Hunan Viewtrix Robotics Co., Ltd. exhibited a dual-machine collaborative welding workstation, a humanoid welding robot, an intelligent residual-material scanning system, an intelligent grinding workstation, and the KUNWU industrial simulation platform at the 2026 China Marine Economy Expo in Shenzhen.
A team led by Dawei Zhou at Virginia Tech and researchers at Meta propose ReGDiff in a NeurIPS 2026 paper to generate metamaterial voxel structures that satisfy geometric constraints. The method has two steps: first, an RAS mechanism organizes the autoencoder latent space into a form where good and bad structures are clearly separated and good structures are evenly distributed; then, short-range repulsion (SRR) pushes latent variables away from the near neighbors of known samples during sampling, and it takes effect only at inference time.
QuantumBit reports that embodied data company Maxinsights has delivered over 2 million hours of egocentric data to clients including Google DeepMind, Figure AI, Dyna Robotics, and Genesis AI, with ARR near $100 million for several consecutive months.
The authors' team successfully ran headless Isaac Lab training on H100 GPUs for humanoid robot reinforcement learning, validating it by training a walking policy for the Unitree G1.
NVIDIA has expanded the Cosmos platform with new training tools, simulation checkpoints and Cosmos Data Factory blueprints for training, evaluating and deploying world models.
Why it matters: The original lists Cosmos's new training tools, simulation checkpoints and data blueprints, and gives specific data on users such as Porsche and ANYbotics, which can be used to assess changes in the world model development workflow.
NVIDIA announced several implementations of open models and physical intelligence in European retail: Shopify used NVIDIA Megatron Bridge and NeMo RL to post-train Nemotron 3.5 Lightning, cutting serving costs by 90% and delivering 7x faster responses on internal benchmarks.
Origami Robotics has released the S1 skill-capture glove, aimed at high-quality dexterous manipulation data collection. It records 20 finger joint angles via encoders and synchronously captures fingertip and palm pressure and wrist-camera video, all aligned to the same timeline, so it can keep measuring even when the hand is fully occluded and the vision pipeline fails. The original text states that its fingertip accuracy has been verified to reach the millimeter level.
TrackDeform3D is an affordable and autonomous framework for collecting 3D datasets of deformable objects using only RGB-D cameras. It identifies 3D keypoints and robustly tracks their trajectories with motion consistency constraints, producing temporally smooth and geometrically consistent data.
The paper proposes the Collective Experience Learning (CEL) paradigm, in which deployed robots pool their own execution experience to update a shared policy that is then returned to the fleet for use. It is implemented through failure-guided experience retrieval fine-tuning (FERT), which uses failure contexts to retrieve relevant segments from successful trajectories and prioritizes their recorded actions for supervised updates, without additional expert demonstrations or corrective actions; a supporting fault-tolerant asynchronous training system (FATS) coordinates robot execution, cloud inference, and policy optimization.
Why it matters: The paper proposes replacing expert corrections with the deployed robots' own execution experience to iterate on the policy, and reports real-world manipulation results at a scale of 96 robots.
The Lambda-humanoid Team released λ0, a humanoid whole-body loco-manipulation foundation model, along with HumanVerse, a 500-hour open-world human loco-manipulation dataset that pairs first-person video with synchronized body and hand motion.
The research team introduces RoboCap, a 250-gram head-mounted device with six cameras and dual IMUs for in-the-wild egocentric data collection, along with the Grounded API, a device-agnostic suite of 3D algorithms. The report shows how the hardware, calibration, and 3D algorithms work together to achieve state-of-the-art performance on public benchmarks, covering SLAM across diverse settings and rigs, egocentric depth estimation, and hand tracking when adapted to third-party devices.
The paper releases the Hilti-Trimble-Oxford dataset, collected at an active construction site. It contains 30 visual-inertial sequences covering 10 floors over an eight-month span, and captures real-world challenges such as changing lighting, moving workers, fast motion, and repetitive structures. Ground-truth trajectories were collected with a high-precision LiDAR-inertial SLAM system rigidly attached to the 360 camera. The authors also report results from the accompanying open research challenge: 62 teams entered the SLAM track, while only 22 entered floor plan localization, indicating less attention to the latter.
