Core Insights - Ant Group's Lingbo Technology announced the open-source release of the embodied world model LingBot-VA, which integrates large-scale video generation models with robotic control through a self-regressive video-action world modeling framework [1][3] - LingBot-VA demonstrates strong adaptability to complex physical interactions, achieving a 20% average improvement in task success rates compared to the industry baseline Pi0.5, requiring only 30 to 50 real machine demonstration data for adaptation [1][3] Summary by Categories Product Development - LingBot-VA is part of a series of releases including LingBot-World (simulation environment), LingBot-VLA (intelligent base), and LingBot-Depth (spatial perception), exploring a new path of "world model empowering embodied operations" [2][4] Industry Collaboration - Ant Group aims to leverage the InclusionAI community for open-source collaboration, focusing on building foundational capabilities for embodied intelligence and accelerating the development of a deeply integrated open-source AGI ecosystem that serves real industrial scenarios [2][4]
全球首个自回归视频-动作世界模型，蚂蚁灵波开源LingBot-VA
Xin Lang Cai Jing·2026-01-30 02:38