UNITAS: A 3D-Native World Action Model for Embodied Manipulation

2026-10-08

AI Summary

Existing world action models predominantly rely on 2D imagery, limiting their capacity to accurately encode physical distances. This work proposes the first 3D-native world action model that unifies observation, action, and scene dynamics modeling within a metric space. Methodologically, it employs 3D point trajectory flows, a physical-time trajectory tokenizer, and world-aligned positional embeddings to achieve unified representations across diverse embodiments and enable action-conditioned scene prediction. Experimental evaluations demonstrate that the proposed model attains a 99.8% success rate on the LIBERO benchmark and an average success rate of 85% in real-world tasks, while reducing displacement errors by 49%.

Abstract

World action models (WAMs) aim to answer a coupled physical question: given a task instruction, what motion should the robot execute, and how will that motion change the surrounding world? Most existing WAMs build on pretrained video generators and represent world evolution through images or visual latents. Robotic interaction, however, takes place in metric three-dimensional space, while images are view-dependent projections whose pixel distances do not directly encode physical distances. We introduce UNITAS, to our knowledge the first 3D-native world action model that unifies observations, actions, and scene dynamics in a shared metric 3D frame within each interaction, using a common representation across robot embodiments and human hands. Action flow represents human hands and robot grippers as 3D point trajectories, while scene flow describes scene-point displacements conditioned on these trajectories. World-aligned 3D positional embeddings ground visual tokens with or without depth input, and a physical-time trajectory tokenizer encodes each point trajectory as one token anchored at its current 3D position. This interface supports both direct action execution and action-conditioned scene prediction. With 1.7B parameters, UNITAS achieves the best action-conditioned scene prediction on RoboTwin among the compared methods, with up to 49% lower displacement errors than PointWorld, and state-of-the-art manipulation success, including 99.8% on LIBERO and an average of 85% across real-world tasks. The code is available at https://github.com/DexForce/UNITAS.

Problem

• World Action Model
• 3D-native representation
• Embodied manipulation
• Scene dynamics
• Metric 3D space

Innovation

• 3D-native world action model
• action flow
• scene flow
• world-aligned 3D positional embeddings
• trajectory tokenizer