← Back to Videos

A 3D World Model Learned Without Action Labels

Loading video

mu-0 learns a physical language for robot interaction traces directly from video, eliminating the need for action labels when building a world model.

World Models 3D Interaction Trace Video Learning Robot Learning Action-Free

Category: world-model

Author: @jbhuang0604

Date: 2026-11-09T00:00:00

Duration: 169.002s

Source: https://x.com/jbhuang0604/status/2119816319973400577

Reference: https://mu0-wm.github.io/

PointZero: 3D World Model Pretrained Without Robot Data

PointZero: 3D World Model Pretrained Without Robot Data

@bardienus · 672 views · 2026-09-17

World Models3D Dynamics

OpenWAM open world-action model pretraining framework

OpenWAM open world-action model pretraining framework

@KnightNemo_ · 705 views · 2026-09-09

World ModelsRobot Learning

Lucy 2.6 Sharpens Real-Time Interactive Video

Lucy 2.6 Sharpens Real-Time Interactive Video

@DecartAI · 131 views · 2026-11-05

Interactive VideoWorld Models

Genie 4: A World You Can Steer in Real Time

Genie 4: A World You Can Steer in Real Time

@GoogleAI · 174 views · 2026-10-27

UniAction Unifies Four Action Types in World Models

UniAction Unifies Four Action Types in World Models

@zizhang_li · 197 views · 2026-10-26

World ModelsUnified Action Conditioning

SKY-001 Builds an Engineering Flange in 2.87 Seconds

SKY-001 Builds an Engineering Flange in 2.87 Seconds

@skyfallai · 201 views · 2026-10-22

Engineering World ModelSkyfall AI

As an Amazon Associate, we earn from qualifying purchases.