This tutorial explores NVIDIA's Cosmos framework from a practical Colab angle, honestly assessing the hardware needed for real Cosmos 3 checkpoints. It builds and trains a compact omnimodal Mixture-of-Transformers world model using the framework's real structure, CLI surface, and input schema. Using synthetic physical-world data and autoregressive rollout, it shows how the model predicts future latent states across text, vision, and action modalities.

Nextdocs.io is an AI-powered platform that creates beautiful documents and slides from a single prompt. It offers templates, multi-version generation, AI editing, brand kits, and various export formats. Trusted by thousands of users across 100+ countries.

Generative AI companies face high costs, copyright lawsuits, and competition from free open-source models, leading to doubts about their commercial value. If sustainable profits prove elusive, the technology could become a 'toxic asset,' but that might be okay for users.

Skill Retriever is an open-source semantic skill retrieval plugin for Hermes Agent. It pre-filters over 1,200 skills organized in a 10,000-category capability taxonomy to the top-5 most relevant per query, with zero additional API cost. It overcomes limitations of pure semantic retrieval by using LLM-navigated capability hierarchy to surface non-obvious but functionally relevant skills.

The founder of Cartlytics, a patent-backed commerce attribution SaaS with identity graph and AI customer scoring, is selling on Flippa with no reserve. The startup integrates 11 ad platforms and runs on Cloudflare infrastructure, positioning itself as YC 2026 ready.

Attackers exploited a prompt injection vulnerability in GitHub Agentic Workflows to trick the AI agent into reading and publicly posting private repository contents via a crafted GitHub Issue. This demonstrates a fundamental trust boundary issue in agentic AI systems.

At an AI conference in Beijing, experts called for US-China cooperation on AI safety to address cybersecurity and systemic risks from advanced models. The proliferation of open-weight models exacerbates risks, but international collaboration could mitigate systemic threats.

Australian dock workers are demanding a 28-hour work week with no pay cut as DP World expands AI and automation. The union warns over 60% of dock and maintenance jobs are threatened and calls for a 'social dividend' from the technology.

Multi-fisheye camera calibration is challenging as rig size and field of view increase. This paper reveals through failure-oriented analysis that intrinsic initialization is the dominant failure factor and proposes CO-Calib, a plug-in framework that improves success rate from 68.1% to 99.3%.

Researchers developed a social robot that autonomously delivers cognitive stimulation therapy (CST) to people with dementia in their homes. A one-week study with nine participants found high adherence rates (nearly 50% of sessions completed), surpassing typical caregiver-led CST. Family members played a key role in initiating sessions and occasionally joining activities, enhancing engagement. The work demonstrates the feasibility of socially assistive robots for scalable in-home dementia care.

A novel method named IMR is proposed for multi-agent motion prediction. It uses a mode-world weighted regression loss to mitigate mode collapse while improving world ranking and top-1 confidence. An iterative decoder recurrently and segmentally generates trajectories, enhancing prediction accuracy. The method achieves first place on the Argoverse 2 multi-agent motion forecasting benchmark.

A learning-based architecture transforms onboard vehicle sensor data into a virtual velocity sensor for IMU drift correction without additional hardware. It uses a Mamba-based state space model and evidential deep learning for uncertainty quantification, achieving localization accuracy within 10% of a dedicated external sensor at 40 Hz on edge hardware.

This study proposes a neural network-based transfer learning framework for modeling the dynamics of soft, fin-actuated underwater robots. Using an autoencoder-based domain adaptation approach, a model trained on a larger robot is adapted to a smaller one without labeled data, achieving accurate body-frame velocity estimation. The work demonstrates efficient cross-robot dynamics transfer among morphologically similar platforms.

This paper presents PRML2, a hybrid framework combining Kalman filtering and machine learning. By end-to-end training through a differentiable Kalman filter, PRML2 achieves physics-regularized learning for vehicle pose estimation from onboard sensors. It demonstrates superior localization accuracy and real-time capability on a public dataset, and introduces a new dataset for low-friction conditions. Accepted at IROS 2026.

This paper presents a control-aware optimal trajectory planning framework that incorporates manipulator dynamics and actuator effort. A midpoint linearization strategy improves accuracy for large motions. Simulations on a UR5 robot show consistent reductions in tracking error, corrective torque, and execution cost compared to classical kinematic planners, demonstrating that kinematic smoothness does not guarantee dynamic efficiency.

This paper introduces the HIOcc hierarchical indoor occupancy benchmark and the GEM-Occ Gaussian Evidence Memory framework, which treats local visual geometry predictions as transient evidence and fuses them into a persistent hierarchical memory, enabling semantic occupancy mapping from single-view to building-level, and showing improvements in stability and scalability across multiple datasets.

MECo-WAM injects action-relevant 4D geometric priors into video-action representations to improve robotic manipulation performance without increasing inference cost. It uses multi-expert co-training, decayed 4D read-mask attention, and action-aware temporal geometric distillation. Achieves 98.2% on LIBERO and 92.6% on RoboTwin 2.0.

