srinivijay

I help curious minds experience the euphoria of building with insights from 12+ years of building robots alongside NASA, Boeing, Yamaha, and more.

[srinivijay @srini_vij]

The rarest thing in engineering today isn't the technology. It's the patience to spend years on one hard thing.

Before software, spacecraft still needed to know where they were. The Soviet Globus did it with gears, cams and rotating parts. Clockwork, tracking an orbit.

I came across a similar idea recently: physical maps that tracked a car's position, doing a GPS's job long before GPS existed.

What gets me isn't the cleverness. It's the effort behind it. Someone had to understand every gear, design it, and assemble it meticulously enough to get clockwork precision.

We have far better technology now. What we're running short of is people willing to give years, sometimes a big part of their lives, to something others might call trivial, and see it through.

Those people still exist. They're just getting sparser. And this AI-driven world is speeding that up, not slowing it down.

But talk to one of them and you realize something. There's a real happiness in the flow state you get from working uninterrupted on a hard technical problem. You don't find it anywhere else. It's addictive enough that they'll build their lives around it.

I really hope our generation, and the ones after us, don't miss out on that high.

[srinivijay @srini_vij]

Watching a 3D print is pure magic.
Making one yourself is a different story.

I could watch videos of 3D printed toys like this for hours.

First, the layers. One on top of the next, until the thing slowly takes shape.

Then the sound. That clicky noise the prints make, especially in PLA.

And then the part I love most: it actually works. The pieces move exactly the way they were designed to. That gives you a different kind of high.

Now the honest part.

When you're the one doing the printing, the magic gets buried. It's slow. Prints fail more often than most of us like to admit. And designing even something simple takes real effort.

But I still think everyone should go through it at least once.

Watching someone else's print is satisfying.

Watching something you created take shape, layer by layer, is a whole other experience.

[srinivijay @srini_vij]

@sebify https://t.co/jSAR0AMmH7

[srinivijay @srini_vij]

Fifteen years ago, doing SLAM with a single camera didn't even stand a chance.

I was working on SLAM back then. Point Cloud Library had just been released. ROS1 was picking up steam. And if you wanted good accuracy, you used a LiDAR of some sort.

Cameras alone didn't cut it. Not even stereo. Bring up a single camera and nobody took it seriously.

Then I came across this open-source model rebuilding 3D scenes from one camera, live, at around 20 frames per second, and staying stable over long sequences.

That still blows my mind.

And it isn't only perception and mapping that grew up. Position estimation came a really long way too. Factor graphs, IMU preintegration and a stack of other techniques each pushed the field forward in their own way.

If you're a robotics nerd, you know how exciting this is.

The setup that didn't stand a chance in my SLAM days is now code you can clone and run.

[srinivijay @srini_vij]

@m_emanaftab You can check out the repo here https://t.co/jSAR0ALORz

[srinivijay @srini_vij]

You hold an egg and a hammer with almost the same hand shape.
The force behind each grip is nothing alike.

A camera can't see that difference.

I came across this demo of Physical AI that pairs vision with sEMG (surface electromyography): a wristband that senses muscle activation, fatigue and coordination in real time.

What amazes me is how accurate the hand tracking is.

If you thought vision alone could track a hand well, this proves otherwise.

Vision is necessary. It tells you the pose of the hand.
But it doesn't give you the full picture.

sEMG tells you the force behind that pose. And for manipulation, force is everything. Grip the egg like the hammer and it's gone.

This kind of multimodal input matters even more for the foundation models learning manipulation. Vision alone is not going to solve it.

The real question now: will it generalize?

[srinivijay @srini_vij]

Why would a robot jump with a CO2 piston instead of an electric motor?

I came across Sand Flea from Boston Dynamics. A small, simple robot that clears a 10-meter wall. And lands precisely where it's aimed.

For emergency professionals, that precision is crucial.

What makes it work is impulse. A huge force, delivered in a tiny instant.

A CO2-powered piston does exactly that.

An electric motor couldn't. Not that much force, not that fast. At least not when Sand Flea was built.

That's the part I love thinking about.

Look at the humanoids doing stunts today. Electric motors are now starting to deliver the kind of explosive force that once needed a gas piston.

The limit that pushed Sand Flea to CO2 is quietly fading.

It's fascinating how fast this technology is evolving.

[srinivijay @srini_vij]

Thrust vectoring used to belong to multimillion-dollar fighter jets.

Now it fits on a hobby plane.

I'm an airplane enthusiast, so when I see miniature machines like this, I can't look away.

