Alibaba announces 'Qwen-Robot Suite,' a suite of AI models for automated robot operation.
Qwen, the AI research team of Alibaba, a major Chinese technology company, announced the ' Qwen-Robot Suite, ' a suite of AI models for robot development, on June 16, 2026. The Qwen-Robot Suite includes three types of models: ' Qwen-RobotNav ' for target tracking and autonomous driving, ' Qwen-RobotManip ' for controlling robot arms (manipulators), and ' Qwen-RobotWorld ' for robot world recognition.
Qwen-Robot Suite: A Foundation Model Suite for Physical World Intelligence
Qwen-RobotNav: A Scalable Navigation Model Designed for an Agentic Navigation System
https://qwen.ai/blog?id=qwen-robotnav
Qwen-RobotManip: Alignment Unlocks Scale for Robotic Manipulation Foundation Models
https://qwen.ai/blog?id=qwen-robotmanip
Qwen-RobotWorld: Boundless Worlds for Embodied Agents
https://qwen.ai/blog?id=qwen-robotworld
◆Qwen-RobotNav
Qwen-RobotNav is an AI model designed to automatically control robots to reach their destinations. Based on Qwen3-VL and trained with 15.6 million samples, it can perform navigation tasks such as 'reaching a certain point' and 'tracking a certain target' with high accuracy.
By using Qwen-RobotNav, robots can follow natural language instructions even in unfamiliar locations.
Its high performance has been confirmed in various benchmark tests.
◆Qwen-RobotManip
Qwen-RobotManip is an AI model for controlling manipulators. It is based on Qwen3.5-4B VL and trained with over 38,100 hours of data.
Manipulators vary in arm length, number of joints, and other specifications from product to product, but Qwen-RobotManip can handle manipulators with different structures and enables high-precision operation.
◆Qwen-RobotWorld
Qwen-RobotWorld is a model developed to improve robots' world perception capabilities. Its visual language model allows it to represent the world in natural language.
Qwen-RobotWorld enables operations such as 'generating manipulator movements based on videos of human manual work' and 'generating multi-angle videos that are consistent both temporally and spatially based on a single instruction.' Qwen-RobotWorld is expected to be useful in the development of physical AI.
Related Posts:
in AI, Posted by log1o_hf