Digest

Reset

1167 papers

2026-10-14 · arXiv:2610.14228

Latent World Action Model for GUI Agents

Seungjun Yu, Hojun Choi, Seojeong Park et al.

The authors propose a latent GUI world-action model that shifts GUI world modeling from full pixel reconstruction to action-induced future information in latent space. Using a LoRA-adapted LMM with separate action and transition heads plus an action-aware contrastive objective, the model supports action selection with a single forward pass (transition head removed at inference). They report up to 31.9% (offline) and 40.6% (online) relative success-rate gains over pixel-space world models, and up to 98.9% lower inference latency.

Latent DynamicsCode availableAI summary
2026-10-14 · arXiv:2610.14248

Akule: Fast and Scalable Multi-Robot Motion Planning via Sparse Interaction Diffusion

Manav Vora, Gokul Puthumanaillam, Aditya Penumarti et al.

Akule augments a frozen single-robot diffusion prior for multi-robot motion planning with (1) a reusable pairwise correction model Rϕ, (2) a contextual selector Uψ that chooses which directed pairwise corrections to evaluate sparsely, and (3) a composer Gω that combines selected corrections using context-dependent signed weights. Joint proposals are passed to downstream hard repair. The authors report competitive success, up to 25× speedups, zero-shot population transfer, and scalability to 256 robots.

RoboticsAI summary
2026-10-14 · arXiv:2610.14302

Learning to See through Action: Predictive Curiosity for Robots

Stelios Zarifis, Panagiotis Filntisis, George Retsinas et al.

The authors introduce DySCOVER, a self-supervised active visual learning framework for robots. A robot uses a visual autoencoder to encode RGB observations into latents and trains an action-conditioned recurrent world model to predict the next visual latent. The policy is trained with intrinsic reward that mixes visual reconstruction error (visual novelty) and world-model prediction error (predictive curiosity), improving visual reconstruction over visual-curiosity-only across multiple settings.

DrivingAI summary
2026-10-14 · arXiv:2610.14324

EV-JEPA: Elastic V-JEPA for Any Device and Cross-Device Inference

Anxhelo Diko, Jieyu Lin, Stefano Ambrogio et al.

EV-JEPA trains a single elastic V-JEPA model once for both standalone width changes and hybrid cross-device execution. Using Transition-Aware Triplet Training (T3), it jointly optimizes a smaller standalone path, a larger standalone path, and a hybrid path that switches width at a split layer where device handoff occurs. The same weights support many-to-one and variable-split configurations across benchmarks.

JEPAAI summary
2026-10-14 · arXiv:2610.14335

PILOT: Guiding Long-Horizon World-Model Planning for Robot Manipulation with Visual Subgoals and Policy Priors

Yang Liu, Hanchen Cui, Karthik Desingh

PILOT guides long-horizon latent world-model planning for robot manipulation by generating intermediate visual subgoals with a video model and using a learned policy to initialize CEM for action search. In real UR5e experiments across four tasks, the authors report 68.8% success, versus 55.0% without policy initialization, 13.8% for policy-only execution, and 0.0% for direct final-goal planning. Simulation results report persistent gains on longer horizons.

RoboticsAI summary
2026-10-14 · arXiv:2610.14351

Learning Transferable World-Action Models from Task-Paired Human Videos

Jeongeun Park, Taekyung Kim, Dongyoon Han et al.

WATCH is a cross-video world-action pre-training from task-paired human videos. Conditioned on one demonstration’s video and actions, the model predicts another demonstration’s actions and future frames from the query’s current observation. The authors report improved motion alignment across embodiments and better action/video prediction on unseen robots, and higher MimicDroid success after robot post-training.

DrivingAI summary
2026-10-14 · arXiv:2610.14400

Sparse-TD-JEPA: Improving Zero-Shot Reinforcement Learning With Sparse Latent Task Representations

Duaa Nakshbandi, Zhiyu An, Wan Du

Sparse-TD-JEPA adds an ℓ1 sparsity penalty during training to concentrate reward-predictive structure into a smaller subspace of the task encoder’s latent space, improving zero-shot RL via TD-JEPA. The authors evaluate on 65 tasks across DMC and OGBench and report higher mean performance than TD-JEPA in all 13 domain-level comparisons, with analysis showing a sharper inferred reward landscape.

