TL;DR
DEEP DIVE #1
7-Eleven Japan + Telexistence “Astra”
7-Eleven Japan and robotics startup Telexistence announced “Astra”, a humanoid retail worker powered by AI that can see, plan, and act like a person. They aim to deploy these robots in stores by 2029.
Astra will restock shelves, handle repetitive tasks, and help with labor shortages. It will use Telexistence’s massive data platform to train its behavior from real store footage collected across Japan’s 20,000+ 7-Eleven outlets.
This is the clearest commercial plan yet for humanoids in retail. It’s not just another lab demo - it’s a multi-year commitment tied to real stores, data, and customers.
The timeline assumes big progress in robot hands, power, and safety. Privacy is another concern when thousands of in-store robots collect live data. Plus, moving safely in crowded aisles is still a tough engineering problem.
Telexistence logo
7 Eleven logo
Help us keep sharing these stories
DEEP DIVE #2
Brooks’s Critique: “Dexterity Won’t Be Learned Soon”
Robotics pioneer Rodney Brooks poured a bit of cold water on the hype this week. He argued that humanoids still struggle with dexterity—the simple act of using hands like humans do.
Brooks says the problem isn’t just data or training. Real dexterity depends on sensors, materials, and how the hardware interacts with the physical world. In other words, you can’t teach a robot to fold laundry just by feeding it millions of videos.
His message to investors and engineers: scale isn’t a shortcut. Hardware, safety, and design need just as much attention as software.
Read the full blog here.
..a lot of money will have disappeared, spent on trying to squeeze performance, any performance, from today’s humanoid robots. But those robots will be long gone and mostly conveniently forgotten.
IN FOCUS
HumanoidExo: Learning from Wearable Exoskeleton Data
A new research paper introduces HumanoidExo, a wearable suit that lets humans “teach” robots by capturing full-body motion data. The system records how humans move, touch, and balance, then transfers that data to train humanoids.
Why it matters: one of the biggest challenges in humanoid development is collecting realistic data. Training a robot entirely in simulation is fast but doesn’t always match the real world. The HumanoidExo approach could fill that gap by turning human movement into usable robot lessons.
This isn’t a flashy viral demo, but it’s the kind of quiet engineering step that actually pushes the field forward.
Read the full paper here.
Side by side comparison of HumanoidExo and Humanoid
INDUSTRY WATCH
AheadForm’s Ultra-Realistic Robot Head (Origin M1)
Startup AheadForm revealed the Origin M1, a robot head that can blink, twitch, and make subtle facial movements using over twenty micro-actuators. It even has embedded pupil cameras and microphones to simulate lifelike interaction.
The demo looks impressive, but it also raises questions about how far we should go in making machines that mimic people. It’s a major step for human-robot communication - and a new chapter in the uncanny valley.
Editor’s Take
This week sums up where humanoids really stand: bold business moves, technical reality checks, and flashes of real progress.
7-Eleven’s Astra plan finally puts a date on humanoids in retail. Brooks’s critique brings a necessary shot of realism. HumanoidExo shows that smart data collection might close the learning gap. And AheadForm’s robot head proves we’re getting good at making robots look alive.
▶ Coming Soon: Coverage of Humanoids 2025 conference (IEEE-RAS) in Seoul (30 Sept – 2 Oct)
