Digest

Reset

1167 papers

2026-11-04 · arXiv:2611.04870

PrimeSim: A Generalist World Model Simulator from Large-Scale Cross-Embodiment Demonstrations

Travis Davies, Wentao Ma, Lihao Guo et al.

PrimeSim is an action-conditioned, multiview robot world model simulator built by midtraining a pretrained video generator on 30,110 hours of robot demonstrations across 22 platforms and then posttraining it for fast autoregressive simulation with long-horizon stability. It uses unified cross-embodiment action representation, frame-aligned dual-path action conditioning, view-aware interleaved attention, and hierarchical temporal memory.

DrivingAI summary
2026-11-04 · arXiv:2611.04873

Trusting Imagined Futures: Auditing Imagined Rewards for VLA Post-Training

Zicheng Kong, Zhenbo Xu, Haoyan Zhang et al.

Trusting Imagined Futures (TIF) is an offline supervision selection method for vision-language-action (VLA) post-training that audits reward predictions from a video world model using execution feedback. A multidimensional reward model predicts utility and five execution-quality dimensions for candidate ranking, while an auditor estimates overestimation risk. TIF replaces the current supervision only when predicted advantage exceeds a margin and audit risk is below a threshold.

2026-11-04 · arXiv:2611.05023

HiLo-WAM: Decoupling Long-Horizon Task Evolution and Local Dynamics in World Action Models

Xingyu Wang, Yanjiang Guo, Zhijing Ding et al.

HiLo-WAM is a World Action Model for long-horizon robotic manipulation that decouples visual world modeling into two streams: a sparse long-horizon stream for historical task context and coarse future evolution, and a dense local dynamics stream for short-horizon robot–object interactions. It uses a dual-DiT architecture with structured cross-stream attention and progressive inference (partial/joint denoising and dynamic step skipping).

DrivingAI summary
2026-11-04 · arXiv:2611.05062

FloMo: 3D Scene Flow as a World Action Model Intermediate for Learning from Human Video

Jeremy Collins, Namra Patel, Ayush Agarwal et al.

FloMo argues that dense 3D motion mediates transfer from pretrained video generation to robot action prediction. It renders 3D scene flow as RGB and encodes it in the same latent space as a pretrained Wan 2.2 video model, fine-tuning the backbone to predict scene-flow latents and action tokens using flow-matching. Co-trained on action-free human video and action-labeled robot teleoperation, it improves in-distribution policy success and out-of-distribution generalization versus video prediction or joint video+motion prediction.

Latent DynamicsAI summary
2026-11-04 · arXiv:2611.05072

Trajectory Evaluation via Rollout-Free World Model for End-to-end Autonomous Driving

Junzhe Wang, Jinqing Zhang, Qingjie Liu

RFWorld proposes rollout-free world-model evaluation for end-to-end autonomous driving. Instead of predicting candidate-conditioned future scenes (rollout) for each candidate, it builds a shared scene memory from the current BEV and shapes it via lightweight training-time temporal readout with semantic supervision. At inference, candidate trajectories score by sparse spatial readout from the shared memory, reducing computation and avoiding rollout error accumulation.

DrivingAI summary
2026-11-04 · arXiv:2611.05080

CRONOS: Benchmarking Non-Episodic Reinforcement Learning for Robotic Manipulation

Po-Yi Wu, Djengo Cyun-Jyun Fang, Dian Cheng et al.

CRONOS is a simulation benchmark for fine-tuning vision-language-action (VLA) policies with non-episodic scene resets. The authors report that naive non-episodic fine-tuning is ineffective due to drift in object configurations and end-effector poses, but targeted reset mechanisms (end-effector resets, curriculum reset schedules, and heuristic resets of unrecoverable scenes) improve training and generalization.

RoboticsAI summary
2026-11-04 · arXiv:2611.05280

YANAMI: A Large-Scale Multimodal Dataset and Benchmark for Multiturn Stateful Interaction

Shengyu Wang, Tengyue Nong, Junming Liu et al.