μ0 is a scalable world model based on 3D traces. Rather than predicting dense pixels or directly modeling actions, it forecasts smooth 3D trajectories for salient interaction points such as objects, tools, hands, and contact regions, forming a compact, embodiment-agnostic motion interface.
WARPED is a framework that synthesizes realistic wrist-view observations and actions from human demonstration videos, enabling visuomotor policy training with only monocular RGB data. The system uses vision foundation models to initialize the interactive scene, employs a hand-object interaction pipeline to track the hand and manipulated object and retarget trajectories to a robotic end-effector, and finally uses Gaussian Splatting to synthesize photorealistic wrist-view observations for directly training policies.
This work proposes A2A, a vision-and-language navigation benchmark for agricultural robots, comprising 1,560 episodes across 6 classes of agricultural scenes, using step-by-step instructions and a continuous-action environment. Evaluation found that existing VLN methods perform well on short instructions but poorly on long instructions; to address this, the authors propose an STL module that decomposes complex instructions into lists of simple subtasks and integrates it into the NavGPT backbone, raising SR from 0.33 to 0.47 while NE slightly increases from 2.76 m to 2.91 m. Ablation studies and case analyses are also included.
AgiBot World Colosseo introduces the AgiBot World large-scale manipulation platform, containing over 1 million trajectories covering 217 tasks across 5 deployment scenarios, an order-of-magnitude increase in data scale over existing datasets, and uses a standardized collection pipeline with human-in-the-loop verification.
A study proposes introducing a laser pointer into a vision-based portable pipeline for target disambiguation in multi-object scenes, to support grasp selection and preshaping in upper-limb prostheses. The method uses a mixed reality strategy to reuse existing high-quality segmentation masks to generate training images and retrains a deep convolutional neural network. Experimental results show target disambiguation accuracy improves by up to 9% over existing solutions while computational cost remains unchanged.
This study examines the degradation of world-model predictions when physical conditions fall outside the training distribution, comparing two causal modeling elements: the inputs available to the predictor and how its mechanism computes their effects. On a synthetic benchmark with shortcut features, a frozen dependency mask derived from paired interventions reduced the out-of-distribution RMSE of a fully connected predictor from 1.769 to 0.697, close to the generator-oracle mask's 0.694.
PhysiCalWorld proposes calibrating action-conditioned world models' physics via in-context prompting, without fine-tuning or analytical physics modeling, to adapt to instance-level dynamics.
The study trained 12 policies with the same recipe on a simulated Unitree G1, varying gravity, terrain geometry, and ground contact models (rigid or 3D resistive force theory, 3D RFT), and tested Earth-to-Moon fine-tuning and gravity randomization.
DONIC is the first generalist bimanual dexterous tracker jointly trained across manipulation categories and human motion sources. It is released alongside BiHOI, a multi-view dataset containing 1,575 sequences covering 229 objects, in which reconstructed interactions are retargeted into robot references.
An arXiv paper formalizes zero-shot generalization in physical AI as coverage of the deployment distribution, noting that more samples from the same collector can only slowly shrink the unseen region, while each additional contributor can expand coverage. Based on this, the paper proposes a shared latent space for robot actions that combines multimodal alignment, generative augmentation, and noisy-label learning, and suggests that developers build applications around the standard robot Origmi to provide diverse data. It concludes with a testable hypothesis and evaluation protocol linking real-robot zero-shot success rates to data diversity.
SpatialDex is a framework that learns 3D-grounded bimanual dexterous manipulation from RGB-only human videos through reconstruction-augmented visuomotor pre-training, learning human behavior together with its spatial context. It reconstructs scene geometry and estimates hand motion from each video, aligns the two at a video-level scale using the hand-size prior of 3D hand estimates, and uses the reconstructed geometry as both a policy input and a prediction target.
AirPlayground is a unified benchmark for training and evaluating aerial navigation policies, built on 94 photorealistic 3D Gaussian Splatting (3DGS) environments spanning indoor, urban outdoor, and vegetation settings, with aligned collision geometry.
The paper proposes the PRISM framework, which disentangles depth noise into two modalities—sensing invalidation and measurement inaccuracy—and distills features from 3D visual foundation models as spatial-semantic priors.
