China is preparing to take a stronger role in the race for embodied AI. Alibaba has introduced Qwen-Robot, a new family of AI models designed to help robots move through the real world, interact with objects, and make decisions based on their surroundings.
As artificial intelligence continues to advance rapidly, robotics is also entering a new phase. Robots designed for different tasks may soon become a more visible part of everyday life. At the center of this transformation is a field known as embodied AI, which focuses on systems that can connect digital intelligence with physical action.
In recent months, companies such as NVIDIA, Google, and Figure AI have made major moves in this area. Now, Alibaba is joining that competition with its own robotics-focused AI model family.
Alibaba Introduces Qwen-Robot for Real-World Robots
Alibaba has unveiled Qwen-Robot, its first embodied AI model family developed for robots that operate in the physical world. Created by the company’s Tongyi Lab research team, the new model family aims to combine the reasoning abilities of large language models with the physical movements of robots.
Qwen-Robot has entered pilot testing with selected enterprise customers through Alibaba Cloud. The system is designed to help robots perceive their environment, make decisions, and complete tasks in real-world settings.
The model family consists of three different systems, each focused on a specific area of robotic intelligence.
Why Embodied AI Is a Major Challenge
Large language models have made impressive progress in understanding text, images, and instructions. However, using that intelligence in the physical world is much more complicated.
For example, an AI model can understand a command such as: “Go to the kitchen, find the red mug, and place it on the shelf.” But understanding the sentence is only the first step. A robot must also analyze the room, identify the target object, plan its movement, avoid obstacles, reach the object, grasp it properly, and complete the task safely.
This is one of the biggest challenges in robotics: closing the gap between digital understanding and physical action.
According to Alibaba, one of the key reasons this problem is difficult is the nature of robotics data. Unlike internet data, which is widely available and relatively easy to organize, robotics data comes from many different sources, including robotic arms, vehicles, sensors, and cameras. These data types often have different formats and are expensive to collect.
Bringing all of this information together into a single system does not always improve performance. In some cases, it can even create new technical problems.
Three Models Power the Qwen-Robot Family
To address these challenges, Alibaba developed three separate models under the Qwen-Robot family.
The first model, Qwen-RobotNav, focuses on movement and navigation. It allows robots to follow instructions, reach specific destinations, track objects, and perform tasks similar to autonomous driving.
The second model, Qwen-RobotManip, is designed for physical interaction. It helps robots grasp, move, and manipulate objects. This model was trained using a broad dataset collected from different robotic platforms.
The third model, Qwen-RobotWorld, functions as a type of world model. It predicts possible changes in the environment and helps robots evaluate the likely results of their actions before carrying them out.
When these three models work together, Alibaba says robots can do more than simply understand commands. They can interact with objects, move through complex environments, and make real-time decisions based on what is happening around them.
Bringing Digital Assistants Into the Physical World
Alibaba’s long-term goal is to bring the logic of digital assistants into the physical world. Instead of only responding to text or voice commands on a screen, future AI systems could use robots as physical agents capable of performing real tasks.
The company’s early demonstrations offer a glimpse of this vision.
In one test, Alibaba used Qwen-RobotNav on Unitree’s four-legged Go2 robot. The robot was powered by NVIDIA’s Jetson Thor hardware and used only a single low-resolution camera. Despite never having seen the apartment before, it was able to follow spoken instructions and move between rooms without relying on a pre-built map.
This is an important step because real-world environments are unpredictable. For robots to become useful outside controlled labs, they must be able to understand unfamiliar spaces and respond to changing conditions.
Alibaba Also Introduces Qwen-RobotClaw
Alongside the model family, Alibaba also introduced a robot agent system called Qwen-RobotClaw. This system allows Qwen models to act more like physical-world agents, using tools and making decisions to complete different tasks.
In one demonstration, an AI-powered agent searches for a restroom, notices a sign showing that it is out of service, and then creates an alternative route to find another restroom.
Alibaba has also released an open-source platform called Chat2Robot, allowing developers to test and experiment with these systems.
China Enters the Embodied AI Race
The launch of Qwen-Robot shows that Chinese technology companies are not only competing in large language models and humanoid robotics, but also in the AI systems that could power the next generation of robots.
As companies in the United States and China continue to invest heavily in embodied AI, the race is shifting beyond chatbots and digital assistants. The next major frontier may be robots that can understand the world, make decisions, and act with greater independence.
Alibaba’s Qwen-Robot family suggests that this future is moving closer.
You Might Also Like;
