Skip to content

#Google DeepMind

Today 3 records
Today2026-10-04Sun
  1. Robotics — Chinese web discovery84

    Is It a Valuation Bubble or Real Demand? Investigating the Truth Behind Embodied Intelligence Data Collection Centers

    The article explores the rapid expansion of embodied intelligence data collection centers in China, driven by both policy and market forces. It also reveals issues with some companies creating false revenue through a 'equipment sale—data repurchase' model. The article further analyzes the technical challenges of centralized data collection and the industry's shift toward distributed data collection models.

    Editorial context：The article investigates the rapid expansion of 'body intelligence' data collection centers in China, driven by policy and market forces. It highlights concerns over inflated valuations and potential unsustainable revenue models, particularly through the 'equipment sale—data repurchase' cycle. The piece also explores the technical challenges of centralized data collection and the industry's shift toward distributed data collection models.

2026-10-03Sat
  2. High-tech Robotics — WeChat85

    A 2-Year-Old Embodied Intelligence Company Founded by Fei-Fei Li Sells for 55 Billion Yuan; Midea Group Plans to Acquire Welding Subsidiary for 6.64 Billion Yuan | Major Capital Events of the Week

    The article reports on major capital events in the recent robotics and embodied intelligence sectors, including AMD's acquisition of WorldLabs, a company co-founded by Fei-Fei Li, for approximately 55 billion yuan, Midea Group's plan to acquire a welding subsidiary for 6.64 billion yuan, and several robotics companies such as Benmo Technology, Huanchuang Technology, and Juxi Intelligent completing financing or going public. Additionally, multiple embodied intelligence products and solutions have been launched, covering humanoid robots, service robots, flexible electronic skin, tactile perception systems, and more, showcasing the industry's rapid progress in technology, capital, and application scenarios.

2026-10-02Fri
  2. arXiv Robotics — research abstracts62

    DeepJEPA: Scaling World Models from Within

    This research introduces DeepJEPA, a weight-tied joint-embedding predictive world model that scales computation internally by focusing on decision-critical transitions. It outperforms or matches fixed-depth planners in five visual-control settings while using fewer updates per transition, demonstrating that strategic allocation of internal computation improves planning without requiring uniform state decodability.

2026-09-30Wed
  1. QbitAI — Robotics85

    GPT-6 Astra Connects to Unitree G1, Cleans the Kitchen!

    The Stanford team's HomeBody project demonstrates how GPT-6 Astra controls the Unitree G1 robot to complete kitchen cleaning tasks through a three-step process: spatial exploration, digital twin simulation, and task execution. The system uses pre-trained models and a skill library to perform complex operations in new environments without additional training.

    Editorial context：The article explains how GPT-6 Astra controls the Unitree G1 robot to perform kitchen tasks using a three-step process: spatial exploration, simulation mapping, and task execution. It highlights the integration of perception, planning, and control systems, and discusses the role of pre-trained models and skill-based execution in enabling generalization across new environments.

  2. arXiv Robotics — research abstracts62

    DORA: Divergence-Oriented Data-Relay Algorithm for Partially Connected Robot Teams

    This research presents DORA, a divergence-oriented data-relay algorithm that enhances communication in partially connected UAV teams by prioritizing the value of information to the team. The algorithm quantifies mission-relevant divergence between a robot's information state and its teammates' knowledge, improving MRT resolution delay by up to 74.8% over traditional methods.

  4. arXiv Robotics — research abstracts63

    SkillWeaver: Agentic Exploration over Neural Interaction Skills for Scalable Robot Data Generation

    SkillWeaver is a framework that autonomously generates robot experience by exploring over Neural Interaction Skills (NIS), which are reusable, parameterized, closed-loop policies. It enables the agent to discover successful long-horizon behaviors through verifier-guided tree search, improving generalization across novel objects, tasks, and environments.

  6. Sergey Levine65

    Sergey Levine participated in a 36-hour hackathon where an unreleased checkpoint of π0.7 was used to control an excavator, achieving second place. The policy, trained with about an hour of embodied data, enabled continuous dirt movement with a single prompt. The team also added a Nerf gun for safety, showcasing the practical application of AI in robotics.

    Quoted postsimon@nautsimon_

    Over the weekend, I put an unreleased checkpoint of π0.7 inside an excavator and placed 2nd at the @ActorLabs + @physical_int 36hr hackathon. With about an hour of embodied data (3 cameras and joystick values, no joint angles), the policy could continuously move dirt with a single prompt. This was so straightforward to get working (pi is cooking) we spent the rest of the hackathon mounting a Nerf minigun on the roof. We added a people searching/tracking + firing feature to prevent people from getting too close to the dangerous machinery (ai safety feature). Here’s how we built it + more details (data collection, hardware hax, sim2real, video of me getting shot)🧵

2026-09-29Tue
  1. Andrew Kang62

    Andrew Kang from X Robotics announces the release of Dyna-2.1, the first Physical Agent capable of reliable super long-horizon whole-body autonomy. The system combines new semi-humanoid hardware with an agentic framework to handle ultra-long real-world workflows, demonstrated through an hour-long laundry room task.