JEPAAI summary
2026-10-14 · arXiv:2610.14406

Memory Without Curation: Learning Multi-Stream Latent Memory for Robot Policies

Yilin Wu, Rutav Shah, Shuaiyi Huang et al.

IMPRINT is a policy-agnostic latent memory framework for vision-language-action (VLA) robot policies that avoids history curation. It compresses multi-stream histories (visual observations, actions, proprioception) into fixed-size latent memory updated incrementally at inference. A two-stage training procedure first supervises memory via multi-stream reconstruction and task-progress estimation, then finetunes end-to-end on action prediction.

DrivingAI summary
2026-10-14 · arXiv:2610.14544

InfraWorld: Motion-Aware World Modeling for Roadside 4D Occupancy Forecasting

Songkai Wang, Xiaokai Bai, Lei Yang et al.

InfraWorld is a roadside 4D occupancy forecasting world model that treats forecasting as selective state evolution: a dense BEV state keeps global scene context, while persistent motion queries model sparse localized dynamics. It uses Dynamic Query Initiation (DQI), Long–Short Temporal Fusion (LST), and Motion-Aware Query Injection (MAQI). Experiments on InfraOcc and Occ3D–nus report state-of-the-art performance; code will be released.

2026-10-14 · arXiv:2610.14564

Chunk-wise Model-based Policy Optimization

Weixi Song, Donglin Wang, Yong-Lu Li

CMPO (Chunk-wise Model-based Policy Optimization) learns a chunk world model and a chunk actor so an H-step hypothetical rollout is generated via a single model invocation. This avoids recurrent model rollout, reduces compounding errors and instability from BPTT, and lowers training/inference cost. The authors evaluate on 19 HumanoidBench tasks and demonstrate real-world sim-to-real transfer for quadruped bipedal walking.

Latent DynamicsAI summary
2026-10-14 · arXiv:2610.14583

Remembering What Matters for Long-Horizon Robotic World Models

Hao Wu, Tianyu Shao, Shengju Qian et al.

ReCAP is a training-free inference-time context scheduler for discrete-token autoregressive robot video world models. It keeps the initial scene anchor and a contiguous window of recent visual–action blocks while evicting older generated history, keeping token/KV-cache growth bounded. Using frozen world models, the authors report improved long-horizon visual fidelity, action-conditioned consistency, and robot structure preservation vs Full Context across RT-1, BridgeV2, LIBERO-90, and CALVIN.

RoboticsAI summary
2026-10-14 · arXiv:2610.14603

AnyRank: Reliable Anytime Policy Ranking via Sequential Testing

Apurva Badithela, David Snyder, Tenny Yin et al.

AnyRank is a sequential multi-policy ranking method for robot policy evaluation under limited hardware trials. It adaptively allocates trials to hard-to-separate policies, outputs (partial) rankings at any stopping time with bounded family-wise error rate, and can incorporate prior information (imperfect simulation/world-model/held-out evaluations) to improve efficiency without invalidating statistical assurances.

Code availableAI summary
2026-10-14 · arXiv:2610.14685

Demiurge: Forging Navigable 4D Worlds via Autoregressive Video Rerendering

Hanxin Zhu, Jianqi Ma, Cong Wang et al.

Demiurge is an autoregressive video rerendering pipeline for creating coherent, traversable 4D worlds. It maintains an evolving dynamic 4D point-cloud memory shared across target camera trajectories, using memory-conditioned appearance diffusion, memory-consistent depth completion, memory-aware dual-stream warping, and optional geometry-aware refinement. The authors report state-of-the-art performance and show coherent spatiotemporal lattices, large-parallax rerendering (up to 180° rotations), and geometry-faithful outputs usable for 3D Gaussian Splatting.

2026-10-14 · arXiv:2610.14720

CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

Peizheng Guo, Cheng Zhang, Xingyu Zhang et al.

