Agile Robots has obtained early access to NVIDIA Cosmos 3, with early simulation tests showing strong capabilities. The company will integrate Cosmos 3 for customers, offering new options for simulating, testing, training, and deploying physical AI, and is exploring the Cosmos-based NVIDIA Video Augmentation Skill for automated data curation, generation, and evaluation.
Genesis AI has released and open-sourced its simulation platform Genesis World 1.0, along with the GPU-accelerated computing project Quadrants and the path-tracing rendering engine Nyx, claiming that one hour of reality can become 100 days in simulation.
Why it matters: The original text explains how simulation turns the physical-time bottleneck into a compute problem, and provides links to three open-source repositories along with performance improvement figures.
Genesis AI has released and open-sourced Genesis World 1.0, the second product in its full-stack suite, alongside the Quadrants compute library and the Nyx rendering engine.
Why it matters: The article provides quantitative comparisons of startup and runtime performance, making it easier to judge how far it shifts real-world validation into a compute problem.
Embodied intelligence company OriginFlow announced it has completed multiple financing rounds—an angel round, a strategic round, and a Pre-A1 round—with cumulative funding exceeding RMB 500 million. Puhua Capital participated in two consecutive rounds, and other investors include BlueRun Ventures, Oasis Capital, Monolith Management, and Yuanhe PuHua.
Hugging Face has released the LeRobot Humanoid full-stack project, with a pair of humanoid robot legs priced at $2,500, built from 3D-printed parts and off-the-shelf components. The release includes a bill of materials, 3D printing files, wiring documentation, assembly instructions, and software tools for calibrating and controlling the robot both on the physical body and in simulation.
Why it matters: The original text explains the hardware cost, the open materials, and the simulation-to-real design loop, which can be used to assess the impact of a low-cost humanoid platform on research workflows.
Noble Machines published an article introducing its three-layer robot learning stack for industrial autonomy, along with on-site test results from GTC 2026: over four days, the same model simultaneously drove multiple Moby robots to complete two autonomous tasks at the NVIDIA booth and its own booth.
Pro Universe Robot released its high-precision fusion data collection solution AcCI and the Dabai robot intelligent loading and unloading module at its Global Partner Conference on May 12, 2026, and introduced an all-domain symbiosis ecosystem strategy. AcCI offers sub-millimeter precision, all-modality collection, and an end-to-end closed loop, and supports both master-slave and UMI dual collection modes; the Dabai robot has completed POC validation across multiple industries and has operated stably for over a year on production lines including the Dreame smart factory.
This article is an interview with Dr. Jianlan Luo about his new work LWD (Learning While Deploying), which enables general-purpose robot policies to continuously perform online reinforcement learning from their own experience during real-world deployment. LWD combines offline RL pretraining with online RL fine-tuning to unify offline and online optimization objectives; the value function is trained separately with a VLM and uses both positive and negative samples (roughly over 60% positive and over 30% negative), and deployment data flows back into the training pipeline in about 40 seconds on average.
Genesis AI has released GENE-26.5, calling it the first "robot brain" on the path toward human-level capability. It is a full-stack approach combining a robot-native foundation model, a 1:1 humanoid robotic hand, non-invasive data-collection gloves that capture motion, force, and touch, and a simulator. The model is trained jointly on language, vision, proprioception, touch, and action, and was tested on a set of self-designed tasks; the demonstration was fully autonomous, at 1x speed, using a single model with the same weights.
LimX Dynamics has officially open-sourced FluxVLA Engine, a standardized VLA engineering foundation for embodied intelligence that covers the full pipeline from data and training to simulation evaluation and real-robot deployment.
Why it matters: LimX Dynamics has officially open-sourced its VLA engineering foundation, which lets readers understand how unified configuration and real-robot deployment workflows lower the barrier to research and development.
Dema Technology, together with Maiqing Intelligence, released OmniEgo, a first-person full pipeline for embodied intelligence data collection. The companies claim that a single person can stably collect data for more than 8 hours per day on average, that a team of 1,000 people can produce more than 200,000 hours of valid data per month, that collection efficiency is improved by more than 10 times, and that overall costs are reduced by 80%. The pipeline covers collection, cleaning, self-labeling, and quality inspection. It relies on the self-developed OmniEngine to achieve millimeter-level hand pose perception, with self-labeling coverage exceeding 85% and overall processing efficiency improved by 90%. Dema Technology is also opening this collection capability globally and announced a three-year goal of producing 10 million hours of high-quality embodied data per year.