The authors propose TIDE (Training In DEployment), a human-in-the-loop online post-training system that uses a provenance-conditioned supervised learning framework to update the policy online from human interventions and policy rollouts during deployment, without learning a critic, reward model, or action scorer.
SMR is a streaming neural motion retargeter learned from semantic-conditioned physics experts. It converts human motion into robot task-space targets, which the physics experts track in simulation, and then uses a strictly causal Transformer to learn their executed trajectories. Each new frame uses only current and past human observations to generate a robot motion reference for a downstream whole-body controller. Experiments show that the semantic-conditioned physics experts outperform direct-optimization retargeting in semantic fidelity and physical validity, and that the resulting streaming retargeter performs better in closed loop under a unified whole-body controller than kinematic and physics-refinement baselines.
SimPT is an automated simulation framework that integrates policy rollouts, corrective data collection, policy training, and standalone evaluation for human-intervention-based vision-language-action (VLA) post-training.
HyperDrone is a quadrotor simulator built for high-performance batch rendering, over 20 times faster than existing state-of-the-art batch quadrotor simulators.
COWBOY is a scalable sim-to-real framework for learning contextual whole-body manipulation. On whole-body dexterous grasping tasks, it transfers zero-shot to hundreds of objects and diverse in-the-wild scenes, including cluttered tabletops, shelves, drawers, and other unstructured environments.
Real2Sem2Real (R2Sem2R) proposes a real-semantic-real pipeline that turns a single monocular RGB human demonstration video into thousands of robot demonstrations covering multiple semantic object instances within the same category. The method tracks object SE(3) poses using DINOv3-distilled Gaussian splats, outperforming existing methods and FoundationPose with estimated metric depth, and maps the human hand to a feasible robot grasp that can be transferred to digital twins.
The paper presents DONIC, a generalist bimanual dexterous tracker jointly trained across manipulation categories and human motion sources, and releases BiHOI, a multi-view dataset containing 1,575 sequences covering 229 objects, emphasizing sustained bimanual interaction and retargeting reconstructed interactions into robot references.
Nvidia CEO Jensen Huang announced in Paris a new round of strategic investments in four French startups, covering health tech, chemical materials, business planning, and robotics. Genesis AI is the only robotics target, and it is negotiating a new funding round at a valuation of about $3 billion, roughly 20 billion yuan.
The Dual Exam benchmark evaluates multimodal agents and world models separately: the agent track compares 10 configurations under fixed environment response mechanisms, covering 1,088 items and 15 reporting categories, while the world model track contains 400 cases across 8 scenarios.
The study examines how closely pretrained LeWorldModel (LeWM) reproduces small action changes, measuring action sensitivity with Gram matrices at three perturbation scales across nearly a thousand states from Reacher and OGBench Cube.
SEER Robotics demonstrated its robotic arm completing the full coffee-making process without human intervention, from picking up the portafilter, receiving grounds, distributing and tamping the grounds, locking the portafilter, extraction, to cleaning. This is the longest multi-step robotic coffee task the company has demonstrated to date.
At NVIDIA GTC Berlin, Agile Robots partnered with Audi to demonstrate using Dassault Systèmes' DELMIA virtual twin to evaluate how mobile humanoid robots perform in production logistics systems before deployment.
A Reddit user built a home RGB + iToF depth setup using a 320×240 epc660-class iToF engineering sample, a fixed Pi camera, and Ouster reference scans. After raising exposure from 500 to 2,000 µs, filtered jitter dropped from 65 to 6 mm and raw jitter from 85 to 28 mm, but straight walls still bow 12–14 cm, and three modulation frequencies report 4.14, 3.92, and 3.97 m for the same wall.
Kinetix AI has launched a first-person human data solution for embodied AI, centered on the lightweight head-worn data capture device KAI Halo. It combines multimodal data collection with edge-cloud AI capabilities to turn real human activity into high-quality training data.
ACE Robotics released ACE-Brain-0.6, a vision-spatial-anchoring-enhanced embodied foundation model focused on Grounding capability. It surpasses GPT-6 Astra on six of eight Grounding benchmarks, with an average score of 74.53 across seven public benchmarks (Astra: 73.68), and scores 94.13 on its self-built real-robot evaluation set RobotGrounding (2,591 questions, 14 task types; Astra: 91.10).