Alexander Spiridonov

Robotics @ ETH Zurich

[Alexander Spiridonov @aspiridon0v]

Sneak peak of what I have been working on over the past two months

[Joni Rakipi @jonirakipi]

Egocentric camera alone is not enough to capture fine grained dexterity. We built an affordable and robust glove + headset that fix this problem.

If you are tired of occlusions, motion blur, bad lighting, and noisy keypoints, check out our sample dataset (link below) and dm me.

[Alexander Spiridonov @aspiridon0v]

the robotics services industry will eat the consumer home robot industry

[Barkley Dai @BarkleyDai]

Feeling surreal to order online, and watch a robot from @taurobots cleaning my place the next day...

Although currently it's teleoperated and still clumsy, the feeling of sitting back and watching it do the chores much resembles my FSD driving experience.

With data scaling and larger models, I think this will be the norm in 3-5 years

[Alexander Spiridonov @aspiridon0v]

@HKydlicek Yeah will be interesting to see how it all plays out. There may be a market for this for some time but I feel like every new frontier model eats away at chunk of it.

[Alexander Spiridonov @aspiridon0v]

If SaaS is dead, why do I see a new “data layer for physical AI” company every day?

[Alexander Spiridonov @aspiridon0v]

@NikolausWest Yes I really like how rerun is approaching this. But for many of these data layer solutions I feel like it is becoming more and more viable to build custom infra given how fast agentic coding is advancing.

[Alexander Spiridonov @aspiridon0v]

wait, how did my demo from 8 months ago end up in rando yc company launch 😂

[Y Combinator @ycombinator]

Intelligence Factory is building human intelligence for robots.

They train general-purpose manipulation models on human demonstration data (vision, action, and touch), then deploy them in warehouses, grocery stores, and data centers.

Congrats on the launch, @coldifl and Jalaj!

https://t.co/fj1UJtiZxx

[Joni Rakipi @jonirakipi]

april glove always knows 👀

[Ted Xiao @xiao_ted]

live look at the robotics + AI landscape

[Alexander Spiridonov @aspiridon0v]

@chris_j_paxton Humanoids benefit most from human data

[Arnie Ramesh @arnie_hacker]

In 1.5 weeks:
- Built CS:GO annotated data collection pipeline
- Coded & trained VAEs & Latent Video Flow-matching models from scratch

All on my macbook air (while travelling) lol

Time to scale up - anyone that could sponsor AWS credits and compute?

[Jannik Grothusen @JannikGrothusen]

The aesthetics of robot learning:

[Alexander Spiridonov @aspiridon0v]

@ben_sdl haha 👀

[Alexander Spiridonov @aspiridon0v]

Sneak peak of what I have been working on over the past two months

[Joni Rakipi @jonirakipi]

Egocentric camera alone is not enough to capture fine grained dexterity. We built an affordable and robust glove + headset that fix this problem.

If you are tired of occlusions, motion blur, bad lighting, and noisy keypoints, check out our sample dataset (link below) and dm me.

[Alexander Spiridonov @aspiridon0v]

@ben_sdl Thank you! Its a mix of methods :)

[Alexander Spiridonov @aspiridon0v]

@iamRezaSayar @jonirakipi CV + Glove Sensors

[Alexander Spiridonov @aspiridon0v]

@iabom @jonirakipi 🎯

[Joni Rakipi @jonirakipi]

Egocentric camera alone is not enough to capture fine grained dexterity. We built an affordable and robust glove + headset that fix this problem.

If you are tired of occlusions, motion blur, bad lighting, and noisy keypoints, check out our sample dataset (link below) and dm me.

[Alexander Spiridonov @aspiridon0v]

@jonirakipi good take

[Alexander Spiridonov @aspiridon0v]

the swiss skunk works 🇨🇭🇨🇭🦨🦨🦅🦅

[Dominique Paul @DominiqueCAPaul]

I spent 37 days doing nothing but talking to manufacturers across Europe: Cold emails, cold walk-ins, a trade show, and a week on the line in a Polish electronics factory.

Over the summer, I’d been in San Francisco working on robotic AI. Robotics is deeply technical, and closing the gap between research and reality is the hard part.

But what surprised me was how far most teams are from their customers. Everyone is trying to build “perfect robotic AI”, yet if you're in that field of research, you know that the road ahead is still long. The more interesting question is what we can already do with these models while they’re still flawed.

I wasn’t much better. A few months ago, the only shop floor I’d ever seen was one making rubber gaskets. For someone who wants to lead with distribution, I knew very little.

So I rented a car and spent September and October talking to customers and doing the work myself. Europe gave me an option you simply don’t have in the US: a manufacturing landscape so dense you can talk to customers just by getting into a car - high-quality, high-value production, and dominated by SMEs with owners who think long-term.

I met founders with more agency than most tech entrepreneurs, heard manufacturing horror stories, saw tasks I didn’t realise we still do in Europe, and spent hours talking to workers and owners about how they actually think about automation.

A lot of it was uncomfortable and full of rejection. I got far more noes than yeses. But it was worth it, and I can’t think of a cooler user group to build for.

To make it easier for other builders to talk to manufacturers – and avoid the mistakes I made – I wrote a post summarising everything I learnt and what I’ll be doing differently next time.

Link in thread.

[Alexander Spiridonov @aspiridon0v]

Wow, notice the use of UMI style grippers. Seems like that is how they are scaling data so fast. Seems like top-down camera is not as important as I thought...

[Generalist @GeneralistAI]

To scale GEN-0 capabilities, we are constructing the largest and most diverse real-world manipulation dataset ever built 🌍 including every manipulation task humans can think of – from peeling potatoes, to threading bolts – spanning homes, bakeries, warehouses, factories, & more