How can we use language supervision to learn better visual representations for robotics?

Introducing Voltron: Language-Driven Representation Learning for Robotics!

Paper: https://t.co/gIsRPtSjKz
Models: https://t.co/NOB3cpATYG
Evaluation: https://t.co/aOzQu95J8z

🧵👇(1 / 12)

Videos of humans performing everyday tasks (Something-Something-v2, Ego4D) offer a rich and diverse resource for learning representations for robotic manipulation.

Yet, an underused part of these datasets are the rich, natural language annotations accompanying each video. (2/12)
The Voltron framework offers a simple way to use language supervision to shape representation learning, building off of prior work in representations for robotics like MVP (https://t.co/Pb0mk9hb4i) and R3M (https://t.co/o2Fkc3fP0e).

The secret is *balance* (3/12)
Starting with a masked autoencoder over frames from these video clips, make a choice:

1) Condition on language and improve our ability to reconstruct the scene.

2) Generate language given the visual representation and improve our ability to describe what's happening. (4/12)
By trading off *conditioning* and *generation* we show that we can learn 1) better representations than prior methods, and 2) explicitly shape the balance of low and high-level features captured.

Why is the ability to shape this balance important? (5/12)
Because robotics isn't a single thing! While prior work focuses on learning for control, there are so many problems we care about – problems that require different features!

How do we know?

Because we build an evaluation suite of 5 diverse robotics problem domains! (6/12)
Problems like grasp affordance prediction (per-pixel segmentation) tend to require more *low-level* spatial features; edges, object boundaries, textures.

Evaluation: the ARC Grasping dataset (https://t.co/rRI4ya84DL) – CC @andyzengtweets @SongShuran. (7/12)
Learning for control tasks benefit from representations that mix of low and high-level features.

Modeling *multi-frame* contexts (easy with Voltron) is also high-impact!

Evaluation: Franka Kitchen & Adroit Manipulation domains from R3M – CC @aravindr93 @Vikashplus. (8/12)
Really cool is how we can use the generative language model zero-shot, with no extra data.

Given a video & language intent, we can score – in real time – how well the behavior in the video captures the intent.

Transfers to *robot data* – no robots during pretraining! (9/12)
But don't take our word for it – try out our representations yourself... or evaluate your own!

Models & Pretraining: https://t.co/NOB3cpATYG
Evaluation Suite: https://t.co/aOzQu95J8z

Use our models: `pip install voltron-robotics` (10/12)
This project was a huge endeavor; one that would not have been possible without amazing collaborators and mentors – @SurajNair_1 @_anniechen_ @tkollar @chelseabfinn @DorsaSadigh and @percyliang.

Further thanks to @ToyotaResearch, @stanfordnlp, and the @StanfordAILab ! (11/12)
I'm really excited to see the impact of language on shaping representations for robotics... but this isn't the end. The hard parts of robotics remain hard.

Voltron is a building block – a tool. I can't wait to see how y'all use it. Thanks folks – and stay tuned 🤖🚀! (12/12)

More from All

🌺श्री गरुड़ पुराण - संक्षिप्त वर्णन🌺

हिन्दु धर्म के 18 पुराणों में से एक गरुड़ पुराण का हिन्दु धर्म में बड़ा महत्व है। गरुड़ पुराण में मृत्यु के बाद सद्गती की व्याख्या मिलती है। इस पुराण के अधिष्ठातृ देव भगवान विष्णु हैं, इसलिए ये वैष्णव पुराण है।


गरुड़ पुराण के अनुसार हमारे कर्मों का फल हमें हमारे जीवन-काल में तो मिलता ही है परंतु मृत्यु के बाद भी अच्छे बुरे कार्यों का उनके अनुसार फल मिलता है। इस कारण इस पुराण में निहित ज्ञान को प्राप्त करने के लिए घर के किसी सदस्य की मृत्यु के बाद का समय निर्धारित किया गया है...

