🔬
XDOF scales data collection for physical AI training
AI 實驗室正轉向 XDOF 等專業服務商，以解決物理 AI 訓練中數據收集繁瑣且勞力密集的難題。這標誌著機器人領域對高品質訓練數據的需求日益增長。
Search
Tag: #data-collection31 results
TinyFish launches Bigset for automated web dataset generation
TinyFish has released Bigset, an open-source multi-agent system designed to convert plain-language text prompts into live, self-refreshing datasets from the web. This tool simplifies the data collection pipeline for developers by automating the extraction process.
Human Archive leverages India's gig economy for robotics data
Human Archive, founded by Berkeley and Stanford researchers, is utilizing India's gig economy to collect physical-world training data. Workers wear camera-equipped gear to capture data essential for advancing AI and robotics models.
Recording Household Chores to Train Humanoid Robots
The article explores the process of turning daily household tasks into training data for future humanoid robots. It highlights the personal and privacy implications of contributing human behavior data to AI development.
Embodied AI Craves Human Motion Data
AI recreates Buffett's shareholder speech from past media. Platforms like DoorDash and Micro1 pay gig workers globally to record chores for robot training. Robot firms spend $100M+ yearly on data amid billions in investments and China data center boom.
Meta Staff Rebel Against AI Surveillance PCs
Meta is installing surveillance software on employees' work computers to capture keystrokes for AI training. Staff are unhappy, highlighting the irony given Meta's user-tracking business model. The move is reportedly driven by Zuckerberg to fuel AI development.
🔬
Crowdsourced Data Fuels Humanoids
Apps pay crypto for users to film everyday tasks like microwaving food for robot training. Another platform lets users remotely control robotic arms in China for puzzles. This gathers vast real-world data for humanoid AI.
GenRobot.AI's Precision Hand Glove for AI
GenRobot.AI unveiled the Gen DAS Dex, a lightweight exoskeleton glove for capturing human hand motion. It achieves 0.02-degree precision and includes tactile sensing. The device targets data gaps in training dexterous embodied AI.
Tsinghua Spinout Xingyi Secures Seed Funding
Xingyi Tech, incubated from Tsinghua, raised millions in seed funding led by Shumu Ventures for Ego-centric data collection hardware rivaling Nvidia's EgoScale. The startup focuses on high-precision, multimodal wearables fusing vision, haptics, and pose for embodied AI training data. Founder Song Zhiheng brings experience from Zhiyuan Robotics and Magna Robotics.
Japan Launches 50-Robot Physical AI Data Center
Yamazen announced the 'Physical AI Robot Data Collection Center' to gather training data for AI controlling industrial humanoid robots. The facility will operate 50 humanoid robots. It collaborates with J-HRTI to advance humanoid robot societal implementation.