Sergey Levine

Sergey Levine

Creator
0 followers

UC Berkeley professor; reinforcement learning for robotics; decision-making

Rethinking What Makes a Good World Model
SocialApr 26, 2026

Rethinking What Makes a Good World Model

Tomorrow in the Workshop on World Models at ICLR in Rio (10:30 am) I’ll talk about a… different take on what might make for a good world model. Come find out, 10:30 in Room 202A at ICLR https://t.co/TRzkSqeeKN

By Sergey Levine
RL Boosts Robot Foundation Models; Generative AI Enables Self‑Improvement
SocialApr 25, 2026

RL Boosts Robot Foundation Models; Generative AI Enables Self‑Improvement

I'll give a talk about lifelong learning tmrw (Sun), 9:30 am (Brazil time) in the lifelong agents workshop, about how we can get robot foundation models to improve with RL: https://t.co/nKci27JeaH Then at 11:30 am, I'll talk about how generative models...

By Sergey Levine
π0.7 Demonstrates Emergent Compositional Generalization via Instructions
SocialApr 16, 2026

π0.7 Demonstrates Emergent Compositional Generalization via Instructions

We finished evaluating π0.7, our new model at Physical Intelligence. What I'm most excited about with π0.7 is that it's starting to show some surprising emergent compositional generalization, being able to both perform complex tasks and learn new tasks just...

By Sergey Levine
Pi Models Power Unexpected Flying Gripper Drones
SocialMar 28, 2026

Pi Models Power Unexpected Flying Gripper Drones

Didn't think that "drones with grippers" was in the cards for a likely embodiment for pi models, but there we have it. It's literally a flying gripper. But believe it or not, pi models have been used on even stranger...

By Sergey Levine
Structured Models Generate Higher‑Reward Material Designs
SocialMar 21, 2026

Structured Models Generate Higher‑Reward Material Designs

A while ago we figured out that structure enables data-driven design: if we have data of designs + rewards, we can find a design with *higher* reward if we learn a structured function: https://t.co/HlevRSTMXV In our latest work, @kuba_AI developed a...

By Sergey Levine
Fast Online RL Upgrades Π-06 with 15‑minute Data
SocialMar 19, 2026

Fast Online RL Upgrades Π-06 with 15‑minute Data

Back in Nov we developed Recap and trained π*-06 with RL. Now, we developed a fast *online* RL method that improves π-06 with as little as 15 min of robot data for precise tasks, using "RL tokens" exposed by our...

By Sergey Levine
Multi-Scale Embodied
SocialMar 3, 2026

Multi-Scale Embodied

We made a memory system for our models at PI. We call it Multi-Scale Embodied Memory (MEM). It provides both short-term and long-term memory to enable very long tasks. We tested it on cleaning a kitchen (and yes, washing dishes),...

By Sergey Levine
Run VLAs in Real Time Using Fast Edge Adapter
SocialFeb 18, 2026

Run VLAs in Real Time Using Fast Edge Adapter

Check out Noriaki's thread about a way to get VLAs to run in real time with a fast "edge adapter"!

By Sergey Levine
VLA Reasoning Lets Vehicles Safely Navigate Complex Edge Cases
SocialFeb 17, 2026

VLA Reasoning Lets Vehicles Safely Navigate Complex Edge Cases

VLAs can enable vehicles to better handle complex edge cases: a VLM can "think through" a complex interaction, deduce a common sense behavior, and then a VLA can carry that out to maintain safe(r) behavior even in unusual situations.

By Sergey Levine