Digest

Reset

1167 papers

2026-10-30 · arXiv:2610.39306

CAD-WAM: Anticipating Geometry for Text-to-CAD Generation

Zhengyang Li, Weijie Zheng, Yuxin Cui et al.

CAD-WAM generates text-to-CAD models by predicting both the next executable CAD-DSL action and the next geometric state. It encodes intermediate solids as latent states from surface point clouds, and uses an auxiliary state-prediction head for geometric supervision during training. At inference, each predicted action is executed by a CAD kernel and the resulting solid is re-encoded for the next decision, improving geometric accuracy and validity on DeepCAD.

DrivingCode availableAI summary
2026-10-30 · arXiv:2610.39367

Utility-WAM: Correcting Future-Induced Action Changes

Feihong Zhang, Xi Lin, Yulong Shi et al.

Utility-WAM studies how predicted visual futures change robot actions via an action delta between future-conditioned and future-free execution. Authors report futures barely change average success (+0.8 LIBERO, −0.8 RoboTwin 2.0) but increase inference latency (1.7–1.9×). They introduce a residual adapter plus a pre-generation gate to correct future-induced action changes and skip video when unhelpful.

RoboticsAI summary
2026-10-30 · arXiv:2610.39516

Robust LIDAR Sensor-to-Sensor Domain Generalization via Accumulated-Point Scene Completion

Christoph B Rist, Larissa T Triess, Markus Enzweiler et al.

The authors propose a source-only domain generalization method for LiDAR semantic segmentation that keeps the input scan unchanged. During training, they aggregate source-domain scans into an accumulated world-frame “world model” and use it as enriched supervision via an occupancy prediction auxiliary task (including free-space points). At test time, they drop the occupancy head for pure segmentation or keep it for scene completion.

DrivingAI summary
2026-10-30 · arXiv:2610.39528

PRIME W0: Physical Priors from World Models for Embodied Manipulation

Huabin Peng, Fagang Jin, Yancong Wei et al.

PRIME W0 is a hierarchical conditional flow-matching VLA for embodied manipulation. It injects per-layer VLM KV into an Adaptive MoE action expert, uses asynchronous world-model subgoal-image prediction plus a DINOv2 historical-frame semantic bridge for dynamics-aware conditioning, and generates actions with Action Dynamic Manifold (ADM) flow matching using covariance-structured noise that respects kinematic coupling.

RoboticsAI summary
2026-10-30 · arXiv:2610.39547

Assuring Learning-Enabled Cyber-Physical Systems with Foundation Models: A Neuro-Symbolic Agenda for Testing, Verification, and Runtime Assurance

Xi Zheng, Ivan Ruchkin, Hong Jin Kang et al.

The authors argue for system-level assurance for learning-enabled cyber-physical systems that use foundation models. They propose a “Foundation Model Assurance Stack” combining (i) neuro-symbolic architectures with symbolic world models and contract specifications, (ii) conformance-based, scenario-driven testing with foundation-model-assisted specification mining and synthesis, and (iii) compositional verification plus runtime assurance using monitors and enforcement via a Simplex-style layer.

DrivingAI summary
2026-10-30 · arXiv:2610.39619

FutureWorld: A Plug-and-Play Causal Next-State Predictor for Interactive Video World Exploration

Guangyuan Li, Kelu Yao, Shuolin Xu et al.

FutureWorld introduces a plug-and-play Causal Next-Chunk Predictor (C-NCP) that predicts the next latent state from the current generated latent and the next action, then modulates a frozen autoregressive video DiT to generate the next chunk. The authors aim to reduce error recirculation from reusing imperfect generated history, improving long-horizon interactive consistency across multiple generators and benchmarks.

2026-10-30 · arXiv:2610.39721

WorldProbe: When Video World Models Fall Apart under Long-Horizon Spatial Exploration

Minqi Wang, Jie Peng, Hejun Dong et al.

WorldProbe is a diagnostic benchmark for video world models under long-horizon spatial exploration. It uses a goal-conditioned multimodal agent to adapt camera actions online during generation, with 164 test cases across four probing families. Across 10 state-of-the-art models, the authors run 1,600+ rollouts totaling ~800 minutes with horizons up to 180s, finding local quality degradation and long-range inconsistencies, including weak long-range memory and evolution.