Genrobot has open-sourced the Phase 3 UMI dataset, containing 13K+ hours of dual-arm data that is ready to use and supports sub-1ms frame-level synchronization and high-fidelity trajectory reconstruction. The data was collected by 3K+ contributors in 10K+ real homes, covering diverse everyday manipulation skills. Download at https://huggingface.co/datasets/genrobot2025/10Kh-RealOmin-OpenData.
Why it matters: The data scale and real-home collection background help readers judge whether this kind of UMI data can be used to train their own manipulation tasks.
On April 10, Lingchu Intelligence released Psi-R2, pretrained on human data, and the Action-Conditioned World Model Psi-W0, and open-sourced the first 1,000 hours of a full-modality human hand manipulation dataset, with total data reserves reaching 100,000 hours.
Why it matters: The original text provides the division of labor between the two new models and the open-source data entry point, so readers can assess the feasibility of using human data for robot training.
Lingchu Intelligence announced the completion of a new funding round, with Guotou Xiandao and Jingxi Ruiling as investors; the funds will be used for core technology R&D and commercialization. On April 10, the company released its next-generation embodied AI models Psi-R2 and Psi-W0 and open-sourced a 1,000-hour full-modality human manipulation dataset, stating that Psi-R2 ranked No. 1 globally on the Molmospace benchmark; it will next advance the large-scale application of its dexterous manipulation solutions in logistics, industrial, and new retail scenarios.
Lingchu Intelligence has officially released its self-developed complete robot ψ-SynRobot and simultaneously started large-scale mass production. The model targets logistics, retail, and industrial scenarios and uses an integrated "operation execution + data collection" architecture; each completed task simultaneously accumulates visual, tactile, and motion-trajectory data. It can run continuously 24/7 and suits scenarios such as warehouse sorting, retail guidance, and industrial assembly. The release gives Lingchu Intelligence a complete product matrix from an "embodied intelligence brain" to a complete-machine execution platform, covering forms such as wheeled robots and exoskeleton gloves.
Personalized manufacturing brand UNICUS (formerly Fangzai Photo Studio, Shenzhen Qianzhi Technology Co., Ltd.) announced the completion of a new round of financing of several million USD, led by Linear Capital, with JiuKun Venture Capital and Skyline Capital participating, along with personal participation from the founding team and founders of leading companies in the Maker sector; the funds will be used for building-block generation large model training, AI Agent development, and overseas market expansion.
Damon Technology and UBTech signed a strategic cooperation agreement in Shanghai to jointly develop humanoid robots for scenarios including warehouse sorting, material handling, and production line collaboration, and plan to establish a joint venture. UBTech's Walker S2 industrial humanoid robot began mass production in 2025 and completed its first batch of large-scale deliveries; its full-size embodied intelligent humanoid robot business recorded 820 million yuan in sales revenue and 1,079 units sold for the full year 2025. The cooperation also covers core component development and embodied data collection coordination.
General-purpose embodied intelligent robotics company Simple AI announced the completion of its third funding round within six months, raising hundreds of millions of yuan, led by Linear Capital and Puhua Capital, with existing shareholders Junshan Capital, Shunwei Capital, and BV Baidu Ventures continuing to increase their investment; the funds will be focused on R&D of embodied robot brains and bodies.
Lingchu Intelligence released two models, Psi-R2 and Psi-W0, which it calls the industry's first embodied intelligence models pretrained on human data at the 100,000-hour scale.
Why it matters: The original article, written by the team itself, breaks down the dual-model architecture and data governance conclusions, offering a reusable path for human data utilization along with experimental evidence.
Sanctuary AI demonstrated zero-shot in-hand manipulation in the real world with its proprietary hydraulic hand: a policy trained entirely in simulation continuously reoriented a lettered cube to a target pose ten times in a row without dropping it. The operation relied solely on the fingertips with no palm support, requiring simultaneous stable grasping and approach to the pose target. The original text states that the hand has a relatively high number of active degrees of freedom enabling finger abduction, and that hydraulic actuation and compact hydraulic valves offer a path toward industrial-grade humanoid dexterous manipulation.