..ताकि उस समय हम जीवन-मरण से जुड़े सभी सत्य जान सकें और मृत्यु के कारण बिछडने वाले सदस्य का दुख कम हो सके।
गरुड़ पुराण में विष्णु की भक्ति व अवतारों का विस्तार से उसी प्रकार वर्णन मिलता है जिस प्रकार भगवत पुराण में।आरम्भ में मनु से सृष्टि की उत्पत्ति,ध्रुव चरित्र की कथा मिलती है।


तदुपरांत सुर्य व चंद्र ग्रहों के मंत्र, शिव-पार्वती मंत्र,इन्द्र सम्बंधित मंत्र,सरस्वती मंत्र और नौ शक्तियों के बारे में विस्तार से बताया गया है।
इस पुराण में उन्नीस हज़ार श्लोक बताए जाते हैं और इसे दो भागों में कहा जाता है।
प्रथम भाग में विष्णुभक्ति और पूजा विधियों का उल्लेख है।

मृत्यु के उपरांत गरुड़ पुराण के श्रवण का प्रावधान है ।
पुराण के द्वितीय भाग में 'प्रेतकल्प' का विस्तार से वर्णन और नरकों में जीव के पड़ने का वृत्तांत मिलता है। मरने के बाद मनुष्य की क्या गति होती है, उसका किस प्रकार की योनियों में जन्म होता है, प्रेत योनि से मुक्ति के उपाय...

You May Also Like

The entire discussion around Facebook’s disclosures of what happened in 2016 is very frustrating. No exec stopped any investigations, but there were a lot of heated discussions about what to publish and when.


In the spring and summer of 2016, as reported by the Times, activity we traced to GRU was reported to the FBI. This was the standard model of interaction companies used for nation-state attacks against likely US targeted.

In the Spring of 2017, after a deep dive into the Fake News phenomena, the security team wanted to publish an update that covered what we had learned. At this point, we didn’t have any advertising content or the big IRA cluster, but we did know about the GRU model.

This report when through dozens of edits as different equities were represented. I did not have any meetings with Sheryl on the paper, but I can’t speak to whether she was in the loop with my higher-ups.

In the end, the difficult question of attribution was settled by us pointing to the DNI report instead of saying Russia or GRU directly. In my pre-briefs with members of Congress, I made it clear that we believed this action was GRU.
1/12

RT-PCR corona (test) scam

Symptomatic people are tested for one and only one respiratory virus. This means that other acute respiratory infections are reclassified as


2/12

It is tested exquisitely with a hypersensitive non-specific RT-PCR test / Ct >35 (>30 is nonsense, >35 is madness), without considering Ct and clinical context. This means that more acute respiratory infections are reclassified as


3/12

The Drosten RT-PCR test is fabricated in a way that each country and laboratory perform it differently at too high Ct and that the high rate of false positives increases massively due to cross-reaction with other (corona) viruses in the "flu


4/12

Even asymptomatic, previously called healthy, people are tested (en masse) in this way, although there is no epidemiologically relevant asymptomatic transmission. This means that even healthy people are declared as COVID


5/12

Deaths within 28 days after a positive RT-PCR test from whatever cause are designated as deaths WITH COVID. This means that other causes of death are reclassified as
Great article from @AsheSchow. I lived thru the 'Satanic Panic' of the 1980's/early 1990's asking myself "Has eveyrbody lost their GODDAMN MINDS?!"


The 3 big things that made the 1980's/early 1990's surreal for me.

1) Satanic Panic - satanism in the day cares ahhhh!

2) "Repressed memory" syndrome

3) Facilitated Communication [FC]

All 3 led to massive abuse.

"Therapists" -and I use the term to describe these quacks loosely - would hypnotize people & convince they they were 'reliving' past memories of Mom & Dad killing babies in Satanic rituals in the basement while they were growing up.

Other 'therapists' would badger kids until they invented stories about watching alligators eat babies dropped into a lake from a hot air balloon. Kids would deny anything happened for hours until the therapist 'broke through' and 'found' the 'truth'.

FC was a movement that started with the claim severely handicapped individuals were able to 'type' legible sentences & communicate if a 'helper' guided their hands over a keyboard.