CAG-WAM analyzes action evaluation in World Action Models (WAMs) via a structural causal model that separates action-induced change from environment-driven evolution, including a backdoor path from current observation to action outcomes. For each candidate action, it predicts an action-conditioned future and an environmental future, and uses their evaluation difference as causal action gain (CAG) for training (GRPO) and inference-time action revision.

RoboticsAI summary
2026-10-14 · arXiv:2610.14727

MHWA: Multi-timescale Hierarchical World-Action Model

Pengcheng Pan, Guoqing Ma, Yuhan Zhang et al.

MHWA (Multi-timescale Hierarchical World-Action model) reduces open-loop drift in tokenized autoregressive world models by using a strided high-level context channel updated less frequently, plus context-aware routing to dynamically select a low-level Mixture-of-Experts. On a 14-game Atari subset with 10% subsampled offline data, MHWA (77M) reports IQM human normalized score (HNS) 0.884 across three seeds, outperforming JOWA-70M (0.529) and matching JOWA-150M (0.833) with fewer parameters and lower training FLOPs.

Latent DynamicsAI summary
2026-10-14 · arXiv:2610.14746

BridgeWAM: Bridging Video and Action Experts via Latent Bridge Queries

Jiaxin Fang, Ruitong Liu, Zeyu Lai et al.

BridgeWAM (Bridge-of-Experts) uses 32 Latent Bridge Queries (LBQs) to connect a 30-layer Video Expert with a two-layer Action Expert in a shared limited latent space. Joint video–action supervision routes action gradients through the LBQs to adapt video representations for control, while the Action Expert selectively retrieves from the LBQ latent via cross-attention. The authors report 98.25% success on LIBERO and 72.24% pooled success on LIBERO-Plus.

DrivingAI summary
2026-10-14 · arXiv:2610.14802

YOUDrive: Driving Style as a Composable Direction for End-to-End Autonomous Driving

Guantian Zheng, Lidong Yu, Tianyu Gao et al.

YOUDrive is a vision–language–action planner that controls continuous, composable driving style at inference. Each teacher policy is distilled into a task vector that induces a style direction in a shared VLA planner; a flow-matching decoder uses this conditioned representation to sample kinematically feasible trajectory alternatives. The authors report monotonic assertiveness control on NAVSIM navtest while keeping PDMS above 0.90, reaching 0.954, and propose a Style Alignment Score (SAS).

DrivingAI summary
2026-10-14 · arXiv:2610.14871

MORE: Learning Motion-Oriented Registers for Driving World Models and Planning

Yingdong Gu, Haoyu Wang, Shaocheng Yan et al.

MORE learns motion-oriented registers during video pretraining (a V-JEPA framework) and uses the same compact register state interface for both a non-autoregressive driving world model and trajectory planning. On NAVSIM-v2, the authors report the highest EPDMS (92.5), and in efficiency analysis they report 1/10 the inference latency of DriveLaW while keeping inference latency close to Drive-JEPA (110.0 vs 73.3 ms for Drive-JEPA).

DrivingCode availableAI summary
2026-10-14 · arXiv:2610.14936

CFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free Guidance

Shutong Ding, Zhaoyang Liu, Yimiao Zhou et al.

CFG-Policy is a plug-and-play diffusion/flow RL framework that separates exploration and exploitation into two generative policies and composes their score/velocity fields during sampling via classifier-free guidance. In online RL, the exploration branch is frozen while an exploitation branch learns from task rewards; in offline-to-online post-training, the pretrained exploitation branch is frozen while a trainable branch steers composed behavior. The authors report improved results with QSM and QVPO on 5 MuJoCo tasks and with Flow-SDE+PPO on 4 RoboTwin tasks.

DrivingAI summary
2026-10-14 · arXiv:2610.15112

Drive3R: Block-wise Streaming Geometric Foundation Models for Autonomous Driving

Tianyi Yang, Haojun Yang, Jipeng Xiao et al.

Drive3R is a block-wise streaming geometric foundation model for long-horizon autonomous driving reconstruction. It uses block-wise causal attention to keep bidirectional context within each block while enforcing causality across blocks via KV-cache reuse. It also adds confidence-aware re-anchoring that discards stale KV cache and stitches refreshed segments using SE(3) alignment.

DrivingAI summary