@DankDubey @AgamRobotics @karthikRanga92 @CaptVenk @reddy2go @amritabhinder indigenous flight controllers for drones, mobile robots and underwater systems is exactly the kind of deeptech worth building - we're putting hardware teams on hardwaremap.com , a directory you can filter by stage, funding and country

if you have a qwen3.5 4b around, please try steering neurons L14 N8949 +1 and L13 N7956 -1
now you have evil qwen

recipe: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖` at the down_proj at every input

This is how demonstrations should be. Superb.

Robotics is turning into the physical agent for LLMs.

For the last few years, embodied AI has mostly been about getting robots to learn moves on their own. Train with RL over and over. Collect teleop data to train VLAs. Chase general-purpose humanoids with dexterous hands. What zoo of incompatible hardware, and slow progress.

There's another path that's starting to actually work: don't make the robot think. Let the LLM do the thinking.

The model looks at the camera feed, figures out the task, breaks it into steps, then uses tool calls to hit the arm's "move", "grip" and "release".
Picture a really smart kid with a game controller, staring at a screen, driving a robot arm.
Someone just had GPT-6 Astra run two arms through vessel suturing in a sim built in Godot, a game engine. It's just a sim, nowhere near real surgery. But it shows something: LLMs are getting good enough at spatial reasoning and long-horizon planning to start calling the shots in the physical world.

So here's how the work splits up:

🧠 Brain: LLM / VLM. Thinks every few seconds. Decides what to do.
🦴 Cerebellum and spine: motion planning, low-level control, safety. Reacts in milliseconds. Figures out how to move without crashing into stuff.

My little propeller balancing toy has to tweak its PWM every 10 ms or the stick just falls over. One LLM tool call takes a few seconds. Doesn't matter how smart the brain is, it can't do the cerebellum's job.
But the "what to do" part? That's going to the models, more and more.

So what does this mean for robotics companies?

An agent is basically a reseller. Robots are going to be the LLMs' resellers in the physical world.
Better models make robots more useful, sure. But more of the value keeps flowing upstream to OpenAI, Anthropic and Google.

The robotics companies that make it will look like the resellers that make it. Two kinds:
• The ones that nail the "cerebellum": hardware, control, safety. Stuff the model can't do for you.
• The ones that go really deep on one setting: the workflows, data and customers of a specific factory, lab or warehouse.

If you're just a middle layer, upstream will squeeze you sooner or later.
LLMs won't replace robotics. Robotics will become their hands and feet.

Inspired by @lidangzzz's ideas and @maoconsult511's experiment. Thanks, both.

An interesting reply post on Laksh's latest build. the (ai)r guitar.

Dinner is on Laksh tonight! He is taking us out.
@VishalPeriTS gifted him some coupons in return for a short video byte.

Flow rate calibration.

A drone is only as smart as its brain. 🧠

Agam Robotics is building that brain in India — developing indigenous flight controllers and electronics for drones, mobile robots and underwater systems.

India needs to build the brains, not just the bots. 🇮🇳

#Robotics #DeepTech

2 weeks on X taught me more than I expected.

I chased views.
Got a few viral posts.
Then watched my reach drop to almost zero.
And I realized: none of it was really me.
So I’m starting over.

I’m an 80s-born dad with two boys, 2 years apart.
They’re growing up in a world where verywhere.
Their classmates. Their coworkers. Maybe their friends.

I don’t want them to just use this stuff.
I want them to understand it, question it, and build with it.

So that’s what we’re going to do. Together.
→ Testing AI tools with my kids
→ Playing with robots & embodied AI at home
→ Sharing what works, what fails, and what makes them laugh
I’m not an expert. Just a curious dad figuring it out alongside his sons.

If you’re a parent wondering how to raise kids in the AI era, follow along.
Let’s learn this together. 🤖👨‍👦‍👦

Launching ProtoRun🏁

ProtoRun🏁 is a 2-week online builders hackathon & workshop.

$50,000 in prizes

1: Take a real business
2: Build its app on erp.ai with Proto
3: Get it customers with Meta/Social Ads. We fund your ads.
4: Win

🥇 $5,000
🥈 $3,000
🥉 $2,000
+ $40K across 100+ more winners

Every builder gets $100+ in AI tokens to build with (tbh as many as you need)

Starts Monday 26th Oct. Join the Discord 👇

Just realised one part broke off long ago, causing the issue. The second one has also come loose. Looks like a design issue.

Why does this happen? How to avoid this? Suggestions please.
@Dudechillkar @maahirpanchal @shantanugoel

Laksh designed this in Astra and 3d printed it. Building his home assistant.

The SO-101s were sitting on one corner for a long time. Will be put to use again. No mercy for any hardware.

Battle of Lanka.
Experimental, Magnific Seedance 2.5 + Opus 5.5

Want to get into robotics? These 18 projects let you explore 3D models, mechanical design, electronics, control systems and ROS without starting from scratch.

Humanoids & bipedal robots

1. Microban: github.com/Rhoban/microban
2. MiniLoong: github.com/loongOpen/Mini…
3. Poppy HumaH1Hc85
4. Otto DIY: github.com/OttoDIY
5. pib: pib.rocks

Quadruped robots

6. Mini Pupper: github.com/mangdangroboti…
7. Petoi Bittle: github.com/PetoiCamp/Open…
8. Stanford Doggo: github.com/Nate711/Stanfo…
9. SpotMicro
10. OpenQuadruped: github.com/adham-elarabaw…
11. Puguette: github.com/haruto89610/pu…

Robot arms & manipulation

12. SO-ARM100: github.com/TheRobotStudio…
13. RoArm-M2
14. OpenMANIPULATOR-X: github.com/ROBOTIS-GIT/op…
15. LeKiwi: github.com/SIGRobotics-UI…

Mobile robots

16. TurtleBot3 Burger: github.com/ROBOTIS-GIT/tu…
17. OpenBot: github.com/isl-org/OpenBot
18. Duckiebot: github.com/duckietown

Study their CAD files, inspect their 3D models and learn how the hardware and software work together.

Which robot would you build first?

5k run + 100 ab crunches. That’s it for today.

Easy 5k run. Done.

120 physical to physical ai explorers so far · 4 reached the atoms

More coming soon: fridge, pressure cooker, water pump…

atoms-to-ai.vercel.app