Addressing the problem that world-model predictions may be built on outdated perceptions of the environment, SLAM-MER at CVPR 2026 uses recent keyframe temporal information and a retrievable 3D representation for monocular SLAM, while GaME in the same year continuously updates its 3D Gaussian representation and removes outdated keyframe observations.
StreamSplat at ICLR 2026 achieves purely feedforward, online generation of dynamic 3DGS directly from uncalibrated video streams, at 1,200 times the speed of optimization-based methods. But MoRe at CVPR 2026 actively embeds a BA-like refinement module in its purely feedforward pipeline; removing it raises Sintel ATE from 0.147 to 0.155. BA-Track, an ICCV 2025 Oral, uses a 3D point tracker to decompose camera and object motion, adapting classic BA to dynamic scenes.
Thank you Toru, very insightful! Especially this line: "...don't blindly delegate a piece of work before building the understanding needed to evaluate it."
Silicon Valley 101 invited Pokee.ai founder Zhu Zheqing, Pokebot founder Xu Huazhe, and University of Washington assistant professor Wang Sheng to discuss the impact of GPT-6 Astra's release on three paths: personal agents, robotics, and AI drug discovery.
OpenAI CEO Sam Altman said that OpenAI's internal, unreleased models have reached a level on some mathematical research tasks that existing public models find hard to match. He said AI programming has not eliminated programmers; instead, it keeps developers busier, creates greater value, and may let them earn more, because the world's demand for software far exceeds what was previously imagined. He also expects that in about two years, AI may find the first meaningful candidate treatment for some major disease.
SpaceX's superintelligence business revenue reached $7.56 billion in Q3, about 52% of total revenue. Newly deployed compute pays for itself in under a year at current prices. Anthropic rents all of Colossus 1's compute, and Google has been paying $920 million per month for computing capacity since October.
At the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree Technology founder Wang Xingxing judged that as soon as the end of this year or next year, users may be able to use a single voice command to have a Unitree humanoid robot independently complete household tasks such as tidying a room.
Unitree Technology held the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference in Hangzhou. Founder Wang Xingxing said in a speech that physical AI may see a major breakthrough within one to three years, and said GPT-6 Astra enabled Unitree robots to complete painting and Go-playing experiments under zero-training conditions. He also announced Unitree's in-development Physical AI Robot Self-Evolution V1 system, which lets AI agents write code autonomously, test and train in simulated environments, and validate in real environments, forming a continuous iteration loop; he hopes Unitree robots can enter ordinary households by the end of this year or next year.
On October 22, the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference was held in Hangzhou. In his keynote speech, Unitree's Wang Xingxing judged that for embodied intelligence to pass the ChatGPT moment, the core challenge is not whether the model is smart enough, but the alignment between the input and output of AI models and real robots.
In a keynote speech at the 2026 World Physical AI Conference, Unitree's Wang Xingxing said physical AI is showing signs of a breakthrough, but it may still take 1-2 or 2-3 years. The core challenge of embodied AI is aligning AI model inputs and outputs with real robots; large models can plan navigation but have not solved problems after physical contact. Commercial services will be deployed faster than home and industrial applications.
Using a real-world test report on SenseTime's Astra as its starting point, the article notes that Astra is outstanding at the planning layer (100% on LIBERO-Long++, 81.9% on RoboDojo), but real-machine operation exposes an architectural break between cloud inference's second-level latency and millisecond-level force-feedback closed loops; stacking bowls took an average of about 131 minutes and cost about 19 US dollars.
DeepSeek is reportedly raising about 100 billion yuan (approximately $15 billion) in a new funding round, with Tencent and CATL as the largest investors; the funds will go toward model training, computing infrastructure, and talent recruitment. Tesla's Q3 GAAP net income was $787 million, down 43% year over year, with 486,532 vehicles delivered. Tencent customer service said WeChat has been adapted for Apple's foldable phone iPhone Duo, and WeChat Input Method has also been updated to version 4.0.0 with full adaptation.
