Visited the University of Tokyo for the #RSS2024 area chair meeting. Paper decisions have been made and will be announced on Monday.

I will attend #ICRA2024 in Yokohama next week. Looking forward to connecting with the Japanese robotics community, particularly on humanoid…

Our Eureka follow-up work is out!

Got a taste of Tesla's FSD v12.3.4 last night. By no means flawless, but the human-like driving maneuvers (with no interventions) delivered a magical experience. Excited to witness the recipe of scaling law and data flywheel for full autonomy show signs of life in real products.…

Revisiting David Epstein's book Range: Why Generalists Triumph in a Specialized World as a roboticist unveils a compelling insight: Elite athletes usually start broad and embrace diverse experiences as a generalist prior to delayed specialization.

Given this prevailing pathway…

Check out our work on data-efficient imitation learning with behavior primitives led by Tian Gao!

📢Update announced in today’s #GTC2024 Keynote📢

We are working on Project GR00T, a general-purpose foundation model for humanoid robots. GR00T will enable the robots to follow natural language instructions and learn new skills from human videos and demonstrations.

Generalist…

I will give a talk at #SXSW2024 on How to Train a Humanoid Robot tomorrow from 10 to 11:30 a.m. Come to check out our ramen-cooking DRACO 3 robot developed Texas Robotics and learn the technical stories behind it!

My #CoRL2023 keynote talk on Pathway to Generalist Robots is on YouTube now: youtu.be/hFPZSBnJeLc

I discussed the three key ingredients for building general-purpose robot autonomy: scaling law, data flywheel, and human-like embodiment.

If you want to learn more about our…

Thrilled to co-lead this new team with my long-time collaborator Jim Fan. We are on a mission to build transformative breakthroughs in the landscape of Robotics and Embodied Agents. Come join us and shape the future together!

We are unlikely to create an “ImageNet for Robotics”. In retrospect, ImageNet is such a homogeneous dataset. Labeled images w/ boxes.

Generalist robot models will be fueled by the Data Pyramid, blending diverse data sources from web and synthetic data to real-world experiences.

One of RL's most future-proof ideas is that adaptation is just a memory problem in disguise. Simple in theory, scaling is hard!

Our #ICLR2024 spotlight work AMAGO shows the path to training long-context Transformer models with pure RL. Open-source here: github.com/UT-Austin-RPL/…

Debates abound on the necessity of a human-like form factor for generalist robots. Humanoid robots are overkill now, but they make sense from first principles. A deep tech will always be overkill until we put hard work into it. Why not stop debating and shape the future together?

Most of the impressive robot demos are done by remote control. How to reduce reliance on humans? This paper designs a very useful runtime monitoring system that predicts failure preemptively and notifies humans in advance to intervene. Kind of like 'Minority Report' but for real…

We'll witness more and more demos of humanoid robots doing the same tasks the robotics community has mastered with simpler systems. Yet people will still be awed.

It speaks more about human psychology than technology. Humanoids make sense in domains requiring social interaction.

Check out our new survey paper on Foundation Models in Robotics!