Farhan Khan

CEO at @shotwellst, Previously @sundayrobotics, @tesla, 🐝🐝

[Farhan Khan @farhunky]

@BrentHazman did you see the surprise 👀

[Farhan Khan @farhunky]

If anyone needs teleop data for the next robot olympics, I've been training for years

[sur4js @sur4js]

It’s a small world after all!

cc @neldien_ and neighbors

[Farhan Khan @farhunky]

@max_simchowitz woo! they’re lucky to have you, excited to see what comes out of your time there!

[Farhan Khan @farhunky]

Congrats to Dyna, especially to their data team! Getting that much high-quality data together must have been an insane effort

[Jason Ma @JasonMa2020]

Super excited to share Dyna-2, the first robot foundation model trained on over 1 million hours of human data. At this scale, we saw the emergence of a cross-embodiment transfer scaling law: training on increasing amount of human data not only improves model prediction on held-out human data but also on robot data the model has never seen before 🤯

We validate that this transfer scaling law does translate to on-robot performance across 3 different robot platforms, and uncovers that both training objective and data matter greatly for this emergence. Crucially, our results position video as a new scaling axis for physical AI.

Beyond these scaling-law oriented results, we also dissect Dyna-2 greatly, going in-depth on its many capabilities, including language following via world modeling, enhanced robustness and precision, zero-shot customer site deployment, and some cool video generation results!

Read our technical blog post for more details! I do think this is a very important result that shows a very different path for robot foundation models from the ones we are marching on. Really excited for the road ahead. We have more releases coming, stay tuned!

[Farhan Khan @farhunky]

htrllap chapters 16-17.

[andi (twocents.com) @Nexuist]

Just watched The Odyssey and I have a bone to pick with Nolan. You put a sex scene in Oppenheimer but not the movie about having a string of affairs with nymphs for a decade? What the hell?

[Walden Robotics @WaldenRobotics]

Autonomy isn't binary. Neither is robot learning.
In our latest post, CTO & co-founder @Ben_Burchfiel explains why deployment, remote assistance, and high-quality training data are deeply connected.
https://t.co/wKMBeUerjr

[Farhan Khan @farhunky]

@abidlabs @ShotwellSt @xdof @huggingface We have our HF experts on it ;)

[nour e @neldien_]

13,692 episodes with @ShotwellSt labels just open-sourced! Sub-second accuracy on boundaries!

[Farhan Khan @farhunky]

@catboosted stop by next time!

[Farhan Khan @farhunky]

“Annotations” are a loaded term. Every company has different needs, different techniques, different strategies. And it’s difficult to have conversations that arent grounded in data. That’s why we’re sharing these annotations.

Play around with the demo here: https://t.co/jJprFZT8MO

[Shotwell. @ShotwellSt]

Today, we are releasing over 415 hrs of annotations on top of the expansive ABC 130K data set from @xdof. We took 5 tasks and ran them through the Shotwell engine, and here are the results: https://t.co/5zVQ0BAVH5

[ali @si3luwa]

Automated temporal segmentation, dense descriptions of the segments, and scene descriptions of the data, with a bar for quality only humans can hit. Play around here: https://t.co/WYMhkiBkdE

[Farhan Khan @farhunky]

We are always looking to collaborate: if you’re a research or academic group looking for annotations for your research, let us know or upload your data today: https://t.co/8QEfO9qV0k

[Farhan Khan @farhunky]

@c0dyswain @ShotwellSt appreciate it @c0dyswain!! 🙏

[Farhan Khan @farhunky]

Today, we're introducing @shotwellst.

We're solving the hardest problem in robotics: understanding your data

There’s no point collecting robotics data if you don't understand whats happening in it, but your options are limited. If you need dense, accurate labels with precise subtask boundaries, VLMs don't cut it, an in-house annotation army is a huge operational sink, and today's vendors charge an arm and a leg for low quality work

Shotwell combines the best of both worlds by training our own annotation models and letting humans deal with the ugly edge cases.

Its the ONLY way to consistently guarantee 100% quality in a scalable way. Humans are inconsistent, and models are inaccurate.

[Farhan Khan @farhunky]

@PrapthiAgarwala @Ultraroboticsco It totally should! Just DMed

[Farhan Khan @farhunky]

@Ultraroboticsco We’re already processing thousands of hours a week for our customers today. If you have data you want us to look at, drop it here and we'll respond within the next 24 hours https://t.co/3mx4l7uA7o

[Farhan Khan @farhunky]

@Abdella6if @abidlabs This is dope!!

[Farhan Khan @farhunky]

Send us your data today, and get it annotated by tomorrow. The best demos happen on your own data!

[Farhan Khan @farhunky]

@Ultraroboticsco We’re already processing thousands of hours a week for our customers today. If you have data you want us to look at, drop it here and we'll respond within the next 24 hours https://t.co/3mx4l7uA7o

[Farhan Khan @farhunky]

@viktor_malstrom @ShotwellSt We've labeled some pretty chaotic data from live deployments :P Would love to do a free sample for you! Feel free to DM me

[ali @si3luwa]

People think we’re named after the street in the mission but it’s actually the other way around. The street was named after us!

Excited to announce @ShotwellSt today :)

[Abdulrahman Alfozan | عبدالرحمن الفوزان @alfozan]

@farhunky @ShotwellSt Congrats on the launch team 🙌! @farhunky , @neldien_ , @si3luwa , @aliiiiiiiabid