Robotics & AI: simplified | 44.000+ Followers | @micropsi | Ex-Siemens

Most robotics companies have incredible tech. Nobody knows it exists.

Your LinkedIn post gets 400 views when it should get 40,000. Your launch disappears in 48 hours.

Meanwhile, mediocre companies with great storytelling sell more, raise larger rounds and recruit better talent.

The gap isn’t tech. It’s narrative reach.

I’ve built one of the largest dedicated distribution channels for robotics and automation worldwide:

120,000+ followers, 2,000,000+ weekly views reaching VCs, operators, engineers, and developers.

Companies pay me to amplify product launches and announcements to this audience.

Not generic PR. 

Storytelling from someone who’s operated inside 50+ robotics deployments at SIEMENS and understands what actually resonates.

If you’re launching a product, raising a round, or recruiting senior talent, and nobody’s paying attention… that’s a reach problem.

You built the product, I’ve built your distribution.

For select Series A companies, I also provide strategic advisory on commercial scaling:
helping you go from pilots to repeatable revenue systems. Limited.

Open-source robot arm meets hand tracking
[📍GitHub below]

It is designed with an industrial mindset but built as a 3D-printed desktop system. 

PAROL6 paired with a LEAP Motion controller is a nice example of how accessible robot teleoperation has become.

• Hand motion is streamed to the robot at 100 Hz via UDP
• A pneumatic gripper is controlled by simple fist open and close gestures
• The entire robot stack is open source, from mechanics to control software

Combine that with low-latency hand tracking and you get a very practical platform for learning manipulation, teleoperation, and human-robot interfaces.

This kind of setup is great for experimentation, teleop, data collection, and teaching robots by demonstration

All without proprietary hardware or locked software.

Credit to @SourceRobotics

📍Code: https://lnkd.in/evJpDAB7


——

Weekly robotics and AI insights.
Subscribe free: scalingdeep.tech

Ilir Aliu

AI & Robotics | 120k+ | Scaling Deep Tech

If you think Physical AI is “just train a policy and ship it,” you’re about to waste months.

If you work on robotics,
this one is worth bookmarking‼️ 

The bottleneck is not the robot.
It’s the world.

Again: you do not fail because your model is dumb.

You fail because the real world is infinite, messy, and full of edge cases.

That’s why NVIDIA’s new Cosmos and GR00T updates are interesting.

They are basically trying to give robotics a full stack:
a better brain
a world generator
and a humanoid policy that can actually use the brain

Here’s the mental model.

Stop thinking “robot learning.”
Start thinking “robot development.”

Same way software dev needed
- compilers
- unit tests
- simulators
- CI pipelines

Physical AI needs the same thing
• reasoning that understands physics and intent
• synthetic worlds that cover the long tail
• evaluation loops that catch failures before hardware does

Cosmos Reason 2
This is a reasoning vision language model for physical AI.

It is positioned as an open model and NVIDIA says it tops Physical AI Bench and Physical Reasoning leaderboards.  

It also supports long context up to 256K tokens, plus capabilities like 2D and 3D point localization, bounding boxes, trajectories, and OCR.  

Translation in normal people terms:
Your robot can look at a scene or a video and do higher quality “what is happening, what will happen next, what should I do” reasoning.

Cosmos Predict 2.5 and Transfer 2.5
This is the world side.
Predict generates future world states in video form.  

Transfer 2.5 is built on Predict 2.5 and generates high quality world simulations conditioned on multiple spatial control inputs.  

You can manufacture training and validation videos at scale.
Different environments.
Different lighting.
Different camera angles.
Different “the thing you forgot would happen in production.”

This is how you make progress without waiting for reality to hurt you.

Isaac GR00T N1.6… This is THE humanoid action model.

NVIDIA describes GR00T N1.6 as an open vision language action model for generalized humanoid skills, trained on a mixture of robot datasets, and adaptable via fine tuning.    

Translation: Cosmos Reason helps the robot think.
GR00T turns that thinking into full body actions.

The important shift
You can now iterate like this

- generate worlds and edge cases
- reason about what is happening and what to do
- train and fine tune the action policy
- benchmark in simulation before touching hardware

That loop is the difference between
cool demos and robots that survive Monday morning in a factory.

