-flow readout that predicts actions and future camera latents. On LIBERO, it reports 95.5% success over 400 episodes with 57 ms per 16-action chunk on an A100, claiming 11× lower latency and 3.2× lower memory than the five-step teacher pipeline.

RoboticsAI summary
2026-10-12 · arXiv:2610.13816

Dream to Remember: Continual Steering of Generalist Robot Policies with World Models

Bahey Tharwat, Iman Nematollahi, Marvin Alles et al.

Dream to Remember (D2R) continually steers a frozen generalist robot policy toward sequential new tasks while retaining earlier skills. It learns a small steering module using imagined rollouts from a frozen world model; the same world model rehearses prior tasks by regenerating their rollouts on demand, regularizing the steering module via a KL penalty. On CALVIN it improves 8 tasks entirely in imagination, and on LIBERO-90 it learns unseen tasks in the environment while preserving the policy’s earlier skills.

RoboticsAI summary
2026-10-09 · arXiv:2610.10571

SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction

Minsu Kim, Ye-Sung Kim, Hyeseong Jeon et al.

SPERA is an EEG foundation model that adopts a JEPA-style latent prediction objective and predicts masked target latents from visible context. It adds a Legendre-polynomial spherical prior to encode electrode geometry in attention, uses factorized temporal/spatial attention with periodic full-attention blocks, and applies a relational spectral regularizer for frequency-aware latents. Pretrained on ~80,000 hours from 29,048 subjects across 106 datasets, it reports the highest average balanced accuracy on nine downstream EEG tasks.

JEPACode availableAI summary
2026-10-09 · arXiv:2610.10778

MemoWM: How World Models Change What Agents Need to Remember

Bingfan Zeng, Zhisheng Chen, Chenbo Sang et al.

MemoWM is a long-term episodic memory framework that uses a shared, action-conditioned world model as a predictive prior to compress stored experience. It allocates what to retain per discrete coordinate by balancing expected task impact of reconstruction errors against storage cost, then reconstructs omitted content in a closed loop without transmitting retention masks. Across five benchmarks, it reports 42.42% average answer accuracy and 53.9% lower experience-specific storage vs. MIRIX, while releasing code.

Latent DynamicsCode availableAI summary
2026-10-09 · arXiv:2610.10810

Diagnosing and Recovering from Observation-Space Shift at Long-Horizon Skill Seams

Pranav Wagh, Yu Fang, Yue Yang et al.

The authors study Observation-Space Shift (OSS) in long-horizon robotic manipulation where chaining independently trained skills causes downstream failures at skill seams. Using privileged simulator resets, they attribute the dominant shift to displaced articulated scene state and secondary objects. They build a detect–restore–resume system; on BOSS-44 it raises full-chain success from 7.6% to 26.5%, while tested alternatives (resampling, Diffusion Policy, and world-model baselines) fail from the evaluated seam states.

RoboticsAI summary
2026-10-09 · arXiv:2610.10846

Cross-Embodiment Robot Foundation World Models with Latent Actions

Huang Huang, Sriram Yenamandra, Arjun Majumdar et al.

The authors introduce LAC-WM, a robot world model that uses a learned unified latent action space shared across embodiments. Compared to EAC-WM, which conditions on explicit motion labels via embodiment-specific action encoders, LAC-WM improves cross-embodiment downstream performance: up to +46.7% on dexterous manipulation and +11.7% on a modified LIBERO. The unified latent action space also enables downstream scaling with the number of pretraining embodiments.

DrivingAI summary
2026-10-09 · arXiv:2610.10850

Similar Predictive Fit but Different Latent Dynamics: Characterizing Learned Dynamical Structure in Personalized Models of Brain Disorders

Rita Huan-Ting Peng, Nhat Bui

The authors study whether personalized latent dynamics differ between epilepsy and non-epilepsy when predictive fit is similar. Using frozen EEG foundation representations and subject-specific mLTD transition-dependency graphs, they find epilepsy models retain denser dependency structure (e.g., nonzero dependencies and densities) at k=4 and k=6, while held-out log-likelihood and within-subject next-state AUROC are nearly matched.

Latent DynamicsAI summary
2026-10-09 · arXiv:2610.10932

World-Model Policy Arbiter for Goal-Conditioned Reinforcement Learning

Junwei Quan, Evgenii Opryshko, Nicholas Rhinehart et al.

The authors propose World-Model Policy Arbiter (WMPA), a test-time framework that arbitrates among a bank of frozen goal-conditioned policies. WMPA rolls out each candidate in a learned state-space world model, scores imagined futures with a shared goal-conditioned value function, and executes the highest-scoring policy for a commitment interval before re-arbitrating. On 18 OGBench state-based datasets, it improves macro-average success from 44% to 58%.

DrivingAI summary
2026-10-09 · arXiv:2610.11060

AffordDrive3D: Affordance-Aware World-Action Modeling with Spatial Understanding

Tianhui Cai, Xinglong Sun, Chao Fang et al.

AffordDrive3D is an affordance- and geometry-aware world-action model for autonomous driving. It uses a VLM backbone to jointly forecast action-relevant regions (drivable areas and collision-critical regions) and RGB-conditioned future geometry from RGB world-model latents, while an action diffusion model generates future ego trajectories. On NAVSIM, the authors report state-of-the-art 91.3 PDMS and 89.9 EPDMS.

DrivingAI summary
2026-10-09 · arXiv:2610.11161

VGGTWorld-VLA: Intent-Conditioned 3D World Evolution for Autonomous Driving

Zhaoyang Liu, Kun Jiang, Ziying Song et al.

VGGTWorld-VLA extends VGGT-World with driving-intention control for autonomous driving. It injects ego action and VLA-derived semantic tokens into the future-token stream via a geometry–language–action bridge and layer-wise conditioning, enabling different future 3D geometry predictions from the same observation under different ego actions. Evaluated on NAVSIM for future geometry/depth prediction, the authors report improved Token MSE, RMSE, and δ1 versus VGGT-World.

DrivingAI summary
2026-10-09 · arXiv:2610.11174

IntactWorld: Joint World Modeling with Intact Features

Boming Tan, Xiangdong Zhang, Yan Xia et al.

IntactWorld is a joint video world-modeling architecture that uses “intact” (uncompressed) semantic/spatial/temporal features. To avoid a manifold-gap bottleneck, it predicts clean world features x0 at intermediate layers while keeping velocity prediction for VAE latents, computing the training loss in velocity space. It also introduces a Full-to-Compact training and compact CLS inner-guidance to reduce inference overhead.