Video Friday is a weekly selection of great robotics videos from our friends. IEEE spectrum Robotics. We’ll also post a weekly calendar of robot events happening over the next few months. please Please send an event to include.
Summer School on Multi-Robot Systems: 29 July – 4 August 2026, Prague
Actuate 2026: August 18-19, 2026, San Francisco
IROS 2026: September 27-October 1, 2026, Pittsburgh
Humanoid Summit Seoul: September 22-23, 2026, Seoul
Enjoy today’s video!
In just six months, our team turned GENE.01 into a fully functional humanoid platform that can walk, sense, and interact. Full-body multimodal skin senses touch, proximity, force, and temperature, bringing physical AI closer to safe and natural collaboration with people. It’s not a rendering. It’s not a concept. This is GENE.01. The future of physical AI is taking its first steps.
[ Generative Bionics ]
Why create a one-handed robot intelligence when you can learn from many robots? Our latest embodied foundation model, GEN-1, now supports a wide range of end effectors, from five-fingered hands to specialized tools and everything in between. Each hand is a different sensorimotor interface for GEN-1 to experience the physical world. By scaling pre-training across these thousands of interfaces, you teach your GEN-1 universal physical common sense that transfers to new hands and teaches them new ways to grip, push, pull, twist, and more.
And to illustrate this concept, a surprise spatula appears.
[ Generalist ]
This paper describes the design, fabrication, and flight validation of a flat-packable flying wing made primarily of cardboard. The aircraft is manufactured from three laser-cut sheets and assembled with a fold-and-lock structure that forms a load-bearing wing structure with minimal tooling and no permanent fasteners. The complete airframe can be assembled within 15 minutes, demonstrating strong potential for rapid deployment, low-cost logistics, and scalable field use.
[ AIR Lab ]
A $14,000 open source data collection system with beatdown capabilities.
[ MEVION ]
Thank you, Kent!
In collaboration with Niantic Spatial and Nvidia, [we] You can now use off-the-shelf hardware to scan real deployment sites, reconstruct them into photorealistic Gaussian splats, and run massively parallel RL. [reinforcement learning] training. Policies trained in a gym environment transfer zero shots to the real robot and trained environment. This allows you to quickly deploy more functional and robust policies for your end users.
[ Flexion ]
I don’t know why, but the version with the stubby little feet in Tron 2 is just adorable.
[ LimX Dynamics ]
What, have you already got a real job…?
[ PNDbotics ]
Well, you can stop asking what humanoid robots are good for.
[ EngineAI ]
I think the right thing to do here is to only post the disclaimer that is included in this video. “This film is a conceptual creative work and the specific scenes are presented for demonstration purposes only and do not represent the actual in-store operation process. The final store environment, robot appearance and functionality are subject to actual deployment. During actual operation, the robot autonomously performs only the specified preparation steps of the specified ice cream product, and its hands are fitted with protective gloves that comply with applicable food safety requirements.”
[ Sharpa ]
Drone delivery is now a key part of South West London Pathology’s (SWLP) modernization agenda. Since February 2026, our highly automated aircraft have been delivering urgent NHS samples across South West London, providing a service up to 85% faster than ground transport. We are excited to be part of this initiative, supporting clinicians to deliver timely and effective care to patients, and contributing to a greener and more resilient NHS.
[ Wing ]
Let’s take a closer look at what’s next for Aurora Driver. Designed to move cargo farther, faster, and more efficiently, this next-generation Aurora Driver delivers better performance, 1 million miles of durability, and cuts hardware costs in half.
[ Aurora ]
How can a robot learn to recognize objects it has never encountered? In this case, a demonstration is worth a thousand words. Using short human demonstrations, you can create fully automated training datasets and avoid the irritating limitations that hinder visual language models. Rather than describing objects verbally, the system tracks what a person touches or interacts with during a demo, tracks those objects over time, and clusters detections to handle objects that combine or separate in a scene. This avoids a major weakness of VLM, which struggles to reliably detect unusual or new objects, even with repeated and carefully designed prompts.
[ Robotics and AI Institute ]
From an article on your site
Related articles on the web
