AgiBot GE-Act 2.0: A World Action Model Trained From Scratch
Loading video
Loading videoAgiBot released GE-Act 2.0, a native world action model trained entirely from scratch on embodied manipulation data instead of inheriting a pretrained video generator. Scaling data from 300 to 30,000 hours lifted overall success on G1-OP from 17.1% to 44.1%.
World-action modelsVLARobot ManipulationZero-ShotData ScalingPretrainingAgiBotWorld ModelsEmbodied AIAgiBot Robotics
Category: world-model
Author: @Chinazhidx
Date: 2026-09-09T00:00:00
Duration: 228.633s
Reference: https://arxiv.org/abs/2609.05588
