@TBytes_Cyber
Manchester
Joined May 2023
A stable grasp depends on more than what a robot can see—it also depends on what its fingertips feel over time. “Temporal Visuo-Tactile Learning for Dexterous Grasp Stability” studies how high-resolution, dynamic touch can help a multi-fingered robotic hand predict whether an object will remain stable after lifting. Using 10,000 grasp trials across 200 objects, the researchers trained temporal models that combine vision, proprioception, and tactile signals. Adding touch improved stability prediction, while visuo-tactile model-guided regrasping increased the success rate of executed lifts by 10.5 percentage points over a non-tactile gate. The dataset and project details are publicly available. Title: Temporal Visuo-Tactile Learning for Dexterous Grasp Stability Project URL: lasr-lab.github.io/dexterous… Paper: arxiv.org/abs/2610.10283v1
1
1
117
Tasha Bytes retweeted
A tube can look right while it is held in the tooling, then open out slightly once released. This is springback, a normal part of metal forming that needs to be accounted for when producing accurate components. Read more from SMI- sheetmetalindustries.com/und…
1
3
74
"Math was made by humans for humans" And now humans have given math the ability to think about itself, and reach into the dark unknown, bringing back new knowledge. If you can't see how gloriously transcendent and divine this is, I feel bad for you.
Why wouldn’t mathematicians want solutions to these huge open maths problems? shoutout @mathandcobb for being a consistent voice in this conversation
28
26
3
318
9,112
Tasha Bytes retweeted
Model and Simulate (almost) Anything 🤙 Cartesian is now open for early release at: cartesianbyformas.com
1
4
47
2,480
Rest in power 🫡
We’ve lost an icon. Margaret Hamilton, a computer science pioneer best known for her work at MIT leading development of the flight software for the Apollo missions to the moon, has died at 90. Read about her legacy via MIT News: news.mit.edu/2026/margaret-h…
3
4
128
Tasha Bytes retweeted
Want to build your AI skills through hands-on experimentation? Join us October 14 as Arduino Product Evangelist Andrea Richetta explores how Arduino UNO Q brings AI to the edge, using person recognition to demonstrate vision AI in action. The session will also introduce "AI Upskilling Certificate: Hands-On Development from Model to App," a free, project-driven learning experience that takes you beyond the demo with practical, end-to-end AI projects using UNO Q. Be sure to save the date and tune in! youtube.com/live/BzSz8nFKJ-M
2
11
1
60
6,297
Tasha Bytes retweeted
Object-centric 3D generators collapse on full scenes—canonical boxes fail outdoors. Building Rome from a Single Image fixes this: adaptive chunking + 2D-3D lifts turn one photo into complete indoor/outdoor meshes 🏛️ 📄 arxiv.org/abs/2610.08790v1 🌐 build-rome.github.io/
5
55
2,429
Turkish team building a desktop CNC, extremely tempting: ‣ Ø200 mm × 180 mm five-axis work area ‣ 10-tool changer ‣ 2.2 kW spindle ‣ $6k
12
5
84
6,647
Tasha Bytes retweeted
Wrist cameras help robots a lot. Taking them off and letting the robot look at the right thing helps more. EyeRobot 2.0, from Kush Hari, @justkerrding and the team at @UCBerkeley and Amazon FAR (advised by @Ken_Goldberg and @akanazawa), has no wrist cameras at all. One stereo head moves its gaze to whatever matters at each step. Sharp in the middle, blurry at the edges, like our own eyes. Seven real tasks, including capping a marker, zipping a bag and pulling a pan out of a toaster oven. Same demonstrations and same policy architecture throughout, 25 trials per task. Fixed stereo camera: 27%. Wrist cameras added: 52%. Active gaze: 67%. When a grasped tool blocks the wrist cameras, that policy drops to roughly the level of the fixed camera. Gaze keeps working. The gaze is trained with RL, with no gaze demonstrations. Where it looks is where it reaches. When the gaze lands on the wrong object, the robot picks up the wrong object. That makes failures easy to read, and it suggests where to improve next. Paper: arxiv.org/html/2610.03710v1 Source: eyerobot2.github.io/ #PhysicalAI #RobotLearning #VLA
3
10
1
95
6,584
Tasha Bytes retweeted
Today we are releasing security-one: a system one model trained to catch attacks across agents, infrastructure and applications. security-one is designed as an always-on that protects your stack from the Hugging Face incident.
