Karol Hausman's banner

Karol Hausman

@hausman_k • 44,259 subscribers

@Physical_int

Shorts

Very exited to announce our largest deep RL deployment to date: robots sorting trash end-to-end in real offices! (aka RLS) This project took a long time (started before SayCan/RT-1/other newer works) but the learnings from it have been really valuable.🧵

Very exited to announce our largest deep RL deployment to date: robots sorting trash end-to-end in real offices! (aka RLS) This project took a long time (started before SayCan/RT-1/other newer works) but the learnings from it have been really valuable.🧵

243,862 Aufrufe

PaLM-E or GPT-4 can speak in many languages and understand images. What if they could speak robot actions? Introducing RT-2: our new model that uses a VLM (up to 55B params) backbone and fine-tunes it to directly output robot actions!

PaLM-E or GPT-4 can speak in many languages and understand images. What if they could speak robot actions? Introducing RT-2: our new model that uses a VLM (up to 55B params) backbone and fine-tunes it to directly output robot actions!

182,937 Aufrufe

There are many cool subtasks that π-0.5 can perform (yes, all in unseen homes) Here are a few examples. Folding and hanging a towel:

There are many cool subtasks that π-0.5 can perform (yes, all in unseen homes) Here are a few examples. Folding and hanging a towel:

56,473 Aufrufe

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

Sometimes things don't go as planned...

Sometimes things don't go as planned...

276,229 Aufrufe • vor 1 Jahr

We developed a general recipe that allows VLAs to improve from experience. RL is back. (yes, this is 13 hours of coffee making)

We developed a general recipe that allows VLAs to improve from experience. RL is back. (yes, this is 13 hours of coffee making)

111,643 Aufrufe • vor 11 Monaten

Sometimes it's not humans that mess with the robots, it's the real world. Two pieces of clothing on the table:

Sometimes it's not humans that mess with the robots, it's the real world. Two pieces of clothing on the table:

151,833 Aufrufe • vor 1 Jahr

PaLM-E or GPT-4 can speak in many languages and understand images. What if they could speak robot actions? Introducing RT-2: our new model that uses a VLM (up to 55B params) backbone and fine-tunes it to directly output robot actions!

PaLM-E or GPT-4 can speak in many languages and understand images. What if they could speak robot actions? Introducing RT-2: our new model that uses a VLM (up to 55B params) backbone and fine-tunes it to directly output robot actions!

182,937 Aufrufe • vor 3 Jahren

A few examples of π0.7 in action Turn the shirt right-side out and fold it:

A few examples of π0.7 in action Turn the shirt right-side out and fold it:

25,762 Aufrufe • vor 6 Monaten

Treat yourself to a 4-hour timelapse of a robot assembling boxes off-site:

Treat yourself to a 4-hour timelapse of a robot assembling boxes off-site:

27,608 Aufrufe • vor 11 Monaten

Seems like people find these 1x videos soothing, here is 50 mins of uninterrupted autonomous coffee making

Seems like people find these 1x videos soothing, here is 50 mins of uninterrupted autonomous coffee making

25,411 Aufrufe • vor 11 Monaten

🚨 New (Offline) RL Method 🦾 🚨 Introducing Q-Transformer - new RL approach that works at scale with large models and many tasks. This is the best method we found so far that works with demos and autonomous (also negative) data at large scale. 🧵

🚨 New (Offline) RL Method 🦾 🚨 Introducing Q-Transformer - new RL approach that works at scale with large models and many tasks. This is the best method we found so far that works with demos and autonomous (also negative) data at large scale. 🧵

56,054 Aufrufe • vor 3 Jahren

Congrats on the Gemini 3 release Real AGI will need to be physical:

Congrats on the Gemini 3 release Real AGI will need to be physical:

23,211 Aufrufe • vor 11 Monaten

Many things can go wrong when running very long-horizon policies. One way to test that is by messing with the robot and seeing if it can recover. We've done quite a bit of that for π-0.5 testing:

Many things can go wrong when running very long-horizon policies. One way to test that is by messing with the robot and seeing if it can recover. We've done quite a bit of that for π-0.5 testing:

27,869 Aufrufe • vor 1 Jahr

Keine weiteren Inhalte verfügbar