YANAMI introduces YANAMI-400K (400K multi-turn multimodal dialogues with 2.0M+ images and 2.6M dialogue turns) and YANAMI-Bench (400 multi-turn dialogues) for multimodal stateful interaction. It evaluates Dialogue State Understanding (DSU) and Visual Realization (VR). The authors report that fine-tuning BAGEL on YANAMI-400K improves image editing, generation, and multimodal stateful interaction, while evaluations on YANAMI-Bench show existing UMMs struggle with long-horizon state maintenance and updates.

AI summary
2026-11-04 · arXiv:2611.05376

From Autonomous Agents to Cognitive Symbionts: Toward AI-Augmented Human Intelligence

Hao Fei

The authors propose AI-augmented human intelligence (AHI) as an “equal endgame” to AGI: measure how much intelligence an AI enables in a particular person via a lifelong “cognitive symbiont.” They define a symbiosis loop (sensing, keeping, understanding, living alongside, acting/thinking with, growing) under a personal runtime and trust, and discuss risks and governance.

AI summary
2026-11-04 · arXiv:2611.05377

DreamX: Video Diffusion Model is a Universal Latent Action Learner

Jisoo Kim, Chaehyun Kim, Junsu Kim et al.

DreamX repurposes a pretrained video diffusion transformer as a universal latent action learner. It appends learnable query tokens to extract compact latent-action tokens (compressing 1,200 frame tokens to 4 tokens per latent frame via a 300× reduction). Trained with future-frame reconstruction, it enables any-to-any video transfer and zero-shot human-to-robot action transfer, and improves compact latent-action policy learning. On LARYBench, it reports state-of-the-art among latent action models.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.00023

FORCE: Factorized Object Relations for Counterfactual Editing in Video World Models

Suraj K. Sood, Yugyung Lee

FORCE is a video world model for counterfactual editing that infers an explicit latent for each ordered object pair from a short observation window without supervision, then holds that relation latent fixed for the whole rollout. Edits directly overwrite the coefficients used by the interaction head to compute forces, enabling removing/changing/transferring/introducing relations. The authors evaluate against simulator counterfactual changes and find worse editing from video than from exact states due to perception precision.

2026-11-03 · arXiv:2611.00044

SparseWAM: Driving World-Action Models are Worth 16 Register Tokens

Yixiang Yang, Lizhong Zhang, Xingyu Zhang et al.

SparseWAM is a driving world-action model that forecasts future scene dynamics using a sparse register space of 16 tokens per frame, instead of dense future videos or long patch-token features. The method uses a modality-decoupled Mixture-of-Transformers with a causal attention mask for imagine-then-act reasoning at deployment and supports multimodal sampling via rectified flow. The authors report 92.6 PDMS on NAVSIM v1, 90.1 EPDMS on NAVSIM v2, and 39.6 on NavHard with 164 ms latency, and release code upon acceptance.

DrivingCode availableAI summary
2026-11-03 · arXiv:2611.00158

Scaling Trusted Autonomy: Towards Generally Capable Verifiable Robots

Benedict Quartey

The authors argue that scaling trusted autonomy requires a neurosymbolic approach that unifies general learned capabilities with explicit symbolic structure for reasoning and verification. They propose three interfaces—LIMP for language-to-plan grounding and inspectable interpretation, REALM for evolving 3D scene state that is updated after predicted manipulation, and PACTS for composing visuomotor skills with predictions of symbolic effects.

RoboticsAI summary
2026-11-03 · arXiv:2611.00177

AuroraWorld: Structured 4D World Modeling for Exo-to-Ego Video Generation in Robot Manipulation

Hao Xue, Peng-Fei Zhang, Ying Cheng et al.

AuroraWorld is a structured 4D world model for exo-to-ego video generation in robot manipulation. It infers a structured 4D embodied representation from sparse exocentric observations using spatial (3D geometric tokens), temporal (motion-stabilized keyframes), and motion (inferred end-effector trajectories) components, then uses these as priors for diffusion-based egocentric video generation. The authors evaluate on visual generation and downstream embodied policy learning in simulation and real robotic environments.

2026-11-03 · arXiv:2611.00241

IsoAlign: Isoform-Aware DNA-Protein Alignment via Joint-Embedding Predictive Architectures

Niousha Sadjadi, Arman Seyed-Ahmadi, Tony Tu et al.

