Welcome, sci-fi realists…

A team of researchers created a framework that allows two humanoids to play table tennis together through coordinated movement and alternating strikes. The PingpongDuo “framework” is a system that combines AI, risk estimation, and motion learning to control the robots. Researchers from NVIDIA, the University of Hong Kong, Archon Robotics, and the Shanghai Innovation Institute worked together, training the framework’s policies using actual footage of humans playing ping pong. 

How well does it work? After just one test run, two of Unitree’s G1 humanoid robots were coordinated enough to keep the ball on the table for more than 10 consecutive strikes. While it’s easy to write this off as pure spectacle, with little practical value, the researchers advanced the knowledge of how a framework like this can get humanoids to work in tandem on industry-related tasks. 

Still, the spectacle alone might be worth the price of admission. We wouldn’t be surprised to see full-blown humanoid table tennis tournaments in the near future. Imagine the viral clips from that. 

In the meantime, here's what you need to know about robotics right now:

NEURA Robotics releases a new foundation model for humanoids

German-based NEURA Robotics has just released Neuron AI, a foundation model that helps humanoids grasp the world around them through multiple types of input: 3D vision, sound, language, and touch. It’s no wonder they’re calling it a “robot brain.” The goal is to use these capabilities to get companies to adapt humanoids to their own specific use cases.

NEURA calls this approach Perception-Language-Action (PLA) in technical terms. A PLA model takes all of this sensory input and translates it into physical actions. 

The press release also highlights Neuron AI’s memory component, which retains information about things like tasks, observations, and objects so humanoids can complete multistep tasks without losing context. 

What this likely means for the future: NEURA will be testing Neuron AI at upcoming events, starting with NVIDIA GTC Berlin. The company will have its own humanoids perform domestic tasks such as folding laundry. If the demonstrations prove successful, NEURA will likely promote this model as viable for factories and other industry settings.   

Robots in Action

Ping pong is not the only sport humanoids can play, apparently. Google DeepMind recently tried a similar experiment by getting one to play cornhole. DeepMind trained frontier AI model Gemini 4 Argon to get a humanoid to be successful at the popular beanbag toss game. This one also involved an AI model trained on human demonstrations. 

DeepMind posted the video to its YouTube channel. It shows an Apollo 2 humanoid made by Apptronik throwing bean bags closer and closer to the hole until it finally lands one on target. We thought the promise of robots was that they’ll take over the jobs we don’t want, freeing us up for more fun pursuits, but it looks like they might be angling for our weekend pastimes, too.

Around the Horn

If you want to get more out of AI, we have one more link for you. The Neuron Academy offers practical, self-paced courses to help professionals use AI in their work. You’ll find courses on the best ways to research, create content, and write better prompts, plus how to build workflows, agents, and apps. The academy will even recommend courses based on your specific goals. Check out the scope of courses we offer, and get a leg up on the humanoids… 

Weekly Feature: Bots Behaving Badly

What’s something that isn’t working so well?

A humanoid in Cologne, Germany, got picked up by the police after a 16-year old girl accused it of sexual assault for groping her inappropriately at a restaurant. Yes, you read that right. Police seized the $5,600 commercially available robot along with the phones of the alleged owners, who livestreamed the alleged assault. 

Police are investigating the incident, but legal experts say prosecution might pose challenges under German law, which requires direct physical contact between people for a criminal sexual harassment charge to be viable. We said before that robot failures are funny until they aren’t, but this one feels like it might end up being a failure of the humans at the controls.

Industry Signals

P.S. If you haven’t already… follow us on Youtube!

Not subscribed? Subscribe here.