    Quoted postDyna Robotics@DynaRobotics

    We are releasing Dyna-2.1, the first Physical Agent that achieves reliable super long-horizon whole-body autonomy. It combines our brand-new semi-humanoid hardware with an agentic system built around Dyna-2 to handle ultra-long real-world workflows. Here is an uncut footage of Dyna-2.1 completing an entire hour-long laundry room workflow, just like a human does.

  2. Remi Cadene25

    Wow many robots :)

    Quoted postNicolas Keller@Nicolas_Keller

    The Era of Evals is coming to AI robotics. Today we’re launching Reality Check: our leaderboard & first public robot manipulation benchmark. 14,400 real-world rollouts. Four models. Multiple tasks & data regimes. Object placements. Confidence intervals. FR3 Duo stations. 1/10

2026-09-28Mon
  1. Fei-Fei Li15

    **Summary:** Fei-Fei Li, co-founder of World Labs, announced the company's partnership with AMD, emphasizing its focus on spatial intelligence and robotics simulation. World Labs, founded in 2024, aims to solve critical challenges in AI by developing models that reason over the physical world. The company has released Atlas, an omni-model architecture that predicts new camera views from 2D images, outperforming state-of-the-art results. The acquisition of SceniX is part of World Labs' strategy to build industry-leading robotics simulation capabilities.

  2. Yuke Zhu22

    **Summary:** Yuke Zhu presents key insights from an ICRA'26 keynote, emphasizing the rapid progress in building human-like machines, the scalability of human data for training robot foundation models, and the role of world models in enabling policy generalization. The talk highlights NVIDIA's work on projects like GR00T, EgoScale, SONIC, and WAM, including a demo of a model learning to assemble the YCB airplane with zero teleoperation data. The discussion underscores ongoing advancements in embodied AI and robotics.

  3. Eric Jang15

    If you're a robotics researcher working with GPT 6 Astra to do interesting things with robotic control or agentic real2sim, and would like access to open source models (Kimi K3, Qwen 3.8 Flash Next) as a comparison, please DM me! I'd love to provide you with free access to tokens to help you benchmark the agentic capabilities of these open models, and would be super eager to see what you can do with open source models in this domain.

2026-09-23Wed
  1. Robohub85

    Robotics Roadmaps from Around the World: Spotlight of the Month - China

    China is pursuing a strategic roadmap through its Five-Year Plans to dominate the global robotics industry, focusing on embodied AI and humanoid robotics. The 13th and 14th FYPs laid the foundation for industrial automation, while the 15th FYP emphasizes embodied intelligence, with state support for R&D and mass deployment of humanoids. Despite achievements in market share and production scale, challenges remain in precision hardware and semiconductor self-sufficiency, with risks of overcapacity and market volatility.

    Editorial context：This analysis provides a detailed overview of China's strategic approach to robotics through its Five-Year Plans, highlighting the shift from industrial automation to embodied AI and the challenges in achieving self-sufficiency in critical components. It emphasizes the role of state coordination, domestic manufacturing capabilities, and the risks associated with the rapid expansion of humanoid机器人.

2026-09-22Tue
  1. Pulkit Agrawal15

    Policy learning is not just about learning :)... a lot depends on the physical manifestation underlying what is learned. Uncover things that matter to make robots work!

    Quoted postYounghyo Park@younghyo_park

    Robot-learning folks: have you ever found that a seemingly minor system choice beneath the policy—controller tuning, hardware selection, or inference infrastructure—made a much bigger difference to your robot’s performance than an architecture or algorithm tweak? How do you usually isolate and report that effect? That’s the question behind our Workshop on Everything Beneath the Policy at #CoRL2026.

  2. Sergey Levine23

    How do we run RL with real-time chunking (RTC)? In this work we figured out how to use a small RL policy with a large robot foundation model, where the RL policy observed more recent images (due to faster inference) and steers the policy toward better behaviors! A fun collaboration with Siemens, led by Brian Zhu, Momen Khalil, Emanuele Poggi from Siemens and @ehharrison4 from Berkeley, with lots of amazing contributors!

    Quoted postE Harrison@ehharrison4

    Asynchronous VLA inference reduces inference delay, but breaks the Markovian assumption necessary for RL fine-tuning. How can we enable RL fine-tuning of VLAs with async inference? We introduce ARLI: Asynchronous RL with Intermediate Information! https://async-rl-intermediate-information.github.io/ (1/n)

2026-09-21Mon
2026-04-21Tue
2025-07-15Tue