At CoRL, NVIDIA unveiled the GR00T 2 robot learning model, built on NVIDIA Cosmos, which can learn from human demonstrations and transfer skills across different robot embodiments. Of the 50 tasks it evaluated, 40 were successfully completed at least once with no task demonstrations; after one demonstration per task, the average success rate was 71%, and after five demonstrations it reached 95%.
At CoRL 2026, NVIDIA launched the open-source world-action model series GR00T 2, comprising two versions, GR00T 2 and GR00T 2 Turbo. The former focuses on generalization and fine manipulation, while the latter has fewer parameters and can be deployed directly on the Unitree G1, with a single inference taking 124 milliseconds on the onboard Jetson Thor.
Why it matters: NVIDIA released the GR00T 2 series at CoRL 2026 and disclosed the scale of three types of pre-training data along with success rates for cross-embodiment transfer and few-shot adaptation, which can be used to assess the current capability boundaries of world-action models.
CASBOT has partnered with AMD to complete on-device large-model inference, multimodal interaction, and Agent scheduling validation on the CASBOT 02 by externally connecting the AMD Ryzen AI Embedded X100 series, enabling local multilingual multi-turn dialogue and semantic invocation of motion skills. The next-generation humanoid robot plans to upgrade the X100 series from an external connection to a built-in chest "brain" computing unit, and to adopt the P100 series as the "cerebellum" to handle real-time tasks including bipedal locomotion, dual-arm coordination, and joint control. The two series share a unified architecture and software stack and can be flexibly configured as single-processor or multi-processor solutions according to modular requirements.
NVIDIA's GEAR team released its next-generation world action model GR00T 2 at the CoRL conference. It can simultaneously predict future frames and robot actions. The model, data, simulation environment, and collection hardware will be open-sourced progressively, and 100,000 hours of human data have already been released first.
Why it matters: NVIDIA GR00T 2 pretrains on a mix of human video, UMI, and teleoperation data, and reports quantitative success rates for cross-embodiment transfer, which can be used to assess the data strategy for robot foundation models.
In early November, former Baidu Group Vice President and Baidu Research Vice President Wu Tian officially joined Sany Group as Group Vice President, Dean of the Yun AI Institute, and concurrently Dean of the Artificial Intelligence Research Institute. Previously, in October, former Chery Automobile Vice President Xie Baojun had also joined the Sany system. The article states that Sany designated 2025 as its "first year of intelligence," seeking to make AI a general capability across all roles. Vice President Li Hongwei said Sany will develop into an AI-native robotics company in the future, but currently unmanned operations still rely mainly on rule-based algorithms rather than being data-driven.
NVIDIA launched the open world action model GR00T 2 and the open-source robot learning framework Isaac Lab 3.0 at CoRL. GR00T 2 was trained on 100,000 hours of NVIDIA open-source first-person data; with only one demonstration per task, it reached an average success rate of 71% across 50 tasks, rising to 95% with five demonstrations, and without task-specific demonstrations it completed at least once each of 40 of the 50 tasks.
Why it matters: GR00T 2's few-shot success rates and Isaac Lab 3.0's training throughput gains provide the current baseline for embodied models and the simulation training stack.
Ken Goldberg's team at UC Berkeley and Spencer Huang, NVIDIA's head of robotics engineering, proposed an "agentic robotics" method that lets AI agents generate, test, and correct robot control programs in simulation without additional data collection.
Why it matters: The report describes a method for using multi-agent systems to generate, test, and correct robot programs in simulation, and presents real-world data from warehouse sorting.
At the MSIT-NVIDIA AI Startup Partnership Forum, NVIDIA announced an investment in Korean AI interconnect company Point2 and plans to make additional investments in Motif Technologies, RLWRLD, and Upstage. RLWRLD develops robot foundation models for humanoid and industrial robots, supporting five-fingered dexterous manipulation such as precisely inserting parts, tightening nuts, and sorting parts from a moving conveyor belt. These investments cover the full AI stack, from infrastructure and models to robotics applications.