IsoAlign is a multimodal genomic pretraining framework for isoform-aware DNA-protein representations using two JEPA architectures: DP-JEPA (predict protein representations from DNA via contrastive alignment) and IM-JEPA (mask protein isoforms and predict them using DNA anchors and visible isoform context). The authors evaluate DNA representations on tasks including fitness, variant effect prediction, gene expression, and TF binding.

JEPAAI summary
2026-11-03 · arXiv:2611.00406

TAP-Out: Tracking Any Point in 360{\deg} via Outpainting

Finlay G. C. Hudson, James A. D. Gardner, William A. P. Smith

TAP-Out is an offline pipeline for tracking points in 360° even after they leave the camera’s field of view. From a narrow-FoV video, it outpaints a complete 360° equirectangular video (Argus), segments/locates the target (SAM2), generates object-centered perspective crops, and runs standard in-frame point tracking on the crops to recover 360° direction tracks.

2026-11-03 · arXiv:2611.00427

Spoofing the Latent State: Imperceptible Goal-Image Attacks Derail Planning in LeWorldModel

Kunal Bhosikar, Nicole Meng, Charu Sharma et al.

The authors study test-time robustness of LeWorldModel (LeWM), a JEPA-style latent planner that minimizes squared Euclidean distance between predicted and goal embeddings. They find block pose dominates latent distance. A white-box ℓ∞ perturbation (ε=4/255) of a single input frame can steer the encoder to an attacker-chosen block pose in 73–95% of scenes while leaving encoded agent state largely unchanged, severely reducing planning success across benchmarks.

Latent DynamicsAI summary
2026-11-03 · arXiv:2611.00455

TacMind: Benchmarking and Modeling Tactile Interaction States for Robotic Manipulation

Jiaxin Shi, Zibin Wu, Yichi Zhang et al.

TacMind: Benchmarking and Modeling Tactile Interaction States for Robotic Manipulation proposes TacMind-Bench (128 simulation tasks) to evaluate how tactile information helps precise contact, recognition under visual ambiguity, and dynamic interaction under 12 perturbations. The authors introduce TacMind-0: an action-conditioned predictive latent world model (Tac-LeWM) that learns streaming tactile states and refines actions using predictive tactile reflection.

RoboticsAI summary
2026-11-03 · arXiv:2611.00513

WorldSketch: Sketching World Models for Zero-Shot Mobile Manipulation

Jiaming Wang

WorldSketch is an agentic robotics paradigm where a VLM builds a minimal kinematic world model (a “world sketch”) using perception tools and visual checks, then a geometric planner optimizes contact and whole-body motion on the sketch. Execution feedback verifies outcomes, updates uncertain joint hypotheses, and triggers sketch revisions. The authors report zero-shot success of 91% (RoboCasa), 97% (LIBERO-PRO), and 76% (R2R-CE), with fewer VLM input tokens.

RoboticsAI summary
2026-11-03 · arXiv:2611.00596

X-World 1.6: Building Controllable Virtual Cities with World Models

XPeng, :, Boyang Wang et al.

X-World 1.6 advances X-World 1.0 toward a more reliable, controllable driving simulator. The authors add finer-grained traffic semantics control (traffic lights and road markings) and improved spatial consistency across cameras. They introduce absolute static scene control (point clouds, 3D Gaussian Splatting, or reference images), an enhanced long-horizon post-training pipeline (Connected Self Forcing), and a dedicated evaluation/iteration framework.

DrivingAI summary
2026-11-03 · arXiv:2611.00612

Roz: A Taste Test for World Model Latents

Kenneth Ge, Dhruv Sheth, Kunvar Thaman

Roz evaluates whether latent dynamics world models encode object-relevant physical properties beyond image similarity. The authors introduce Roz, a training-free, model-agnostic benchmark using two evaluation methods (contrastive and continuous nearest-neighbor) with 10 contrastive tasks, 3 continuous tasks, and 1734 scenarios. They show that popular models do well on endpoint/image robustness but fail on tasks where history-revealed properties (e.g., friction, mass, deformation) matter.

Latent DynamicsAI summary