Video World ModelsCode availableAI summary
2026-10-30 · arXiv:2610.39812

GROME: GROunding Factorized MEmory into Video World Models

Fang Yu, Xiaoxue Wu, Haiyu Zhang et al.

GROME augments action-conditioned video world models with factorized, local video memory for long-horizon coherence. It decouples historical evidence into static (scene) and dynamic (entity) components, retrieves them with distinct criteria to avoid geometric drift and stale dynamics, and injects the retrieved references into a frozen backbone via training-free conditioning (including value adaptation and disentangled attention).

2026-10-30 · arXiv:2610.39816

PASA-KV: Attacking Action-Facing Visual Routes in World Action Models

Zhongbo Wu, Jiakai Huang, Zheng Zhang et al.

The authors study when perturbations to internal visual K/V routes in world action models become behaviorally consequential. They find a mismatch: layers with larger representation displacement are not necessarily more vulnerable. They attribute this to action-conditioned readout, where keys route action queries and values determine what content is read. They introduce Tail-AV and PASA-KV to select action-facing K/V targets before training a task-shared patch attack.

DrivingAI summary
2026-10-30 · arXiv:2610.40016

LOGIV: Graph-Verified Execution with Causal Local Repair for Long-Horizon Robot Tasks

Enqiao Lu (CUHKSZ, THU), Yuxiang Xie (CUHKSZ) et al.

LOGIV is a graph-based execution and repair layer for long-horizon robot tasks. It compiles a validated symbolic plan into a task graph with condition-support and resource-ordering edges, checks physical validity online during execution, and uses dependency tracing plus constrained local repair to update only affected parts while retaining usable unfinished actions. It operates above existing VLA/WAM policies without adding trainable components.

RoboticsCode availableAI summary
2026-10-30 · arXiv:2610.40026

RoboSeed: Multidimensional Evaluation of Vision-Language-Action and World-Action Models: Capability, Resource Affordability, and Research Evidence

Xin Ma, Furong Chen, Xinlong Man et al.

The authors propose RoboSeed, a multidimensional evaluation framework for vision-language-action (VLA) and world-action models (WAM). It scores models on five 0–100 dimensions (capability, training affordability, deployment affordability, locally verified research value, reproduction evidence) using fixed normalization anchors, explicit missing-data rules, and audit criteria. They apply it to 12 checkpoint units across VLA/WAM families with reproducible scoring code and machine-readable evidence, but with incomplete deployment measurement in the current snapshot.

DrivingAI summary
2026-10-30 · arXiv:2610.40030

Cheap Retrieval and Observation Credit Assignment from One Quantity: The Value of Information

Daniel Strauss, Nikolaos Sotirakis, Zhenshan Bing et al.

The authors propose “value of information” for credit assignment and retrieval in memory-based RL. They estimate information value from archived rollouts using a learned world model. When read at the observation, it yields a provable lower bound that floors TD targets and attributes the floor to the observation that earned it; read at the decision, it defines an optimal retrieval set. A pilot “Memory Navigation” system uses a reduced-context retrieval setup.

DrivingAI summary
2026-10-30 · arXiv:2610.40191

Video Generation Meets Agent Harness: Agentic Video Generation, Agentic World Modeling and Recursive Self-Improvement

Liaoyuan Fan, Yuhao Zheng, Dongxing Mao et al.

The authors survey how agent harnesses—executables that manage context, tools, state, feedback, and budgets—connect agentic video generation, agentic world modeling, and recursive self-improvement (RSI) for video models. They frame harness roles in production workflows and interactive environments, then discuss RSI as improving the procedure that produces later improvements, with harnesses as an inspectable surface.

AI summary
2026-10-30 · arXiv:2610.40207

RoboAtom: Large-scale Atomic-Action Supervision for Robot Learning

Xueyuan Wei, Xuanxu Wang, Xudong Li et al.