Nozzles that swivel. Thrust pointed exactly where the pilot wants it. Manoeuvres that used to need a fighter jet's budget, flown by a hobby aircraft.

It reminds me of something people often say: the comforts an ordinary person has today would have been beyond the reach of kings and emperors a few centuries ago.

Look at it that way, and we're living in a golden age.

But it rarely feels like one.

Our aspirations keep moving ahead of us. Whatever we can do today quietly becomes the new normal.

We already have more than the emperors did, and it still feels like we have places to go.

[srinivijay @srini_vij]

A decade ago, if you'd told me even one of these robots was the product of ten years of academic research, I would have believed you.

Today, one builder makes them for a YouTube channel.

These come from Geng Yupeng in China. Capable, seriously impressive machines. Just a few years ago, any single one of them would have made headlines.

Robotics has clearly come a long way. But what grabs me isn't the distance. It's the velocity.

For decades, progress in this field was steady. In the last few years, it has picked up drastically. Better motors and motor controllers, pulled forward by the drone boom. Software libraries. Specialized hardware and compute, with far more choices than we ever had. All of it compounding.

It's a golden time to be around.

And if you've wanted to get into robotics but it feels impossible to crack open, I feel you. The learning curve is steep, and the sheer amount of information out there makes it overwhelming to even think about starting.

For a while I've wanted to help with exactly that. So I recently put together Roboticist Foundations, a program for anyone who wants to get started.

Not sure where you stand? Take the free five-minute quiz: https://t.co/qctvPKo56m

[srinivijay @srini_vij]

Some feats look impressive to a human and are completely mundane to a robot.

This is one of them.

I came across this clip of Tesla FSD threading a narrow cliff road in Slovenia, and my first reaction was the human one: that's hard.

And for you and me, it is. You need years behind the wheel to pull this off. Years to build a detailed mental model of the world, and of exactly where your car sits inside it.

For an autonomous car with multiple sensors watching its surroundings? The driving itself is a cakewalk.

What I love is where the hard part hides for the robot. It's not the steering. It's putting all that sensory information together into one accurate world model. No drift. No warping. Precise enough to plan the maneuver and then execute it.

So the thing that takes a human years of experience is the same thing the robot has to fight for.

In the end, it's the world model for both of us.

We just arrive at it in different ways.

[srinivijay @srini_vij]

I got fooled by this video. And honestly, part of me wanted to be.

Hundreds of people hanging from ropes, arranged in mid-air to form the shape of a car. At first I thought it was real.

Then I started thinking about the mechanics.

Hanging one person from a rope is already unstable. Now hang hundreds of them. That's thousands of pounds in the air. And the really hard part is the tangling. What keeps one rope from wrapping around the next?

It's impossible. This is AI-generated.

But what fascinates me isn't the fake. It's how my mind reacted to it. It looks so good that you want to believe it's true. Only when you take a deeper look does it fall apart.

Learning works the same way.

From the outside, it looks like you can learn anything. Then you go deeper and realize how much effort it takes, and how much you need proper direction. That's where a lot of people fall off the train.

Real progress starts once you get past that initial illusion.

If you're up for that and want to learn robotics, let me know. I might be able to help.

[srinivijay @srini_vij]

For an idea to survive, it has to stay in people's minds. If it doesn't, it's lost. Most times permanently.

I came across a car navigation system Honda built in 1981. No satellites. Just a gyroscope and a distance sensor in the gearbox.

It blew my mind to see that kind of effort.

I'm not completely surprised that it existed. There have always been people who try weird things and end up far ahead of their time.

What surprises me is how much of it gets lost as time goes on. As a society, we record the well-circulated facts. The details and the alternate methods just get buried.

Think about when the car was first invented. There were so many car companies with innovative designs. There were even electric cars. A lot of them, actually. Only a few survived, and those are the ones we talk about now. Look at the rest today and you get this same jolt of surprise.

The 60s and 70s feel the same to me. So much innovation happened around the Apollo moon missions. A lot of it is probably recorded somewhere. But a lot of that knowledge is probably gone too, which is sad.

A navigation system in a car, in 1981. And I'm only seeing it now.

[srinivijay @srini_vij]

Some machines make you smile before you understand a single thing about them.

A 3D printer that prints in chocolate is one of them. Technology nerd or someone who has never touched a printer, I don't think it matters. You're going to smile.

I did.

My mind immediately started racing through everything I could print with a machine like this. It's a kid's dream.

Then the technologist in me took over. And now I can't stop thinking about how it actually works.

Chocolate is a really tricky material for a 3D printer. It melts easily. But getting it solid enough for the next layer to sit on takes a while.

