paper/arxiv/2608.26103

Robotics Dataset

Zero-WAM is a causal video-action model for zero-shot cross-task generalization in robotic manipulation, trained on HumanGen, a dataset of 74.2K human-robot in-context learning pairs across 8.6K tasks, and evaluated on unseen tasks in simulation and real world.

Catalog metadata

Repository owner

Jiaming Zhou, Qihang Zhang, Gangwei Xu et al.

Source

Research paper

Task

Zero-WAM: In-Context World-Action Modeling from Human Videos for Open-Ended Task Generalization

Cameras

0

Browse the catalog to compare robots, tasks, sensors, and licenses. Sign in to open protected download and model links.