If you’re building humanoids or mobile manipulators, the question is not “is it open?”

The question is “does it compress iteration time?”

Cosmos and GR00T are a very direct attempt to compress it.

📍 https://lnkd.in/d9VsJ8bW

📍 https://lnkd.in/dCM_Y-wV

Over the last decade, I went from selling trees to working for a robotics software scaleup, that just closed a $30M Series B funding.

Winter 2010, selling Christmas trees at -17 degree.

I decided I was going to leave that path. So, I walked away.

It took a while: ups and downs, for every step forward, two steps back...

In times I didn’t believe in myself, I had people believe in me, helping me, guiding me. It worked out in the end. 

Help and learning from other played a major part in my successful transition into the tech world.

So, I started sharing my learnings on LinkedIn as well. 

And the results were crazy.

In just 6 months I grew from zero followers to over 25,000.

Last year I left my corporate job, to work for a startup, scary, but dynamic and fun.

I worked like a maniac and stopped sharing my journey, but it didn’t work out. So I changed jobs again, and start sharing again. 

My mission is to enable non-tech people to start their own tech startups 

From fast & furious pizza-delivering in Spain during my youth, challenging tree nursery for over a decade and around-the-world traveling or becoming an aerospace/mechanical engineer...

life is always changing, and you have to conquer it!

If you want to learn more about how I did this, just come along…
 
Website: www.Ilir.co
Twitter: @_IlirAliu

Joseph C.

20 Open-Source Robotics Projects
- ON GITHUB -
[📍 Bookmark this or it will get lost]

20 projects that actually teach you how robots work, not just where to clone a repo.

	1.	Hugging Face Robotics Course with LeRobot
A complete open-source learning program that teaches robotics from scratch.
https://lnkd.in/d7ipvwjW

	2.	Open Robotics Hardware Hub
A central directory of open-source robotics hardware projects.
openrobhardware.org

	3.	ROSbot 2.0
An open-source mobile robot platform built for learning ROS in real industrial setups.
husarion.com

	4.	Magni
An affordable heavy-payload mobile base that can carry up to 100 kg.
ubiquityrobotics.com

	5.	Stanford Doggo
A fully open quadruped robot that teaches legged locomotion from scratch.
https://lnkd.in/dNkqK53p

	6.	ANYmal C
A state-of-the-art four-legged robot used in real industrial inspection research.
anybotics.com

	7.	JPL Open Source Rover
NASA’s six-wheel Mars-style rover, rebuilt with off-the-shelf components.
https://lnkd.in/eJg7WXNq

	8.	MuSHR
A low-cost autonomous race car for learning perception, control, and planning.
mushr.io

	9.	AWS DeepRacer
A 1/18th scale race car designed to train reinforcement learning models in the real world.
aws.amazon.com/deepracer

	10.	OpenManipulator
A fully open robotic arm stack from hardware to firmware and ROS integration.
https://lnkd.in/dH25Fj8X

	11.	Niryo One
A 3D-printed six-axis robotic arm for affordable manipulation research.
niryo.com

	12.	Poppy Project
A modular platform for building humanoid and bio-inspired robots.
poppy-project.org

	13.	OpenBionics
An open-source ecosystem for building robotic prosthetics and bionic devices.
openbionics.org

	14.	Nybble
An open-source robotic cat that teaches locomotion and low-cost actuator design.
petoi.com

	15.	TriFinger
A dexterous three-finger robot designed to learn fine manipulation.
https://lnkd.in/eBVvzHBJ

	16.	Open Dynamic Robot Initiative
A torque-controlled modular robot framework for advanced locomotion research.
https://lnkd.in/dbJ4idSH

	17.	Universal Robots ROS2 Driver
Official open ROS2 drivers for controlling UR industrial robot arms.
https://lnkd.in/dtFv6pE9

	18.	Autoware Foundation
The most widely used open-source autonomy stack for self-driving vehicles.
https://lnkd.in/dC7jr6TQ

	19.	Duckietown
A full learning ecosystem that teaches autonomy using small self-driving robots.
duckietown.com

	20.	TortoiseBot
An educational ROS robot designed for beginners to learn navigation and mapping.
https://lnkd.in/dj77n_9F


🔶 BONUS ONE:

PythonRobotics
A massive collection of Python implementations of core robotics algorithms.
https://lnkd.in/dpUZMFRw


♻️ Share this for better reach within the robotics community AND add your own projects below 👇 



—-

Weekly robotics and AI insights.
Subscribe free: scalingdeep.tech

1.7× faster inverse kinematics in pure Python, GPU-accelerated! Fully open-source and built for scale [📍Bookmark for later]

PyRoki, a modular toolkit for robot kinematic optimization, supporting
- inverse kinematics
- trajectory optimization, and
- motion retargeting.