Today we're releasing Security-One, our open-weight 27B model for security decisions. It reads a document, tool call, or code change and returns probabilities for the answers you define. Your code decides what to allow, block, or send for review.
4
5
422
Tasha Bytes retweeted
Building AI Factories: Why Data Movement Is Reshaping Semiconductor Design As AI infrastructure evolves, modern hyperscale data center environments are transitioning into coordinated computing systems known as ai factories... Watch Now>> ow.ly/F53W50ZM65O
1
1
5
163
Most excited for my day today to begin with a talk to founders from Venture Beyond Borders. "Ventures Beyond Borders is the first global accelerator and non-profit fund for refugee founders. We take founders all the way from idea to investment." venturesbeyondborders.com/
2
2
27
2,475
Tasha Bytes retweeted
The #Cybersecurity Spiral of Failure - And How to Break Out of It How decades or #corporate short-termism and lip-service around cybersecurity have led to the endless series of #cyberattacks and #databreaches we see today Everything #security vendors and consultants don't want you to read >> buff.ly/Foijj3H #tech #business #leaders #leadership #managementga #governance #CISO #CIO #CTO #CEO #cyberthreats #cyberrisks
2
7
8
156
Tasha Bytes retweeted
Papers to watch: UniWAM The idea: stop separating "understand the world", "predict what happens next" and "act" Instead their UniWAM combines a VLM-based physical reasoner + video world model + action policy in one architecture. The model was trained on: - robot demos - human egocentric video - VQA data Results - achieved SOTA across several benchmarks: 99.2% LIBERO 92.6% LIBERO-Plus 68.3% RoboTwin Clean2Rand UniWAM The part I find most interesting: human + robot co-training appears to follow a log-linear scaling trend. Another datapoint that human video may matter a lot for scaling robot foundation models. project page: uniwam.github.io
11
14
4
154
12,564
Tasha Bytes retweeted
Feed-forward models distort geometry even when given camera poses. In paper to appear in #NeurIPS, we recast multi-view stereo as sequence-to-sequence: a camera-aware transformer with a unified global cost volume predicts geometry for all views jointly. arxiv.org/abs/2609.24850
2
20
1
202
11,607
Tasha Bytes retweeted
Congrats to the AlephAlpha team on a very cool German-English model and very detailed technical report! Check it out and give the model a spin!
Small bird, fast wings, Kolibri is here. 78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe. Now the weights are yours. Run it on your own hardware, under Apache 2.0.
11
13
220
12,004
This is one of the clearest examples of why vibe manufacturing is going to be real. Someone without a background in metal fabrication or electrical engineering can now create a custom physical product and actually get it manufactured. But this is only the first layer. The bigger unlock is when AI also handles the manufacturing knowledge around the design: • requirements • process selection • simulation • manufacturability • sourcing • testing • cost • revisions • failure feedback That is when building physical products starts becoming dramatically more accessible. nitter.cf/konstantinsaifo/status…
We are now entering into an era where any product can be created exactly to a consumer's preferences & needs. I vibe-fabricated a dog door with a wifi-controlled lock, perfectly to the specifications & design of my house. I know nothing about metal fabrication or electrical engineering. It's now getting manufactured and delivered in 2 weeks -- for almost the same cost if I bought a mass-produced item off-the-shelf.
2
4
25
1,735
Tasha Bytes retweeted
I've been paying the $20 Gemini subscription and not using it at all. Finally (hopefully 🤞) I'll get to use it?
Lots of discussion out there about our next model(!), so I wanted to give an early look as soon as possible. Introducing Gemini 4 Argon! It shows frontier performance in complex workflows, cyber defense and software engineering. Teams are using it extensively at Google, from coding to quantum computing, great feedback. Here’s a look at the benchmarks:
1
8
773