AgiBot Launches World's First Open Source Video Platform for Robots

Published on Aug 14, 2025.
AgiBot Launches World's First Open Source Video Platform for Robots

Shanghai-based robotics start-up AgiBot has made a significant advancement in robotic control with the launch of Genie Envisioner, or GE, a novel platform designed to enhance training through unified video generation.

This innovative platform integrates prediction, policy learning, and neural simulation within a single video-generative framework, marking it as the first of its kind in the robotics industry.

According to the company, the philosophy behind Genie Envisioner emphasizes the importance of world models in robotics learning, acting, and evaluating in a cohesive loop.

In a recent communication on social media, AgiBot articulated, 'We're releasing Genie Envisioner: a unified, video generative platform that integrates prediction, policy learning, and neural simulation together,' showcasing their commitment to pioneering robotics technology.

The Genie Envisioner provides a strong foundation for constructing general-purpose, instruction-driven embodied intelligence, and AgiBot plans to make all related code, models, and benchmarks open source.

This platform's vision-centric approach to world modeling is expected to revolutionize robot learning by transitioning from a passive to an active framework of 'imagine-verify-act,' according to AgiBot.

The research team is focused on expanding sensor capabilities to improve full-body mobility and enhance collaboration between humans and robots, which is vital for advancing intelligent manufacturing and service robotics.

Traditional systems for training robots often rely on disjointed stages of data collection, training, and evaluation; however, GE aims to unify these processes within a single platform for greater efficiency.

At the core of Genie Envisioner is GE-Base, a large-scale video diffusion model trained on roughly 3,000 hours of video language-paired data, mapping language instructions to an embodied visual space.

Extensive testing has demonstrated the system's improved capabilities, particularly in task planning for practical activities such as folding clothes and sorting items on conveyor belts.

At the World Robot Conference in Beijing, robots equipped with artificial general intelligence showcased their ability to perform complex tasks with a success rate exceeding industry norms, reinforcing AgiBot's technological edge.

Industry observer Zhong Xiangyun commented on the significance of AgiBot's GE platform, stating that it lays a robust foundation for instruction-driven embodied intelligence and heralds a new era in robotics.

TECHNOLOGYINNOVATION

Read These Next