TL;DR
INDUSTRY WATCH #1
Nov 13, 2025
AIDOL Humanoid Robot Demo and Fall
AIDOL, Russia’s first humanoid, walked onstage in Moscow and fell seconds later during its public unveiling. Multiple outlets published the event footage. The video circulated widely on social media and was the source of many vodka-related jokes.
The CEO of the company, Vladimir Vitushin, said lighting and calibration problems caused the loss of balance. The public demo exposed the gaps between lab performance and real environment. It also highlighted persistent perception and control fragilities under variable lighting and stage conditions.
Successful mistakes can turn into knowledge, while failed mistakes can turn into experience. We hope this mistake will provide us with valuable experience.
INDUSTRY WATCH #2
Oct 21, 2025
UBTECH Walker S2 - World’s First Mass Delivery of Humanoid Robots
UBTECH reports the first mass deliveries of Walker S2 humanoids to industrial customers. The rollout is presented as commercial scale rather than pilot shipments. The company claimed that Walker S2 is ready for continuous, real-world logistics and factory tasks. Read more about Walker S2 here.
If genuine, this is a shift from prototypes to deployed fleets. Practical issues follow: maintenance at scale, safety in mixed human-robot workspaces, and potential job displacement. UBTECH frames Walker S2 as the bridge from lab prototypes to continuous industrial operation.
The announcement was met with skepticism about marketing vs reality. Brett Adcock, the CEO of competitor Figure AI, flagged the promotional footage as possibly CGI on X. This prompted UBTECH to release another drone footage of their army of Walkers.
IN FOCUS
Unified Humanoid Fall-Safety Policy
Researchers published a unified policy that combines three behaviors: prevent a fall when possible, mitigate impact during an unavoidable fall, and recover to standing afterward. The method trains from a few human demonstrations combined with reinforcement learning and a memory model to handle real disturbances.
This paper solves one of the biggest unsolved problems in humanoid robotics: safe, autonomous, adaptive fall handling. By combining sparse human priors, reinforcement learning, and diffusion-based memory, the robot learns a single instinct that redirects dangerous falls and stands up smoothly. The approach is more robust, more adaptable, and more practical than anything in current use, and could be a step forward for building resilient humanoids that can operate safely in messy real-world environments.
The approach was validated in simulation and on physical Unitree G1 hardware. Results show lower impact forces during falls and faster, more reliable recovery. This is a practical advance for on-body safety and long-term robot survivability.
Read the full paper here.
Real-world deployment on the Unitree G1
QUICK HITS
XPENG IRON’s Reverse Turing Test
When Xpeng unveiled its humanoid robot IRON at its AI Day in Guangzhou, the bot moved with such fluidity and realism that many onlookers openly speculated the “robot” was actually a person inside a high-end costume.
In response, CEO He Xiaopeng and his engineering team took to the stage the next day and literally cut open its left leg while it was powered on—revealing a metallic skeleton of rods, cables, and servos underneath synthetic skin. The reveal was as theatrical as it was convincing: once exposed, IRON continued walking without missing a beat, offering powerful proof that it is indeed a machine, not a human in disguise.
Later, Xiaopeng admitted that he and his team were driven to do this to overcome rampant online doubt. According to reports, they made the decision to perform the cut just an hour before going on stage. During the demonstration, you could hear the robot’s internal cooling system humming as the synthetic “muscle” layers peeled back.
Editor’s Take
The week was a concentrated reminder of where humanoids actually are: not magic, not ready for every floor, but moving from spectacle toward usable capability.
The AIDOL fall is embarrassing, but it is a crisp real-world signal: perception and closed-loop balance fail in messy conditions. By contrast, UBTECH pushing Walker S2 into customer hands marks the industry pivoting to operations and logistics.
The research on unified fall safety and adaptive controllers supplies exactly the missing tools: making robots less fragile and more durable in human spaces. Expect next steps to focus less on flashy demos and more on reliability, maintainability, and honest demonstrations.
