Toggle light / dark theme

Robotok: Mining Internet Videos for Dexterous Robot Learning

What if robots could learn dexterous skills by watching videos on the internet?

Teaching robots to manipulate objects is challenging because collecting training data is expensive, time-consuming, and difficult to scale across different tasks and environments. Humans, on the other hand, demonstrate countless everyday skills in videos uploaded online. The challenge is finding the right demonstrations that can help a robot learn how to perform a specific action.

A new system called RoboTok tackles this problem by turning internet videos into a searchable resource for robot learning. Given a reference video of a person performing a manipulation task, RoboTok searches for other videos showing similar hand movements and interactions with objects. Rather than relying mainly on visual similarities, it analyzes the motion trajectories of the hands in three-dimensional space, using a representation centered on the person performing the action.

This approach allows the system to recognize similar manipulation behaviors even when videos differ in camera angles, backgrounds, visual appearances, or when parts of the person’s body are obscured. Because the motion representation is compact, RoboTok can also index and search large collections of online videos efficiently.

The researchers evaluated the system using video-retrieval benchmarks and real-world robot experiments. Their results showed that RoboTok retrieved more relevant demonstrations than existing approaches and helped robots perform manipulation tasks more successfully. In real-world tests, RoboTok-guided robot policies achieved an average improvement of 42.2 percentage points in task success rates compared with the strongest baseline for each task.

The broader implication is significant: the internet could become a vast, continuously growing source of training data for robots. Instead of relying exclusively on costly, specially collected robot demonstrations, researchers could extract useful manipulation examples from videos of humans performing everyday tasks.

RoboTok does not mean robots can automatically master every skill simply by watching videos. They still need suitable learning methods and robot-specific training. But by making human demonstrations easier to find and reuse, this approach offers a promising path toward more scalable robot learning and more dexterous machines.

#ArtificialIntelligence #robotics #ImitationLearning #DexterousManipulation.

Linus Torvalds says AI is a gateway drug for coding

Linus Torvalds is using AI to develop Linux kernel code now https://www.zdnet.com/innovation/linus-torvalds-uses-ai-to-w…nel-patch/

Linux creator Linus Torvalds says AI is like a “gateway drug”


Linus Torvalds AI stance has softened as he says tools boost productivity but still create false bug reports. Read what his comments reveal.

I Tested AI Cloning… and It’s Way Worse Than You Think!

I used AI to reverse engineer real binaries, understand how they work, and then clone their functionality into brand-new applications. In this video I show the full process — what the AI can actually figure out, where it falls short, and how far you can push the technique of turning an existing binary into something new.

This is the practical side of AI-assisted reverse engineering, not theory.

What you’ll see:

How AI helps tear apart and understand binaries.
Extracting logic, behavior, and structure.
Cloning that functionality into a clean new application.
Real examples of what works and what still needs a human engineer.
The limits, risks, and surprising capabilities of current tools.

If you’ve ever wondered how far AI can go in reverse-engineering and recreating software from binaries, this is the deep dive.

For a free download of Task Manager TMOG, visit: https://tmog.org

AI-assisted math advance highlights gap between finding answers and judging their significance

AI models are good at finding answers to mathematical puzzles but are not as good at research judgment, according to theoretical computer scientists.

The researchers, co-led by Princeton’s Pravesh Kothari, documented their use of AI in a case study shared on arXiv, outlining the ways in which AI tools excelled and fell short in reaching an important solution to an open mathematical problem.

AI is excellent at finding answers, they argue, because the models are trained on published papers, textbooks and solved problems. But, as of August 2026, these models were not as good at the process of recognizing the value of a theoretical result over numerical evidence or determining the significance, generality or downstream use of a result, according to the authors. This is because training data rarely includes information on the long process of mathematical discovery, which includes failed approaches, unlikely results and changes in strategy.

Storing memory on a chip with flowing water

Like the folds in crumpled sheets or the footprints on recently trodden sand, certain materials mark their histories. Now, Yale researchers show that networks of flowing water and other fluids can store their own memories—similar to how a computer chip does—an innovation that could steer robots and provide insight into some simple organisms. The results are published in Physical Review X.

In Professor Amir Pahlavan’s lab, researchers created a small chip with multiple tiny channels, each with a depth about half the width of a human hair. Inside these channels, they placed a series of flexible microfibers that stand in the way of the fluid flowing through the channels. Pressure from the flow can cause the fibers to switch between two states, affecting how the fluid passes.

This forms a “memory” of past flows, making each fiber’s current state depend on its history and, when coupled, on the states of other fibers. The two stable configurations of each fiber can represent a zero or a one, much like a bit in computer memory.

Robotic ‘time crystal’ organizes itself into a single rhythm despite random disturbances

Scientists have developed a material that can grow and coordinate its own movement until the entire structure pulses in unison, paving the way for versatile robots that behave more like living systems.

Publishing their findings today (Oct. 14) in Nature, engineers from the University of Birmingham working with the universities of Amsterdam, Lyon, Delft, Leiden and Cambridge reveal that they have created a mechatronic metamaterial that behaves as a ‘time crystal.’ Time crystals are an unusual form of matter in which particles endlessly move in regular, repeating cycles. But here, the “particles” were not the atoms of molecules: they were robots.

The scientists’ experimental material combines elastic structures, sensors, electronics and actuators so that neighboring components can influence one another. The key ingredient to generate the time-crystalline state was making the interactions between components “nonreciprocal,” meaning that one component’s response to another does not have to be mirrored in reverse.

Hirudo Android Trojan Uses Patched Banking Apps to Bypass Their Built-In Security Checks

“However, Play Protect also provides offline detection for known potentially harmful applications, meaning that the mechanism does not completely disable its protection,” ThreatFabric noted last month. “We therefore assess that the technique is more likely intended to reduce the effectiveness of online reputation and code-analysis mechanisms rather than bypass Play Protect entirely.”

Depending on the nature of the lure app used, the dropper is also designed to serve bogus screens to capture payment PINs, serve fake screens requesting login credentials for banks and wallet apps, display fake lock screen pattern and PIN overlays, and facilitate total remote control of the device.

While all this behavior is par for the course when it comes to Android banking trojans, it’s their attempts to replace the target banking app with a tampered copy that stand out. Specifically, the purpose-built replacements embed a hooking engine to blind signature, accessibility, screen-capture, and Play Integrity checks from within the app. The replacement targets eight Indonesian apps.