Humanoid (almost) Daily: Apptronik, Google, Mercedes, China's robot winners, Tutor Intelligence, and more ...
Today in Humanoid Daily ... Apptronik's Apollo 2 used Gemini Robotics 2.5 to hit a 96% success rate at Mercedes-Benz, it looks like Beijing wants just 5 world-class humanoid makers, military humanoids are in the news again, Tutor Intelligences shares big physical AI news, Carnegie Mellon opens sources a training glove, China has new bionic robot news anchors, an Antarctic robot runs out of juice, and the number of the day is $1.66 billion.
Apptronik and Google hit 96% success rate with Mercedes
Google DeepMind just released Gemini Robotics 2.5. Apparently, it's very good. Apptronik's Apollo 2 humanoid, which is powered by Gemini Robotics 2.5, has been sorting automotive parts on active Mercedes-Benz production lines in Hungary and Germany.
The task is called kitting: picking metal parts from bulk bins and arranging them in precisely the right sequence for vehicle assembly.
The success rate? 96% after just a short post-training period.
That's a real-world result, not just a lab benchmark, although Google calls it a proof-of-concept for now. In broader testing across more than 200 tasks, the new model achieved success rates 30% to 100% higher than previous baselines, while completing tasks 20% to 30% faster.
Three AI models work together:
- Gemini Robotics 2.5 controls physical movement
- Gemini Robotics ER handles workflow reasoning
- Gemini 4 Argon accelerates training with thousands of simulated practice runs
Google also reports significant safety improvements near obstacles and people, which is important, of course. The big picture: AI foundation models are moving from answering questions to controlling robots doing useful work in real factories.
And, of course, Google DeepMind is getting better and better.
Apparently Beijing wants five world-class humanoid robot makers
China has hundreds of companies pursuing humanoid robotics. But apparently, Beijing wants to make sure some of them become global champions.
A new three-year plan released November 11 calls for developing at least five world-class humanoid manufacturers and ten nationally leading manufacturers by 2028. That's not necessarily a national cap on competitors, but look at the bigger context.
- Chinese regulators have recently slowed the rush of humanoid robot IPOs, questioning inflated valuations and whether government-backed robot orders represent genuine commercial demand.
- Now Beijing is setting specific targets: 25 percent of China's humanoid production capacity in the greater Beijing region, more than 60 percent locally sourced components, and over one million hours of robotics training data.
I'm thinking China's humanoid industry may be entering a new phase, going from funding almost everyone with a promising robot to cultivating companies with the scale, technology, and business models to dominate globally.
Military humanoids in the news again
An American robotics startup backed by Eric Trump is reportedly in talks to supply Israel's military with up to 1,000 humanoid robots. The company is Foundation Future Industries and its robot is called Phantom MK1.
According to the Financial Times, company executives have discussed deploying the robots in Gaza's tunnel networks, where they could detect trip wires, identify explosives, and potentially neutralize booby traps. Company co-founder Mike LeBlanc says combat could eventually become their most important job.
The discussions are preliminary and there's no confirmed order yet.
But there's a fundamental question here: should autonomous humanoids ever be permitted to make lethal decisions? And, if we decide they should not, how would we enforce that?
Tutor Intelligence teaches robots with just a few demonstrations
Boston-based Tutor Intelligence's new TI-0.1 system learns robot tasks from just a handful of human demonstrations, then uses AI-generated simulations and reinforcement learning to practice thousands of times before deployment.
The result: robots performing tasks autonomously at human speed, without requiring massive real-world training datasets. It's another approach to one of robotics' biggest challenges: making robots smarter without needing millions of hours of expensive physical training data.
Very impressive, and it reminds me of Standard Bots' similar tech.
Carnegie Mellon open-sources a glove that teaches robots touch
Researchers at Carnegie Mellon developed ART-Glove, a wearable device that captures 22 degrees of hand movement and 2,048 tactile sensing points at 120 times per second.
Even better, they've open-sourced its hardware designs and assembly instructions, allowing others to build their own.
The goal: scale the collection of human manipulation data so robots can learn not just how our hands move, but how they touch and handle objects.
China's newest news anchors have bionic skin
China's Xinhua News Agency and UBTECH Robotics unveiled two bionic humanoid news anchors at the World Media Summit in Beijing.
Their names are Xinyan and Xinyu and they're full-sized humanoids with lifelike artificial skin, designed to work with Xinhua's AI news systems. The idea is to connect a physical robot to an authoritative, real-time news pipeline, letting it present information and interact with audiences. And yes, China's state news agency emphasizes that the entire process is traceable, secure, and controllable.
The robots could eventually appear in educational settings, international cultural exchanges, and even, apparently, deep space reporting.
The first Antarctic ultramarathon robot lasted three kilometers
A humanoid named Qestr entered the Antarctic Ice Ultra on November 8, competing alongside human runners in temperatures around minus 24 degrees Celsius. It wore a special thermal jacket, but that was not quite enough.
Three kilometers, its battery got too cold and tapped out, while the human athletes continued running distances ranging from 50 to 160 kilometers.
It's funny, but it also illustrates a well-known engineering problem: extreme cold can dramatically reduce lithium-ion battery performance. Turns out that good old-fashioned humans still have a few advantages.
Number of the day: $1.66 billion
$1.66 billion. That's approximately how much South Korea plans to invest in developing a domestic humanoid robotics industry through 2030. The 2.3 trillion won initiative, announced in August, focuses on building a complete domestic ecosystem, from robot bodies and AI systems to critical components.
As part of this, South Korea wants to increase domestic sourcing of key humanoid components from 45 percent to 80 percent: actuators, sensors, motors, and other technologies that are part of effective humanoid robots.
The reason, of course: China is in the lead here, and Korea doesn't want to be left too far behind.
Sources
Written by
I tell stories with data and build startups. I've raised angel and VC funds, led software companies and startups, and chronicled the rise of the mobile economy. Currently I write for Forbes, host several podcasts, and am starting up HumanoidMegaHub. I was TUNE's mobile economist and Singular's VP of Insights, built the research team at VentureBeat, and managed teams creating software for partners like Intel and Disney. In past lives I built websites, e-learning platforms, and mobile apps, and, a long time ago, was named to Folio's top 100 of the media industry's "most innovative entrepreneurs and market shaker-uppers.”