Datasets Latest coverage

538 Hours of 'Extremely Messy' Human Data, Real Robot Takes Less Than an Hour: Hong Kong University Teaches Robots to Unpack Packages, Use Hot Glue Guns, and Make Cold Drinks

The MMLab team at the University of Hong Kong proposes the EgoWild2Dex method, which uses uncurated real human first-person operation data to train robots, enabling complex dual-arm dexterous tasks such as unpacking packages, using hot glue guns to fix figurines, and making cold drinks, with an average success rate of 96.7%. The method transfers natural human operation data to robotic execution space through GeoFormer perspective alignment and progressive training.

RoboSpeak: Capital Flows into the Cold Robot Track of Underwater Robotics

The article explores the growing interest in underwater robotics as a specialized niche within embodied AI, contrasting it with the crowded general-purpose robotics market. It highlights the economic and operational advantages of deploying underwater robots in high-risk environments, such as oil and gas inspection and maintenance. Companies like Weilang and Shandong Future Robotics are leading the charge, demonstrating the potential for deep-sea robotics to become a valuable and scalable industry. The piece also outlines the technical challenges, including perception, reliability, and energy, and discusses the commercial strategies needed to succeed in this niche market.

I Found the Largest 'Underground Organization' in the Robotics Circle, Right at HIT

The article reveals Harbin Institute of Technology's (HIT) profound impact on the robotics industry, showcasing the dense industrial network formed by its alumni and faculty entrepreneurs. HIT-affiliated companies span multiple critical segments of the robotics supply chain, from perception layer to actuation and application. The article also analyzes HIT's unique advantages in talent cultivation, industry-academia-research integration, and cross-university collaboration, as well as its central role in embodied intelligence and the robotics industry.

Joint Module Manufacturer Secures 100,000-Unit Purchase Order, Exceeding 100 Million Yuan; 10,000-Unit Sewing Robot Manufacturing Base Officially Operational | Market Observation

The article outlines multiple key developments in the robotics industry, including a 100,000-unit planetary joint module purchase order worth over 100 million yuan, the official launch of a 10,000-unit sewing robot manufacturing base, and the establishment of a new 'Embodied Intelligence International Application Training Field' in Hong Kong. It also covers strategic collaborations, product launches, and industry reports, reflecting the ongoing growth and innovation in robotics and embodied AI.

Harbin Institute of Technology and the Robotics Industry: Four Decades from Lab to Industrial Network

The article traces Harbin Institute of Technology’s role in robotics, from early arc-welding research to companies working on sensors, robot hardware, control systems and industrial applications. It describes how alumni, research teams and regional partnerships connect university work with the robotics supply chain.

538 Hours of 'Extremely Messy' Human Data, Real Robot Takes Less Than an Hour: Hong Kong University Teaches Robots to Unpack Packages, Use Hot Glue Guns, and Make Cold Drinks

The MMLab team at the University of Hong Kong proposes EgoWild2Dex, which uses 538.9 hours of uncurated, real-world first-person human operation data to train robots, enabling complex dual-arm dexterous tasks such as unpacking packages, using a hot glue gun to fix figurines, and making cold drinks, with an average success rate of 96.7%. The method aligns human and robot perspectives using GeoFormer and adopts a progressive human-robot joint training paradigm, transforming natural human operation data into executable strategies for robots.

Jilin University Team Releases Y-Hand M2, the Globally Highest Degree-of-Freedom Biomimetic Dexterous Hand

The Y-Hand M2 biomimetic dexterous hand, developed by Professor Ren Lei's team at Jilin University, is currently the globally highest-degree-of-freedom biomimetic hand with 38 degrees of freedom. It achieves high flexibility and compliance through biomimetic tensile-compression robotic technology, successfully completing 33 types of human-like dexterous operations with a success rate of 96%. The product has already entered mass production and is applied in multiple fields such as intelligent manufacturing and healthcare.

Standing on the Shoulders of Midea: Welling Robotics Components Unveils Full-Stack Joint Solutions at the Industrial Expo

Welling Robotics Components, a subsidiary of Midea, has launched a new product lineup at the Industrial Expo, including force-controlled joint modules, high-performance frameless torque motors, and specialized harmonic and RV reducers. These products are designed to support the commercialization of humanoid robots and industrial automation, with performance metrics and production capacity highlighted as key differentiators.

9999 RMB Opens the 'Dexterous Era' of Robotics: Huiling Tech Launches 20-Degree-of-Freedom Dexterous Hand

Huiling Tech launched the eHand-20 dexterous hand at the 2026 China International Industrial Fair, priced at 9999 RMB. This 20-degree-of-freedom dexterous hand combines high degrees of freedom with low cost through a full direct-drive control solution, modular design, and scalable manufacturing capabilities. The eHand-20 not only enhances fine manipulation capabilities but also advances embodied intelligence through a hand-brain closed-loop system, marking a significant turning point in the transition of dexterous hands from laboratory research to industrial application.