NaiveAI has open-sourced Naive-N0.5-Flash. Users only need to write a one-page program.md task brief, and an AI agent connected to Claude Code can then complete the entire workflow of data cleaning, training, evaluation, and troubleshooting.
At the World Internet Conference in Wuzhen, Moxin Technology released MoWorld 2.0, a new-generation general-purpose 4D world model that integrates more than ten physical elements, including gravity, elasticity, buoyancy, lighting, soft bodies, fluids, rigid bodies, deformable bodies, and thermal dynamics, into a single model to more accurately simulate object motion, collisions, and environmental changes.
American humanoid robotics company Agility Robotics officially listed on Nasdaq on November 5, 2026, under the ticker AGLT, going public through a merger with Churchill Capital Corp XI. The transaction and related financing are expected to bring in about $465 million in raised funds, including a PIPE financing of about $201 million led by Foxconn.
Why it matters: Agility's listing makes its orders, deployment duration, and financing structure public, allowing comparison against the real progress of humanoid robot commercialization.
Looking forward to this:
NVIDIA has released VFlash, a general-purpose decoding framework for fast visual planning that replaces token-by-token coordinate generation with parallel decoding of complete visual primitives, allowing it to predict robot grasp positions and motion trajectories in one pass. At batch size 1 on H100, single-generation speed improved by about 25x; across four MolmoSpaces tasks, the world action model averaged a 70.2% success rate versus a 62.6% baseline. Jim Fan shared that it achieves up to 12x faster action decoding, 5.1x overall robot inference speedup, and up to 61% lower 3D trajectory error.
Why it matters: NVIDIA's new decoding framework changes visual planning from token-by-token generation to parallel prediction, and provides speedup and success-rate comparisons across multiple robot and autonomous-driving benchmarks.
According to an October 31 report by The Information, Meta and Chinese robotics company Fourier jointly built the humanoid robot prototype Max in about three months, and in September of this year began development of a second-generation prototype, Max 2. Max is about 1.5 meters tall and weighs about 48 kilograms, is equipped with an NVIDIA Jetson Thor chip, and features a display that can simulate facial expressions, built-in speakers, and five-fingered dexterous hands for grasping small objects. A Meta spokesperson said the collaboration focused on internal research and experimentation, and there are no plans to sell these robots.
One year after NVIDIA open sourced its Aerial wireless software, global developer adoption of its AI Aerial and Sionna software has grown 10x, downloads of its open source 5G/6G tools have surpassed 1 million, and more than 8,000 research and academic institutions are using them.
Shanghai Kepler has built a closed loop from simulation training to real-machine validation for industrial humanoid robots based on NVIDIA Isaac Lab, Isaac GR00T, and Cosmos. The company develops its own planetary roller screws, linear and rotary actuators, joint modules, and robot bodies, introduces real dynamics parameters into Isaac Lab for Sim2Real iteration, and validates them in scenarios such as workshop assembly and CNC machine loading and unloading. Kepler is also working with a team from Zhejiang University's College of Artificial Intelligence to develop an embodied intelligence foundation model for long-horizon tasks.
Intel will hold the Embodied Intelligence & Robotics Frontier Trends and Solutions Exchange in Shanghai on November 10, limited to 100 attendees, with a live demo based on Intel Core Ultra X7 358H (Panther Lake).
In Q3 2026, Unitree's revenue was 665 million yuan, up 69.93% year on year, while non-recurring net profit was 105 million yuan, down 17.42% year on year. R&D spending was 127 million yuan, up 249.58% year on year, showing revenue growth without profit growth.
