Apptronik

Andrew Kang

@Rewkang

Nice to see our portfolio companies @Apptronik and @DexmateAI supporting research at Google

Kevin Xu

@kevinxu

$BOT portfolio holding #4 Apptronik (15% of NAV) and #5 Dexmate (10% of NAV) being used at Google btw https://t.co/j9zNyBLuhJ

Techniahqrobot | humanoid robots

@techniahqrobot

People read this announcement and many will miss what it actually means. Apptronik is not just showing Apollo 2. It is building a data factory for humanoid robots. • Robot Park is the key asset Apptronik expanded its Austin facility to nearly 90,000 square feet. Apollo 2 robots work there across logistics, manufacturing, retail and customer-driven tasks. The goal is not a polished demo. The goal is repeated physical work that creates training data. • Apollo 2 is now a training platform Apollo 2 comes in two versions. ➝ A bipedal humanoid for complex environments ➝ A wheeled-base humanoid for industrial sites where current mobile robot safety standards already exist That matters because wheels can enter warehouses faster while legs keep improving for harder spaces. • Google DeepMind is using the data Apptronik says Apollo 2 data helps train and improve Gemini Robotics. This is important because robotics models need contact with the real world. Videos and internet text are not enough when a robot has to lift, sort, carry, inspect or recover from a mistake. • This is not full autonomy yet The press release says Apollo 2 uses teleoperation and autonomous execution. That means humans still guide part of the work while the robot performs some actions through its own systems. The valuable part is the loop: humans help the robot act, the robot collects data and the model improves. • The customer sites matter Apptronik mentions Robot Park locations with Google DeepMind and customers like Mercedes-Benz and GXO. That means Apollo is being exposed to real industrial workflows, not only lab setups. Kitting, inspection, sorting, machine tending and warehouse movement all produce different data. • The back pain angle is practical The first jobs for humanoids are likely repetitive physical tasks that strain the body. Lifting, reaching, carrying totes, loading parts, sorting items and standing through long shifts are exactly the type of work that creates back pain. Apollo is being trained around those environments first. • Apollo 2 is feeding Apollo 3 Apptronik says the data from Apollo 2 will power its next commercial product, Apollo 3. So Apollo 2 is the workhorse. Apollo 3 is the product they want to ship with stronger out-of-the-box intelligence. The big takeaway: humanoid robotics is moving from single demo clips to structured data collection. Apollo 2 is not just a robot. It is part of Apptronik’s system for teaching future humanoids how real work actually happens.

Léo

@LeoKharon

Texas based @Apptronik announced their new Robot Park. This is unfortunately not a theme park, but a ~90,000 sq ft building in Austin, design to be a teleoperation data factory. The first deployments at Mercedes-Benz / GXO facilities were not so much about revenue than they were about data gathering. Most real world humanoid deployments omit to disclose the actual teleop-vs-autonomous ratio; this announcement is a signal that much more data is required before reaching full autonomy. Their video also displayed wheeled humanoids, another signal that bidepal robots are hard to work with, and that in a flat floor factory setting, wheeled robots are just as good and reliable. Also worth mentioning: Apptronik Apollo 2 is trained on Google DeepMind's Gemini Robotics foundation models. The robot hardware becomes the physical-world data-collection layer, while the brain seems to become less and less of a moat.

RoboIndex

@roboindexio

Google just launched its strongest cross-robot intelligence stack yet. Gemini Robotics 2 controls @Apptronik's Apollo 2 across full-body manipulation, but performance varies sharply: > 68.4% picking from a table > 45.7% picking from the floor > 76.3% picking from a shelf > 36% screwing in a bulb > 44% tying a trash bag > 32% using a dustpan > 92% unscrewing a bulb DeepMind says its on-device model can adapt to a new embodiment with fewer than 200 examples and a few hours of training. The new stack consists of three models: - vision-language-action control for whole-body humanoid tasks - embodied reasoning and multi-step planning - on-device inference with fast adaptation to new robot bodies Gemini Robotics 2 remains in private preview, while ER 2 is already available through the Gemini API. The Apptronik relationship is especially notable. Google participated in Apptronik’s USD 935M Series A alongside investors including B Capital, Capital Factory, Mercedes-Benz, John Deere, AT&T Ventures and QIA. Apptronik is also a portfolio holding of @RoboStrategy.

Vedant Nair

@vedantnair__

Walden Robotics came out of stealth today at $1.1B, spun out of @ToyotaResearch. With their arrival, the landscape of automotive companies and their robotics bets is becoming clearer. @Toyota → Walden Robotics @Tesla → @Tesla_Optimus @Rivian → Mind Robotics Hyundai → @BostonDynamics @BMWGroup → @Figure_robot @MercedesBenz → @Apptronik That's not to mention Chinese companies like @XPENG_Global and @CheryAutoCo with their own initiatives. The biggest manufacturing enterprises in the world are putting their money where their mouth is. We are at the beginning of the robotic renaissance. Are there any that I missed?

Jeff Cardenas

Co-founder & CEO

Co-founded Apptronik in 2016 after working in tech commercialization at UT Austin's IC2 Institute.