RealMan, together with Pinpinxiang, presented a "robot tea master" tea-brewing service at the 2026 Zhongguancun Forum. An operator remotely teleoperated it from the Beijing Humanoid Robot Data Training Center, and a RealBOT wheeled humanoid robot replicated in real time on site the Fuding white tea steps of warming the cup, adding tea leaves, pouring water, and decanting the brew. The service relies on RealMan's Global Labor Network (GLN) to turn human expert operating skills into services that can be scheduled across time and space; it is a remote teleoperation demonstration, not autonomous tea brewing by the robot.
Teams including Berkeley AUTOLAB propose CaP-X, an open framework for systematically studying Code-as-Policy agents in robot manipulation, centered on the interactive environment CaP-Gym and CaP-Bench, which evaluates 12 frontier models.
Why it matters: The original text presents a systematic evaluation framework for Code-as-Policy agents and the finding that test-time scaling can mitigate abstraction dependence, offering reference value for research on coding-agent manipulation control.
At NVIDIA GTC 2026, Sharpa demonstrated its dexterous hand Wave in a real-world demo; its skills were trained on the NVIDIA Isaac Lab simulation platform and stably transferred to the real system, enabling fine manipulation such as in-hand operation.
Dexterity published an article introducing system evaluation and production measurement for the Foresight packing agent. On the planning side, it was tested on four real warehouse box-type distributions, with density of 0.65–0.77 and stability of 99.7–100%, and distribution-free order statistics were used to give 95% confidence intervals; the article also found that randomly shuffling the box sequence causes local variance to collapse by about 1.4 times, making evaluation overly optimistic.
Why it matters: The original article fully discloses the evaluation method for the packing agent, including distribution-free confidence intervals and the evaluation bias caused by random shuffling; the method can be transferred to similar systems.
1X Technologies disclosed that the AI development process for its humanoid robot NEO is based on NVIDIA's robotics platform, covering the open-source physical AI agent tools released this week at GTC Taipei.
Noble Machines and Schaeffler have partnered to bring high-fidelity digital twins of high-performance actuators into the open-source physics engine Newton, for developing whole-body control policies before hardware deployment. The models can capture manufacturing variations and predict aging and energy consumption, with energy consumption predictions up to 10 times higher than idealized models, and feedforward compensation can reduce tracking error by up to 68%. Robots can also use control signals as identification signals to enable online self-diagnosis and compensate in real time for wear or environmental drift.
FieldAI presents its collaboration with the NVIDIA Omniverse ecosystem, using NuRec to reconstruct multimodal data collected during robot operations—including vision, depth, and LiDAR—into high-fidelity, interactive 3D scenes without a separate scanning step.
Did you know that the Maya Biosphere Reserve is one of Guatemala's last remaining forests? 🌳 Protecting it means safeguarding biodiversity, ecosystems, and local livelihoods. ACOFOP is using R-evolution's #GreenCubes to advance sustainable forest management. https://hxgn.biz/4cErBmo
Sunday Robotics has closed an oversubscribed $165 million Series B at a valuation of $1.15 billion, led by Coatue, with Thomas Laffont joining the board. The company will stop giving demos and focus on real-world deployment of its robot Memo, with a beta program opening to users this fall. Its engineering team has grown 3x, research 4x, and data operations 5x, and the company says it has a full stack from its Skill Capture Glove and in-house hardware to manufacturing, along with data advantages.
Why it matters: The company announced the funding and disclosed the beta program timeline and the scale of team expansion, giving readers a basis for understanding the pace of its shift to real-world deployment.
The Toyota Research Institute (TRI) published a tutorial article introducing a statistical A/B testing method based on success/failure labels for robot policy evaluation.
Why it matters: The TRI authors systematically explain statistical pitfalls and reusable tools in robot policy A/B testing, with runnable Python examples.
Realman Intelligent announced it has completed nearly 500 million yuan in funding, jointly and strategically invested by industrial parties from multiple listed companies. The funds will go toward product R&D, construction of the AUTRON production line super factory, and global ecosystem development. The article states that its integrated joint modules exceeded an annual production capacity of 100,000 units in 2025, that its robotic arms passed CR L3 certification and achieved a mean time between failures of 50,000 hours, and that the GLN remote operation network and the Beijing Humanoid Robot Data Training Center have accumulated tens of millions of real-machine trajectory segments. The company serves more than 8,000 enterprise users worldwide.