Built for flexibility, but also fast:
- Runs on CPU, GPU, and TPU
- Outperforms cuRobo in speed, success rate, and accuracy
- Fully in Python, easy to integrate and extend

Whether you’re working with industrial arms, simulation environments, or humanoids, PyRoki is designed to fit your stack and scale with your hardware.

Website: https://lnkd.in/d4rzDyxB
Code: https://lnkd.in/diYsBX4b
Paper: https://lnkd.in/dywCbmv7

This one’s for researchers, builders, and anyone tired of debugging their IK stack.






——

Weekly robotics and AI insights.
Subscribe free: http://scalingdeep.tech

Robots don’t need more rules.
They need better ways to learn:
RL with synthetic data.

Group Relative Policy Optimization, or GRPO, is not just another RL trick. It points to a different learning philosophy.

Instead of giving a robot one simple reward, GRPO compares many attempts and learns from the differences.

It is closer to how humans improve through experience.

• Most robots still rely on crude, fixed rewards
• These rewards break down in real, messy environments
• GRPO creates learning signals by comparing behaviors
• That makes learning richer and more scalable

But...

GRPO needs many trials of the same task. In simulation this is cheap. 

In the real world it is slow, risky, and expensive.

That creates a paradox. 

The most promising learning method needs exactly what real robots struggle to provide, massive safe repetition.

This is why simulation is becoming as important as hardware.

For GRPO to work at scale, we need systems that can:
	1.	Run thousands of variations
	2.	Evaluate them in a meaningful way
	3.	Transfer results back to real robots

If this happens, GRPO could unlock real progress in like Navigation, Manipulation, Multi-robot coordination, or Warehouse automation

The bigger lesson is simple.

The future of robotics is not just better algorithms.

It is better learning infrastructure.

Better sims. Better data. Better bridges between virtual and real.

If we build that foundation, methods like GRPO can finally matter outside papers.

That is where real progress will come from.

Credit: @Ashish Kapoor


——

Weekly robotics and AI insights.
Subscribe free: scalingdeep.tech

Abdulrahman S. Al-Batati

Humans usually guess what robots feel.
This system actually lets you feel it.

This is an end-to-end high resolution touch teleoperation demo built with Samsung. 

Real tactile signals from a robot hand were streamed directly into a human fingertip display and a glove in real time.

Most teleoperation has no sense of touch.

That makes delicate manipulation very hard.
High resolution tactile feedback can change this. It also creates better data for training physical AI

It is still early. 

Objects were dropped. 

Control is not perfect. 

But the direction is clear.

Some stats:

• A 22mm haptic fingertip display
• 32 tiny individually controlled actuators
• Tactile sensors in the robot fingers
• Live streaming touch data to the human

If robots are going to handle the real world, touch cannot stay optional.

Seen this at Fluid ￼Reality
Video: Joshua Jung


—-

Weekly robotics and AI insights.
Subscribe free: scalingdeep.tech

What if a robot could simulate the physical world from a single image.

[📍Bookmark Paper & GitHub for later]

PointWorld-1B from Stanford and NVIDIA is a large 3D world model that predicts how an entire scene will move, given RGB-D input and robot actions.

The key idea is simple but powerful:

Actions are not joint angles❗️

They are 3D point flows sampled from the robot’s own geometry. The model reasons in the same space where physics actually happens.

• State and action are unified as 3D point trajectories.
• One forward pass predicts full-scene motion for one second.
• No object masks, no trackers, no material priors.
• Trained on ~500 hours of real and simulated robot interaction data.
• Micrometer-level trajectory error, thinner than a human hair.
• Works across embodiments, from single arm to bimanual humanoid.