In his speech at the 2026 Mobile Network Conference, Qualcomm President and CEO Amon said AI is simultaneously reshaping devices and networks, with agents set to run on phones, PCs, cars, and IoT devices, while the Snapdragon platform supports both on-device AI execution and cloud invocation. Qualcomm's collaboration with China Mobile has expanded from smartphones to AI glasses and robots; the Jiutian on-device AI model has landed on the Snapdragon platform, and the Lingxi 2.0 personal AI agent is among the first agents to run on the fifth-generation Snapdragon 8 Elite mobile platform.
RoboColosseum is a benchmark for testing policies on real robots, covering 150+ scenarios and hundreds of actual robots, with all rollouts (including failures) published on Hugging Face.
Nanjing University has established the Institute of Biological Intelligence, focusing on four directions: the foundations of biological intelligence and large models, bio-inspired perception and computing, biological computing and organoid intelligence, and biological agents and embodied systems. The team has already used a fully connected model of the fruit fly brain, with about 100,000 neuron connections, on a drone, achieving obstacle avoidance in the laboratory without traditional massive-data training. The soon-to-be-released "Guangyu Core 2.0" outputs tokens directly at the sensor end for connection to large language models, and is being adapted and incubated for small and micro special robots and commercial space.
A developer used a 20-second driving record with 7 cameras and a roof-mounted LiDAR from NVIDIA's public PhysicalAI AV dataset to build an end-to-end runnable real-to-sim pipeline.
The article notes that humanoid robots are moving from demonstrations to mass production, but the manipulation success rate drops sharply from 89.4% in simulation environments to 12% in real home scenarios, and reliability must be supplemented by a testing system. Integrated joint modules account for about 40% of the BOM cost of the complete machine, and GB/T 43200-2023 has provided a testing framework in four categories: motor performance, control performance, positioning accuracy, and mechanical performance. NI uses a cRIO joint test bench and a hardware-in-the-loop simulation solution based on Isaac Sim, PXI, and RDMA to try to insert an HIL stage between pure simulation and real machines, narrowing the gap in migration from simulation to reality.
Meta partnered with Chinese robotics company Fourier to complete a humanoid robot prototype named Max in just three months, and has already started development of a second-generation Max 2. Max is about 1.5 meters tall and weighs about 48 kilograms, is equipped with Nvidia's Jetson Thor chip, is customized based on Fourier's GR-series hardware, and needs to recognize voices from 5 meters away and gestures from 3 meters away. Meta says the collaboration is for internal research and experimentation only, with no sales plans, but it is recruiting for supplier development and product design roles in Shanghai and Shenzhen.
At the 2026 Tencent Global Digital Ecosystem Summit, Tencent Cloud showcased AI Infra deployments in embodied intelligence. Manifold Space used a cluster of 512 domestic chips to train stably, improving training performance nearly 3x in about two weeks. Machine Science built a unified data lake in Nanjing with COS, enabling cross-region direct reads of the same data between Beijing and Nanjing and saving tens of millions of yuan that would have gone to building high-performance storage in both locations. Zeroth used COS to handle PB-scale data across multiple campuses in Shanghai, North America, and Europe, with MetaInsight helping engineers quickly extract high-value segments.
Intel will hold an Embodied Intelligence and Robotics Solutions Exchange in Shanghai on November 10, showcasing a multi-channel GMSL real-time perception solution and a 358H embodied intelligence complete system based on the Intel Core Ultra X7 358H (Panther Lake) platform.
According to foreign media reports on October 31, Meta partnered with Chinese robotics company Fourier Intelligence to develop a humanoid robot prototype named Max in only about three months, and a second-generation product is already in development. Max is about 1.5 meters tall and weighs 48 kilograms, is equipped with an NVIDIA Jetson Thor chip, has five-fingered dexterous hands and a display that can simulate human facial expressions, and needs to recognize voices and gestures and maintain eye contact with people.
According to The Information, Meta partnered with Chinese company Fourier to build a humanoid robot prototype named Max. Development took three months; it stands close to 5 feet (about 1.5 meters) tall, weighs about 106 pounds (about 48.1 kilograms), and is powered by an Nvidia Thor chip.