A Reddit user started a discussion on what unmanned aerial manipulators (UAM) might achieve or solve once research, actuators, control, and applications become mature and widespread. The poster noted that these platforms can not only perceive the environment but also change it, yet they have not yet been widely adopted, and invited the community to share ideas about future application scenarios.
Yang Dongsheng, senior vice president of BYD, said at the World Intelligent Connected Vehicles Conference that L3-level conditional autonomous driving is expected to land in 2028, and higher levels may land another 1–2 years after that. Wang Xingxing, founder of Unitree Robotics, said at the 2026 World Physical AI Conference that physical AI is showing signs of being "broken through," with a breakthrough possibly emerging in the next 1–2 or 2–3 years, and that the core challenge of embodied intelligence is aligning AI model inputs and outputs with real robots.
DeepSeek reportedly completed about RMB 100 billion (USD 15 billion) in financing, one of the largest financing deals in China's history, with CATL and Tencent among the largest investors in this round. About 80% of common apps have not completed foldable-screen adaptation for the iPhone Duo; unadapted apps such as WeChat and QQ can only be displayed centered or scaled proportionally, and Tencent customer service said it is actively advancing adaptation and will release updates as soon as possible.
In an exclusive interview with Caixin Time, Xiong Youjun, CEO of the Beijing Humanoid Robot Innovation Center, said the humanoid robot industry has no bubble, only bubble companies, and that the share of prototype robots actually doing real work remains limited. He disclosed that about 80% of Tiangong's data training comes from virtual simulation and 20% from real-world scenarios, that the Pelican-Unify unified representation embodied world model can improve execution accuracy but has high computing power requirements, and that the next funding round is already being planned.
At the 2026 World Physical AI Conference, Unitree founder Wang Xingxing said physical AI is already showing early signs of being "broken through," with a breakthrough possibly arriving in the next 1-2 or 2-3 years. He believes the core challenge in embodied intelligence reaching its "ChatGPT moment" is aligning AI model inputs and outputs with real robots, and that commercial service scenarios will see deployment sooner than home and industrial ones. In internal tests, he said, GPT-6 deployed on a robot could clean paintbrushes on its own and continue painting without being specifically programmed to do so.
In a CES 2025 online interview, Musk said the Tesla Optimus humanoid robot's hand has 22 degrees of freedom, with a target of producing several thousand units this year, first testing them in Tesla factories, and plans to raise output to 50,000 to 100,000 units next year. He said he is confident Tesla's autonomous driving will be safer than human drivers in about three months, that three Neuralink implant patients are doing well, and that he plans to send the first uncrewed spacecraft to Mars within two years.
Brett Adcock shared updates on Figure 04 at an all-hands meeting: the model is now in full design lock, and the team has begun shipping components. He called it the iPhone 1 of humanoid robots. The actuators are being developed in-house, with the goal of completing full characterization and operation of each unique actuator by the end of this week. The new dexterous hand was described as an engineering masterpiece, with the team spending months refining fingertip and contact-surface tracking accuracy to the millimeter level. Brett stressed that large-scale pre-training is the key to winning in robotics, and Figure plans to combine the world's largest data and compute to build the smartest AI for humanoid robots.
Daniela Shchegelskaya, an electronics and robotics engineer at Elimpus, shares two years of experience using Boston Dynamics' Spot for hazard-zone inspections in live power substations. She argues that robots should take on tasks such as carrying instruments into hazardous locations and providing more detailed viewpoints, keeping people away from danger while preserving key judgment.
At the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree Technology CEO Wang Xingxing delivered a roughly 30-minute speech, saying that almost everyone in the world still does not place enough importance on and faith in AI, including himself; physical AI has shown signs of being broken through, likely to happen in the next 1 to 2 years or 2 to 3 years, and the industry is moving toward the ChatGPT moment of embodied intelligence, that is, in 80% of unfamiliar scenarios, robots can complete about 80% of tasks according to voice or text instructions.
