Independent global robotics news from Article module 102.
Each record is classified by both destination market and product layer, including complete robots and core components.
Mobile manipulation combines navigation, torso motion, dual-arm coordination, head control, and dexterous hands. A CoRL 2026 study explores whether demonstrations from hu...
As robot-learning datasets grow to millions of hours, the central challenge is no longer simply collecting more data. This article compares first-person video, real-robot...
Archon Robotics has released Anchor 1, a full-body control model for humanoid robots. Its demonstrations focus on precise whole-body coordination, including writing arbit...
Robotics companies face a distinct go-to-market challenge: translating technical capability into repeatable customer value. This article discusses product positioning, de...
VFlash is designed to let a vision-language model generate an entire robot action trajectory in one pass. The approach targets the latency and planning bottlenecks that a...
A research team used Opus-5.5 to modify the weights of π0.5 and explored whether automated research systems can turn model-editing ideas into reproducible experiments. Th...
A Northwestern, CMU, UNC Chapel Hill, and Amazon team presents WOVEN, a benchmark and training framework for visual world modeling. Across controlled experiments, world-m...
A METR review of Anthropic says the pace of progress on its benchmark capability index has roughly doubled, with about 80% of the improvement attributed to AI-assisted re...
Reports indicate that OpenAI may soon release the GPT-6.2 Astra and Sol models, alongside a cybersecurity-focused model. A published study says the new system identified ...
A Fortune 500 customer is using embodied-AI systems in warehouse operations, creating a demanding test of perception, planning, manipulation, and continual improvement in...
A U.S. humanoid-robot company’s shares surged nearly 80% on their first trading day, drawing renewed attention to the commercial expectations surrounding humanoid robotic...
It's not just a war of color. The most honest robot data of 2026 is on a one-head shipping list. Recently, the Robotics Workshop learned that an American Realbotix deskto...
At the CeMAT site, in order to demonstrate the ability to collaborate in depth with multiple products, PUDU DUDU robots converted the stand into a real in-store operation...
DiT, U-Net, and latent spaces are three important concepts behind modern generative models. This explainer compares their roles in image and video generation and clarifie...
Daimon Robotics is building tactile intelligence around both data collection and model design. The article explains how tactile sensors, datasets, and world models can be...
Yesterday, she sent a farewell letter and jokingly told netizens, "Wait until you get to AGI", and everyone thought she was going to start a business. As a result! Today!...
A founder born in 1998 and educated at Tsinghua has built a world-model startup that has raised about RMB 700 million. The article looks at the company’s technical direct...
However, language is essentially an orderly Token sequence, and vision is a continuous high-dimensional distribution, both of which have a natural gap in information stru...
With the length of the task chain, the computing costs of each step of the model, the long-text capability, affect the end-user experience and cost, the importance of the...
Taobao’s MNN team has open-sourced a 0.3B model optimized for on-device inference. The model reportedly runs offline at up to 350 tokens/s on an iPhone 18 Pro, illustrati...