Bloomberg reporter Mark Gurman says Apple plans to update the entry-level iPad, iPad Air, and iPad Pro in the first half of 2027, with the iPad Air moving to an OLED display and high-end iPad Pro models possibly adding a vapor chamber to improve cooling.
Meta has partnered with Fourier Intelligence to develop a humanoid robot prototype named Max, internally defined as the physical embodiment of its personal AI agent Muse. The prototype was completed in three months, and second-generation development has already begun. Max is about 1.5 meters tall and weighs 48 kilograms, and carries an NVIDIA Jetson Thor chip to support on-device running of vision-language-action models. It builds on Fourier's GR-series chassis and motion control capabilities, with newly customized head, screen, camera, and hand and foot modules.
The "Embodied Intelligence & Robotics Frontier Trends and Intel Solutions Exchange," hosted by Intel and jointly convened by the Shenzhen Robotics Association, will be held on November 13, 2026, at Shenzhen Weixin Software Park, with limited registration open to algorithm, hardware, complete-machine, and supply-chain partners.
The Information reports that Meta has jointly developed a humanoid robot prototype codenamed Max with Fourier Intelligence and has started development of a second-generation Max 2. The Max prototype is about 1.5 meters tall and weighs 48 kilograms, is equipped with Nvidia's Thor chip, and is positioned as the physical embodiment of Meta's Muse large model, with the goal of entering home scenarios. The project took only three months from initiation to the first prototype, which was shipped back to Silicon Valley at the end of August this year and completed a dynamic demonstration at Meta's internal robotics summit.
RoboColosseum is a distributed evaluation framework that evaluates general-purpose robot policies over a network of locally operated robots in multiple locations, using unified policy serving and standardized protocols.
Why it matters: This distributed evaluation framework used 5,699 paired trials to compare VLA policies with frontier agents, revealing inconsistent rankings across embodiments and a ranking reversal after fine-tuning.
Based on a survey of more than 20 mainstream quadruped robot projects on GitHub, the article maps the open-source quadruped technology landscape across five schools: kinematic control, full-stack engineering, optimization algorithms, reinforcement learning, and hardware platforms.
The DeepSeek Harness team unveiled its Super-agent design, which uses a dual-layer architecture, event-driven operation, asynchronous multi-channel interaction, separation of storage and compute, and an external watchdog to let the agent maintain context across sessions and advance multiple tasks in parallel. Internal instances are already involved in reproducing and fixing issues and in code review.
According to The Information, Meta partnered with Chinese company Fourier Intelligence to build a humanoid robot prototype codenamed Max in about three months, and a second version is already in development. The prototype weighs about 48 kg and stands about 1.5 meters tall, runs on NVIDIA Jetson Thor, and features a display that can simulate facial expressions, speakers, and five-fingered hands; Meta says it has no sales plans and positions the robot as internal research.
The robots were told there would be candy. 🍬🎃 Happy Halloween!
Using Unitree H1, Isaac Sim 2024.1, IsaacLab, and PPO as its stack, the article walks through a complete sim-to-real pipeline: converting URDF to USD, scene setup, domain randomization configuration, PPO training, ONNX export, and ROS 2 deployment on the real robot.
At the 2026 Tencent Global Digital Ecosystem Summit, Tencent disclosed that WorkBuddy has completed 9 billion task requests cumulatively since launching in March this year; its Enterprise Edition has been deployed in over 50 industries, covering 50% of China's Fortune 500 companies and 60% of China's Top 100 companies.
The team recreated a real self-service tea station as a physically interactive 3D environment in Isaac Sim, placed a Unitree G1 in it, and used a script to execute the five-step tea-making motion. The robot's motions were not extracted from human video; the entire run took place in simulation with physics enabled. The G1 still shows imperfect contact and slipping when interacting with objects, and there is currently no learned policy or automatic trajectory generation. The team is exploring whether such environments can serve as a source of synthetic robot demonstration data, including for robots with different embodiments.