SLS PA12 Robot Tooling: A Practical Qualification Guide

This guide outlines the essential steps for qualifying SLS PA12-printed robot tooling for production environments. It covers defining the component's function, assessing load paths, designing for manufacturability, and implementing a staged qualification process. The tutorial emphasizes the importance of engineering validation, interface design, and testing under real-world conditions to ensure reliability and safety.

FineART: Fine-Grained Annotated Robotic Trajectory Dataset

FineART is a bimanual robotic manipulation dataset with dense, human-written subtask annotations, offering a structured approach to training hierarchical policies. It includes 40,543 episodes, 151 tasks, and 533,913 subtask labels, collected on a Trossen Stationary AI robot. The dataset supports training models to predict step-by-step language for subtasks, with tools for accessing and processing the data.

2026 China International Industrial Robot Exhibition: New Industrial Embodied Intelligence Products to Be Unveiled

The 2026 China International Industrial Robot Exhibition will take place from October 12 to 16 at the National Exhibition and Convention Center in Shanghai, with the theme 'Embodied Industry, Intelligent Future.' The event will gather over 400 domestic and international robotics companies, showcasing more than 300 new products, including industrial robots, collaborative robots, humanoid robots, mobile robots, and core components. The article details the latest products and solutions from companies such as Estun, ABB, KUKA, Huihuan, and XinSong, including embodied intelligent robots, collaborative robots, and welding solutions, demonstrating the wide application and technological breakthroughs of industrial robots in the field of smart manufacturing.

Shangyu's Industrial Robotics Model: Can It Be Replicated?

The article discusses how Shangyu, a district in Zhejiang, has become a model for integrating robotics into industrial settings. By deploying 40 data collection robots in chemical plants, the region has gathered over 150,000 high-precision operation data points, significantly reducing exposure time in hazardous areas. However, the article raises questions about the scalability of such models, the economic viability of training environments, and the challenges of data collection and monetization. It also highlights the broader industry trends, including the role of government initiatives and the need for standardized data practices.

LeRobot@LeRobotHF· 13:34SignalEditorial score8585

LeRobot's new streaming loader is designed for training on large-scale robot datasets, allowing efficient processing of 100 TB+ data without full local video copies. It reads episodes directly from Hugging Face or Buckets, prepares samples on-the-fly, and balances memory and throughput for varied training batches.

TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning

TouchScale is a 500-hour human vision-touch dataset collected with unified sensors and protocols, enabling scaling benefits in perception and robotic manipulation. The dataset improves cross-sensor generalization and robot task success rates without using action labels, demonstrating the potential of large-scale human touch data for embodied AI.

TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning

TouchScale is a 500-hour human visual-tactile dataset collected with unified sensors and synchronization, showing scaling benefits in both perception and robotic manipulation. The dataset enables cross-sensor generalization and improves robot task success rates from 22.5% to 57.5% through mid-training. The study highlights the potential of large-scale visual-tactile data to enhance embodied AI systems.

Teaching Robots to Work: China's Embodied Intelligence Data Collection and Training Ecosystem

The article describes the process of collecting and annotating real-world data for training robots, including first-person and exoskeleton-based data capture methods. It highlights the role of industrial bases and platforms in creating high-quality datasets for robotic training and deployment, as well as the challenges and opportunities in scaling this data-driven approach.

Vision-Language-Action in Robotics: A Survey of Datasets, Benchmarks, and Data Engines

This paper presents a systematic analysis of Vision-Language-Action (VLA) research, focusing on datasets, benchmarks, and data engines. It identifies key challenges in data fidelity, scalability, and evaluation, and proposes a structured approach to address these issues. The authors release an open-source repository to support the community.

LeRobot@LeRobotHF· 16:04SignalEditorial score6565

The author explores the LeRobot community dataset, using embeddings and FiftyOne to analyze 497 episodes across 50 robot types. They identify near-duplicates, filter sessions, and reveal cross-embodiment patterns like cloth folding. A curated subset is available for further exploration.

Fine-Grained Annotated Robotic Trajectory Dataset and Vision-Language-Action Policy

This paper presents FineART, a densely annotated bimanual manipulation dataset with 40,543 episodes and 533,913 subtask labels, totaling 1,718 hours. It also introduces FineART-VLA, a vision-language-action policy that improves task success rates through dense subtask training. The dataset and model are open-sourced, offering a new benchmark for long-horizon bimanual manipulation tasks.