This paper proposes QAACF for foot-mounted AHRS PDR, using Markley quaternion averaging to fuse sensor data and adaptive weighting to achieve low RMSE and computational cost.

FaceMesh2HPO is a framework for classifying facial phenotypic descriptors aligned with the Human Phenotype Ontology (HPO) to support clinical diagnosis. Using annotations from 124 clinicians across 10 disorders (107 HPO terms) combined with non-syndromic controls, 3D facial meshes (478 landmarks) were generated from 2D images and a hierarchical PointNet-based pipeline with cascading classification and feature elimination was trained. The best models achieved AUROCs between ~0.55 and ~0.89, with higher performance at parent nodes than leaf terms. External validation showed variable generalizability across disorders. Results demonstrate that hierarchical modeling of 3D facial geometry enables interpretable, ontology-linked phenotype classification, though performance on rare leaf terms remains limited. Improved data diversity and feature selection strategies are needed.

This paper presents a training-free method that leverages pretrained generative image models and vision-language models to extract semantic parts from multi-view images of 3D objects and abstract them into superquadric primitives. The approach contains no learned parameters, is category-agnostic and orientation-invariant, achieving the lowest Chamfer distance on HumanPrim and Toys4K datasets with an average of 5–9 primitives per object. The study shows that the current accuracy bottleneck is part segmentation, not primitive fitting.

Researchers propose MuCoDi, a pretraining framework that distills frozen tile embeddings from multiple pathology foundation models into compact edge-oriented encoders using contrastive distillation. RepViT-based MuCoEdge students retain near-teacher performance while reducing model size by orders of magnitude, achieving up to 605-fold speedup on a Raspberry Pi 5.

This paper introduces a rendering-aware Bayesian 3DGS framework that tracks Gaussian geometry with a Normal-Inverse-Wishart posterior, enabling native uncertainty estimation and active view selection. In fixed-budget active-view tasks, the method improves PSNR by +0.453 dB and LPIPS by -0.0146 over baselines, reduces 95% coverage error by approximately 17x, and achieves training cost roughly one-third of a deep ensemble.

This research uncovers naturally occurring statistical signals in vision data that can be exploited like backdoor triggers without malicious insertion. By analyzing ImageNet, the authors identify patterns strongly linked to specific labels, use statistical controls to remove spurious correlations, and demonstrate that these signals directly and predictably alter model predictions. These statistical adversaries are more targeted than generic corruptions and transfer across architectures, suggesting vulnerabilities stem from dataset structure rather than model idiosyncrasies. The study recommends treating spurious structure as a latent attack surface.

Light-Omni is a multimodal agent framework for reflexive video understanding using dual contextual states, achieving 12.1x speedup and 2.6x GPU memory efficiency over M3-Agent, and serving as a memory system for MLLMs.

Ground3D-LMM is a unified model that integrates point cloud and RGB image inputs to enable 3D spatial conversations with explicit point grounding and metric measurements. It introduces the 3D Grounded Measurement task and a large-scale dataset with 2.5M QA pairs, setting a strong baseline for grounded, metric-aware 3D dialogue.

A gaze estimation method using only one camera and one light source is proposed, introducing a virtual light source and virtual glint. Performance is acceptable but degraded compared to two-light-source systems.

This paper presents a task-driven framework linking synthetic fog generation, image restoration, detection, and tracking. Results show fog degrades performance mainly through missed detections; fog-inclusive training offers the most consistent robustness gains.

This paper introduces CanvasAgent, a tool-augmented multimodal agent that learns to orchestrate heterogeneous visual tools through multi-turn interaction for complex image creation and editing. The authors also present CanvasCraft, a large-scale dataset with 140K executable trajectories and 10K RL task specifications. The agent is trained with supervised fine-tuning and then optimized with GRPO using a hybrid reward combining outcome- and process-level signals. Experiments demonstrate effectiveness in both final image quality and trajectory behavior.

NAVER LABS re-implements its IWSLT 2025 instruction-following pipeline for the IWSLT 2026 Shared Task (constrained condition, short audio track), adapting to mandated components: SeamlessM4T-v2-large as speech encoder and Qwen3-4B-Instruct as LLM backbone. The three-stage approach (projector alignment, text-only LoRA pre-training, multimodal merging) is preserved. Additionally, 100k synthetic instruction-following examples across ten speech-centric task types (10k per task) are constructed. The primary model achieves COMET 0.781 on EN-ZH speech translation and BERTScore-F1 0.346 on English SQA on the MCIF benchmark.

BaFCo is a benchmark dataset for Bangla form comprehension, comprising 200 multi-page complex Bangladeshi government forms from diverse sectors. It features 26 fine-grained and 5 coarse entity types. Evaluations of latest MLLMs show limitations in localizing granular form entities.