During the spring 2026 school term, humanoid robots and robot dogs entered primary and secondary schools across various regions in droves as greeters and performers, but most were rented equipment costing several thousand yuan per day; they were loaded onto trucks and hauled away as soon as the ceremonies ended, lacked supporting curricula and teachers, and some became hallway decorations or even malfunctioned, falling over and kicking students. In May 2026, the Beijing Economic-Technological Development Area delivered 20 humanoid robots retired from robot soccer competitions to five secondary schools in the district for school-based curricula and club activities, encouraging students to further develop them, and this has been seen as a reference case distinct from performance-style displays.
Speaking at the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree founder Wang Xingxing said that with improvements in general large models, physical AI may see a major breakthrough within one to three years. He said that GPT-6 Astra, under zero-training conditions, enabled Unitree robots to complete painting and Go-playing experiments, but he also noted that large models still have clear shortcomings in perceiving changes in the real environment and correcting operational errors.
Unitree Robotics held the World Physical AI Conference in Hangzhou to celebrate its 10th anniversary. In a 36-minute speech, founder Wang Xingxing mainly urged people to believe in AI and reflected that he himself had not believed enough. He shortened his expected timeline for embodied intelligence's ChatGPT moment from 2 to 3 years, or 5 to 10 years, as of August this year to 1 to 2 years, or 2 to 3 years at the slowest.
Ozgur Tohumcu, AWS Global General Manager for Automotive and Manufacturing, said the bottleneck to scaling Physical AI is usually not technology but the data foundation, agent guardrails, and organizational tolerance for failure, and that many automakers' AI projects remain stuck at the proof-of-concept stage.
On October 21, Unitree released the H2-X Kung Fu Legend new robot. In the demo video, the robot leaps over a car, kicks a car door until it dents, lifts a sedan with both arms and pins it down, and completes a backflip from a 4-meter-high platform; the comments section debated whether it was remote-controlled. Unitree only disclosed "maximum battery life of about 8 hours" and did not reveal the hardware configuration, nor did it explain whether the movements were manually remote-controlled, pre-scripted, or autonomously controlled. The article notes that humanoid robots are currently mostly in the early L2 stage, that teleoperation is a transitional means toward autonomy, and that manufacturers should clearly distinguish autonomous parts from human-assisted parts.
In his keynote at the 2026 World Physical AI Conference, Unitree founder Wang Xingxing said that with the release of GPT-6, physical AI has shown signs of being broken through, and a breakthrough could come within the next 1 to 2 years or 2 to 3 years. He believes the core challenge for embodied intelligence to pass the ChatGPT moment is aligning AI model inputs and outputs with real robots, and judges that commercial service scenarios will be deployed faster than home and industrial ones. According to the on-site introduction, the G1 robot released in May 2024 has cumulatively produced about 10,000 to 20,000 units.
At the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree Technology's Wang Xingxing gave a roughly 30-minute speech, saying that GPT-6 has been fully deployed on Unitree robots with zero training, enabling them to autonomously paint, clean paintbrushes, play Go and chess, and find kitchens and tools on their own. He judged that physical AI being cracked is very likely to happen in the next 1-2 or 2-3 years, and that the industry is moving toward an embodied-intelligence ChatGPT moment, in which robots can complete about 80% of tasks in 80% of unfamiliar scenarios based on voice or text instructions.
Musk said on social media that money will become useless in the future and advised people not to save for retirement. Netizens shot back with "Let them eat cake," mocking him for being out of touch with ordinary people's lives. The original text provides only this snippet of his remarks and does not involve any robot products or technical content.
The article analyzes the visible and hidden lines behind the convergence of physical AI and industrial scenarios. Han Chen, president of ABB China, believes that working conditions with small batches, multiple varieties, and high-frequency changeovers are best suited to physical AI, but the existing thresholds of safety, efficiency, and cost in industry leave no room for trial and error. Han Chen says that AI hallucinations have very serious consequences in factory operations.