Why it matters: The company describes its own progress on funding, production capacity, certification, and its remote operation network, giving readers a basis to understand how its hardware-plus-data strategy is actually advancing.
Hugging Face LeRobot releases v0.4.4, adding support for OpenArm, OpenArm Mini, and G1 teleoperation, as well as damiao and RobStride CAN motor drivers, and expanding the camera API with CLI support for configuring camera backends.
At the Beijing TV Spring Festival Gala, RealMan's RealBOT wheeled folding robot used remote teleoperation to open a refrigerator, take out soy sauce, pour soy sauce, and stir dumpling filling; it was the only company among the participating recording enterprises to complete on-site work via remote teleoperation. Its integrated joint module has an MTBF of 50,000 hours, opening the refrigerator door requires an instantaneous torque of about 14 kilograms, pouring soy sauce relies on visual guidance and high-frequency posture correction to achieve zero spills, and mixing the filling is controlled by a remote "isomorphic arm" that records data to feed back into training models for the remote operation network (GLN).
Linear Capital reposted LatePost's interview with Tashi Intelligence founder Chen Yilun, who laid out a technical path that differs from the mainstream: he rejects the VLA architecture that grows an action head on top of a VLM, and instead develops its own AWE (AI World Engine), aiming to represent physical quantities such as time, space, and force; it has also abandoned teleoperation data collection in favor of a self-developed collection device combining a wearable glove and a first-person-view camera.
Why it matters: The interview systematically presents technical judgments and a data approach that differ from the mainstream VLA path, giving readers a basis for understanding the specific grounds of the divergence in embodied AI approaches.
Skild AI proposes overcoming the robot data bottleneck by watching human videos, arguing that teleoperation is limited by environment diversity and real-time collection scale and cannot reach the data volume required by foundation models. Its approach is cross-embodiment learning: with less than 1 hour of robot data for fine-tuning, the model can learn new skills from video demonstrations alone, mapping human "grasping" onto the execution actions of different robot embodiments.
Realman is exhibiting at CES 2026 for the third consecutive year, presenting three foundational capabilities: hardware, data, and a remote operation network. It also completed a transoceanic real-time teleoperation demonstration from Beijing to Las Vegas, remotely controlling a RealBOT wheeled folding robot to perform tasks such as delivering items and handing over fruit.
Mentee Robotics demonstrated a MenteeBot learning to replace the battery in another MenteeBot without teleoperation, solely by observing a human demonstration. The learning process has three Real2Sim2Real stages: the robot records a human operation video from its own viewpoint, a foundation model reconstructs the task in simulation and then trains at scale using self-play and reinforcement learning, and finally its proprietary Sim2Real technology transfers the skill to autonomous execution on the real robot.
Zheng Suibing, founder and CEO of Realman, said on December 19 at the 2025 Corporate Social Responsibility Forum that the company is using a "remote workforce network" to let operators direct robots in real time to perform real tasks such as folding towels, utility tunnel inspections, and power control, while accumulating scenario data to feed embodied intelligence. Realman's integrated joint modules have annual production capacity exceeding 100,000 units, its robotic arms have passed CR L3 certification, and their mean time between failures has reached 50,000 hours. The company also co-founded the GPR open-source alliance in February 2023 and released a mobile composite robot development platform in 2024.
On December 19, Tashi Intelligence, founded just 10 months ago, held a livestreamed full-stack technology debut. The world's first robot to achieve autonomous embroidery was officially unveiled, completing thread pulling, needle lowering, and piercing motions with sub-millimeter precision to embroider the brand logo, while also demonstrating the autonomous implementation of flexible fine manipulation in wiring harness assembly.
Great to see @BitRobotNetwork using LeKiwi in its new project TeleArms! TeleArms has officially entered private beta (SN/06 and SN/07). Users can remotely control robotic arms to teach robots to grasp, move, and place objects, and earn rewards. Early access spots are extremely limited; joining Discord, reposting, and tagging friends can improve waiting-list ranking.