The model is then used inside an MPC planner to push objects, manipulate cloth, and use tools, all zero-shot, from a single fixed camera and without finetuning.

This feels like a shift from “learning policies” to “learning physics in 3D”.

Thanks for sharing, @wenlong_huang

📍Project
point-world.github.io

📍Paper
arxiv.org/abs/2601.03782

📍GitHub
https://lnkd.in/dqsjUTxg
(will be published soon)


——

Weekly robotics and AI insights.
Subscribe free: scalingdeep.tech

First fully open Action Reasoning Model (ARM); can ‘think’ in 3D & turn your instructions into real-world actions:

[📍 Bookmark for later]

A model that reasons in space, time, and motion.

It breaks down your command into three steps:
> Grounds the scene with depth-aware perception tokens
> Plans the motion through visual reasoning traces
> Executes low-level commands for real hardware

Think of it as chain-of-thought for physical action.

Give it an instruction like “Pick up the trash” and MolmoAct will:

1. Understand the environment through depth perception
2. Visually plan the sequence of moves
3. Carry them out… while letting you see the plan overlaid on camera frames before anything moves

It’s steerable in real time: draw a path, change the prompt, and the trajectory updates instantly.

AAAANNNDDD:
It’s completely open: checkpoints, code, and evaluation scripts are ALL PUBLIC!

Resources:
Models: tinyurl.com/4fzt3cht
Data: tinyurl.com/3b3skf3f
📍Blog: allenai.org/blog/molmoact

MolmoAct runs across different robot types (from gripper arms to humanoids) and adapts quickly to new tasks.

It outperforms models from major labs like NVIDIA, Google, and Microsoft on benchmark tests for generalization and real-world success rates.

For anyone building robotics systems or studying AI-driven action models, this is worth exploring… and worth sharing! ♻️

🎙️ Stephen James is the founder and CEO of Neuracore, and Assistant Professor of Robot Learning at Imperial College London:

In this episode, Stephen shares his path from growing up in Wales to spending a decade at Imperial, a postdoc at Berkeley, and eventually founding Neuracore. Not because he wanted to be a startup founder, but because he kept running into the same problem again and again: every robotics team rebuilding the same infrastructure from scratch.

We talk about what actually slows robotics teams down, why data pipelines matter more than clever algorithms, and how Neuracore aims to become the infrastructure layer that lets teams focus on deployment instead of plumbing.

Stephen also reflects on imposter syndrome, work ethic, moving between academia and industry, and why Europe can and should build its own robotics infrastructure instead of copying Silicon Valley playbooks.

A very honest conversation about building foundations, not hype, and why scaling robotics is mostly about removing friction.

Building Deep Tech #97
"Why Robotics Keeps Rebuilding the Same Infrastructure"
With Stephen James, Founder & CEO of Neuracore


🎧 Listen on Spotify:
https://lnkd.in/dtmUivbM

🎧 Listen on Apple
https://lnkd.in/d3EGj4MZ

🎬 Watch on YouTube:
https://lnkd.in/dJ8mURZM

These machines don’t look like they build the future of AI. But they do.

What you are seeing is not an industrial robot just move.

It is part of the process that makes the world’s most advanced chips possible.

TRUMPF laser systems are used to manufacture key components for ASML’s EUV lithography machines. 

❗️Without this step, modern AI hardware would simply not exist.

• They machine parts with nanometer-level precision
• They enable the optics and components needed for EUV lithography
• Their stability directly affects chip yield and quality
• No TRUMPF lasers, no ASML EUV. No EUV, no advanced AI chips.

It is easy to focus on GPUs and models.
But the real bottleneck sits much deeper in the industrial stack.


—-

Weekly robotics and AI insights.
Subscribe free: scalingdeep.tech

Talks about #startups, #technology, and #enterprenuership

taplio

Ilir Aliu's Linkedin Analytics

Ilir Aliu

Performance ScoreIlir Aliu's LinkedIn Performance

Ilir Aliu's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Ilir Aliu's Linkedin Analytics

Ilir Aliu

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Ilir Aliu's LinkedIn Performance

Ilir Aliu's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Ilir Aliu's Linkedin Analytics

Performance ScoreIlir Aliu's LinkedIn Performance

Ilir Aliu's Best Posts (last 30 days)