At the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree founder Wang Xingxing said the GPT moment for embodied intelligence means robots can successfully complete about 80% of main tasks in about 80% of unfamiliar scenarios via voice or text commands. He believes large models already have strong capabilities in robot planning and navigation, and that the biggest challenge in passing the GPT moment remains aligning AI model inputs and outputs with real robots. He expects that, at the fastest, by the end of this year or next year, robots may be able to complete tasks such as tidying a room on command.
At the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree founder Wang Xingxing said that with the release of GPT-6, physical AI is showing signs of breaking through and could achieve a breakthrough in the next 1-2 or 2-3 years. He believes the core challenge for embodied intelligence to pass the ChatGPT moment is aligning AI model inputs and outputs with real robots, and that deployment in commercial services will come faster than in home and industrial settings.
Tesla CEO Elon Musk said on the company's earnings call that the humanoid robot sector will become the world's largest industry, and that the number of humanoid robots on Earth may eventually exceed the human population. He expects that by late next year, issues such as robot AI, electromechanical design, reliability, and service life will hopefully be resolved, and only then can Optimus be delivered to companies other than Tesla.
In his GTC Berlin speech, Jensen Huang said the ChatGPT moment for robotics is approaching rapidly rather than having already arrived, and on the same day released upgrades to the Isaac platform, a humanoid robot reference framework, and simulation and data tools to prepare the infrastructure for that moment.
The DreamZero team proposes a cross-embodiment API approach, attempting to use a unified interface to let a general-purpose embodied brain adapt to humanoid robots from different manufacturers, while Unitree emphasizes that embodiment engineering challenges such as precision force control, whole-machine consistency, and long-term reliability have yet to be fully resolved. The article analyzes the divergence between the two approaches: the software camp argues that the brain should be decoupled and that hardware is merely a carrier, while the embodiment camp insists that body dynamics and intelligence are deeply intertwined. Within the short term of 3 years, the integrated software-hardware approach will remain the mainstream commercialization path, and a cross-embodiment API can only be implemented after infrastructure such as unified hardware interfaces and widespread automated calibration tools matures.
CITIC Securities believes the robotics theme rally is already primed to kick off and is bullish on the mass production and application prospects of the robotics industry, recommending attention to core companies in the supply chain. Its basis includes advances by GPT-6 and Gemini 4 in robot planning, spatial perception, and Real-to-Sim modeling, as well as faster mass production ramp-ups by domestic robot makers and Optimus, with Tesla expected to release Optimus V3 in the fourth quarter.
NVIDIA CEO Jensen Huang said at the Build a Claw event in London that OpenClaw is one of the most important models to date, calling it "the first truly trustworthy agent runtime framework" and describing it as providing a real-time runtime framework for LLMs that can be used in robotics. He also stressed that autonomous systems must treat every object as untrusted by default and grant only minimum privileges, and introduced OpenShell, an isolation system NVIDIA developed for this purpose that continuously monitors for privilege violations and issues alerts.
Demand for AI will outpace the supply of data and compute. The winners will be those who extract the most value from every byte and GPU cycle. @flappyairplanes is tackling this from both the kernel and pretraining sides. Congratulations to @amspector100, @spectorb, @aidanmantine, and the Flappy team on closing a $1 billion Series A. ✈️
At Tesla's Q3 earnings call, Optimus has begun production on the Fremont high-volume line. The company says these are the first production-quality complete units, and that existing manufacturing processes already apply to the commercial product. Musk said training data requires a scale of about 10,000 Optimus units, and Tesla is setting up an Optimus Academy to let the robots learn. Humanoid robots lack real-world data, so they must first learn by watching human guidance, then through self-play in simulation and in reality.
Just want to confirm: the SOTA for VLA is an 80% success rate, right?