So there has to be some really interesting fast heating and cooling going on to pull this off. Or maybe they tuned the chocolate formula itself to make it easier.

Either way, the smile stays.

I'm just marveling at the awesome times we're living in. And I can't stop wondering what other surprise material we'll be 3D printing with next.

[srinivijay @srini_vij]

The robot that won the race ran the strangest.

Humanoid racing competitions are stacking up fast, and most competitors are pretrained on human motion data — their gait looks recognizably ours. A few didn't. The winner ran with an oddly stiff, almost non-human stride.

What I love about this: reinforcement learning starts from our biomechanics as a bootstrap, then it's free to wander off. Left alone, it apparently found something more efficient than the human gait it started from.

We taught robots to run by copying us. Is there a version of this where sprinters end up studying robots instead? I hope not. But I wouldn't bet against it.

[srinivijay @srini_vij]

A humanoid robot, unboxed like it's the newest iPhone.

Foam packaging, robot lifted out, stood up, presented like just another gadget. And it makes me wonder: is this trying to make humanoids feel normal, or have they actually become normal, the way a phone is?

I don't think it's the latter. Not yet.

[srinivijay @srini_vij]

A block of steel, sliced by a CNC blade, doesn't break — it flows.

I came across high-speed footage of metal cutting recently. Slowed down enough, the chip peeling off the blade doesn't shear like a rigid solid. It curls and moves like a dense, viscous liquid.

That explains something that never quite made sense to me: why machinists get nervous when you ask for a thin slice. It's not a stiffness problem. It's a fluid-dynamics problem happening at a scale we normally never see.

Slow something down enough, and your understanding of it doubles.

[srinivijay @srini_vij]

My first thought watching this: it has to be CGI. It isn't.

A battalion of Booster T2 humanoids, moving in formation at the World Humanoid Robot Games this August. Not one robot. Dozens, synchronized.

What mesmerizes me isn't the choreography — it's the question underneath it. I've spent years keeping a single field robot alive and coordinated. Doing that for dozens of robots at once isn't a bigger version of the same problem. It's tracking, sequencing, and recovering an entire fleet as one system.

Add a second robot to the picture, and you've stopped doing robotics. You've started doing fleet operations.

[srinivijay @srini_vij]

Big congrats to my friend Humphrey Hu and the team at @RobotWorksGoose on launching prototype 1 of the Goose!

A home robot built with real character — helping seniors stay independent at home, with features that go way beyond the gimmicks.

Think: bringing your walker to the bedside at night, lighting the way to the bathroom, fetching tagged items around the house.

This is the kind of robotics that actually makes everyday life better.

Excited to see where the Goose goes from here. Check it out: https://t.co/fXtuu8m8VR

[The Robot Works @RobotWorksGoose]

Meet prototype 1 of the Goose, the family robot.

We're building for the home: tidy, play, live.

[srinivijay @srini_vij]

We spent decades trying to copy nature exactly — strap on wings, flap, hope to fly. It didn't work until we stopped copying and started extracting the principle instead, which is how we got the airplane.

Festo's Optter is a four-winged robot built on that same insect-flight principle, but it's a step past just re-creating a bug. It's closer to a drone and an ornithopter fused into one machine — something nature never had a reason to build, because nature doesn't have rotors to borrow from.

That's the part I find genuinely new here. For a century, bio-inspired robotics meant getting closer to the animal. Now it means taking the animal's trick and bolting it onto a machine part that never existed in biology. I saw another robot recently that pairs a drone's airframe with a single hopping leg — another combination nothing in nature ever needed.

Nature spent millions of years solving for survival, not for our toolbox. We're finally at the point where we can take just the piece we want and build the rest from scratch. That's a different kind of bio-inspiration than the one we started with.

[srinivijay @srini_vij]

A mosquito-sized reconnaissance drone. A year ago that headline would've stopped me cold. Now I read it and shrug.

That's not the drone's fault — it's mine. We've seen enough of these demos that the "impossible tiny robot" reveal stopped landing as impossible.

What actually holds my attention now isn't whether we can build something this small — clearly we can, at least as a prototype. It's whether it can do anything once it's built. At that scale, range and flight time are brutal: there's barely room for a battery, let alone a useful payload. And that's not a design problem an engineer solves with a clever layout — it's an energy-storage and materials-science ceiling that the whole field is still stuck under.

So we're in an odd spot: we can already imagine and prototype almost any tiny robot shape we want. We just can't power most of them yet. The bottleneck moved from imagination to chemistry — and chemistry doesn't move on a product roadmap.