RoboAtom builds RoboAtom-Data by adding unified, time-aligned atomic-action annotations (skill category, action description, start/end frames) to trajectories from 16 sources. It also introduces RoboAtom-Bench and trains RoboAtom-Anno for automatic atomic-action annotation and hierarchical execution for robot control, reporting improved annotation performance and task success over task-level baselines on multiple benchmarks.

RoboticsAI summary
2026-10-30 · arXiv:2610.40248

LLM-guided Planar Loco-Manipulation for Quadruped Robots

Georges Jetti, Michael Khayyat, Jean-Pierre Sleiman et al.

The authors propose an LLM-guided event-driven planner for interactive navigation with movable obstacles: it jointly selects a skill (navigation or planar manipulation), the relevant object, and a geometric target, producing a variable-length sequence of parameterized actions executed by low-level PPO policies. They fine-tune Qwen2.5-1.5B-Instruct with LoRA on synthetic data and report improved interactive navigation over baselines, plus an end-to-end whole-body planar pushing policy for manipulation.

DrivingAI summary
2026-10-30 · arXiv:2610.40276

TaskFrame: Connecting Task-Level Planning with Frame-Level Control for Game Agents

Guangzhi Wang, Junfu Pu, Yiran Wang et al.

TaskFrame is a hierarchical planner–controller for Minecraft game agents. A vision-language task planner decomposes a high-level goal into an ordered sequence of atomic instructions once per task, using supervised learning and dependency-aware GRPO. Special goal-token hidden states are cached as goal embeddings and reused by a per-frame causal Transformer controller that outputs native keyboard/mouse actions and continuous camera motion every game step. Trained in three stages, it attains higher combat success than evaluated baselines.

DrivingAI summary
2026-10-30 · arXiv:2610.40280

Latent Prediction Yields Cross-Subject Alignable Representations in Sensory-Evoked fMRI

Serwan Jassim, Victoria Bosch, Daniel Anthes et al.

The authors adapt JEPA to pre-train participant-specific fMRI encoders using brain-only data from NSD. Encoders never see stimulus images or stimulus-derived representations. They evaluate frozen encoders with MindEye2 for image reconstruction and CorText for brain captioning, including cross-participant generalization. The authors report that independently trained participant embeddings can be aligned fully unsupervised without shared stimuli, enabling cross-subject decoding without fine-tuning.

JEPAAI summary
2026-10-30 · arXiv:2610.40339

Wizard: Manipulate Any World in Real Time via Agentic Control

Yuzhe Wu, Zijun Lin, Yeying Jin

Wizard is an agentic game world model for real-time, precise character motion control with zero-shot generalization across novel scenes, characters, and actions. It uses structured geometric proxies (scene geometry/depth-height) and articulated skeletons, plus an adaptive masking mechanism that gates reference attention to corresponding character regions. An agent maintains explicit game states and updates them for mechanics-consistent combat during closed-loop play at 24 FPS.

2026-10-30 · arXiv:2610.40356

JEDI: Joint-Embedding Demonstration Compression for In-Context Imitation Learning

Jiazhao Liang, Zhewen He, Shuaihang Yuan et al.

JEDI compresses in-context imitation demonstrations into k latent slots trained to predict the query episode’s next-stage keyframe in a frozen video-JEPA (V-JEPA) space. The predictor selects which cached demonstrations to place in context. Post-training enables conditioning on human-hand demonstrations via keypoint tokens. Experiments on LIBERO and real robots show improved many-shot context performance and better success/demonstration following than action-only compression.

JEPAAI summary
2026-10-30 · arXiv:2610.40373

HiFOR: Hierarchical Foresight with Action-Guided World Grounding for Vision-Language-Action Model

Rongxing Ding, Hongyu Qu, Kehong Wu et al.

HiFOR is a hierarchical foresight VLA framework for robotic manipulation. It jointly predicts short-term state evolution, intermediate semantic milestones, and final goal states using a Hierarchical Dependency Mask (HDM). Predicted futures are organized into a structured latent world (horizon/view/spatial) and retrieved by Action-Guided Factorized World Retrieval that uses denoising-step action representations to select relevant future information for action generation.

DrivingAI summary