Chinese firm Alibaba has launched its first embodied AI model family, which links large language models with real-world robotic actions.

The Qwen-Robot suite was developed by Alibaba’s Tongyi Lab and is undergoing pilot testing with selected Alibaba Cloud enterprise clients.

The suite comprises three models focused on navigation, manipulation, and world modeling for robots operating in physical environments.

Alibaba said the models enable machines to perceive, reason, and interact with the real world, joining a growing global push to advance embodied AI beyond traditional chatbot applications.

Robots meet reasoning

Alibaba says its Qwen family of AI models has become very good at understanding the physical world. These models can recognize objects, understand spatial relationships, follow complex visual instructions, and reason about real-world environments. For example, a model can understand a command such as, “Go to the kitchen, find the red cup, pick it up, and place it on the shelf.”

However, understanding a task is different from actually performing it. While a vision-language model (VLM) can describe the steps needed to complete a task, it cannot directly control a robot’s movements. The challenge is connecting human language and visual understanding with the motor actions required to interact with the physical world.

This problem is difficult because robot training data is very different from internet data. Information collected from navigation systems, robotic arms, vehicles, and cameras comes in different formats and is expensive to gather. Simply combining all this data often creates conflicts rather than improving performance.

To address this, Alibaba developed the Qwen-Robot Suite, which includes three specialized models. Qwen-RobotNav focuses on movement and navigation. It helps robots follow instructions, navigate to locations, track targets, and support autonomous driving.

According to its website, Qwen-RobotManip focuses on physical interaction. It enables robots to grasp, move, and manipulate objects using a large training dataset collected from different robotic systems. Qwen-RobotWorld acts as a world model, predicting how environments may change and helping robots understand the likely outcomes of their actions.

Together, these models aim to enable robots to understand instructions, interact with objects, navigate environments, and make decisions in the real world.

Physical AI accelerates

Alibaba showcased Qwen-RobotNav on a Unitree Go2 quadruped powered by NVIDIA Jetson Thor hardware and a single low-resolution camera. The robot successfully navigated an unfamiliar apartment, following spoken instructions across multiple rooms without preloaded maps, while maintaining an inference latency of 196 milliseconds.

The company claims that Qwen-RobotManip, its robotic manipulation model, was trained on more than 38,000 hours of open-source data covering object handling and interaction tasks. According to Alibaba, the model recently achieved the highest score in the generalist category of the RoboChallenge real-world robotics benchmark, earning a process score of 59.83 and a task success rate of 45 percent.

The company also unveiled Qwen-RobotClaw, a robotics agent framework that enables Qwen models to use the Qwen-Robot suite as physical-world tools. In one demonstration, an agent searched for a restroom, identified an out-of-order sign, and independently rerouted to another location. Alibaba further open-sourced Chat2Robot, a browser-based platform for testing embodied AI interactions.

As competition in embodied AI intensifies worldwide, Alibaba has expanded its ambitions beyond language and multimodal software with the launch of its Qwen-Robot models. The move reflects a broader industry shift toward creating AI systems capable of understanding and interacting with the physical world.

Alibaba’s move comes as competition in physical AI accelerates globally. In the US, Google DeepMind is advancing Gemini Robotics, while Nvidia is expanding its robotics ecosystem through Cosmos, Isaac, and GR00T. Start-ups, including Physical Intelligence, Skild AI, and Figure AI, are also developing general-purpose robotic intelligence, according to the South China Morning Post.

China is strengthening its position by pairing its manufacturing advantages with growing investments in AI software for autonomous decision-making. The sector now spans AI developers, robotics firms, and EV makers. Companies such as Alibaba, Tencent, Unitree, AgiBot, UBTech, Galbot, Spirit AI, GigaAI, Xpeng, and Xiaomi are actively pursuing embodied AI technologies.

An elephant’s trunk can haul a felled log across a clearing and, moments later, peel a single banana without bruising the fruit. That range — from raw strength to delicate precision — comes from a boneless, all-muscle structure that has fascinated soft robotics researchers for years.

A new study has now mapped exactly why the trunk’s skin enables that range, down to its microscopic architecture.

Researcher Lucia Beccai and colleagues studied trunk tissue obtained from a single adult Asian elephant (Elephas maximus) that died of natural causes at Zurich Zoo, Switzerland, in 2020.

Using 35 tissue samples taken across the specimen, the team combined biomechanical testing, histology, multiple imaging techniques, and computational modeling to characterize the trunk’s skin in unprecedented detail.

A skin split into two functional zones

The central finding is that the trunk’s skin is not a uniform material — it is sharply divided into two distinct mechanical zones, each engineered for an entirely different job.

The upper skin functions like an armor-like shield: tough, protective, and 3.14 times stiffer than the skin on the underside of the trunk. That stiffness makes sense functionally — the top of the trunk is what drags across rough bark, scrapes against the ground, and absorbs impact during heavy lifting and pushing.

The underside skin is the opposite: highly flexible and acutely sensitive, built specifically for grasping. Its low stiffness allows it to conform tightly to the contours of whatever object the elephant is handling — a log, a piece of fruit, another elephant’s trunk in greeting — maximizing contact area between the skin and the object and producing a steady, secure grip without requiring excessive muscular force.

A mechanical lens hidden beneath the surface

The most striking discovery lies just beneath the surface of the underside skin: dome-shaped structures called dermal papillae. Using finite element modeling — a computational technique that simulates how physical structures respond to stress — the researchers found that these papillae function like a subsurface mechanical lens.

Rather than simply cushioning the skin, the dome-shaped papillae concentrate and amplify mechanical stress at precisely the locations where the trunk’s sensory nerves are positioned.

In effect, the structure acts as a biological signal booster — taking a faint mechanical input, such as the texture of a banana peel, and intensifying it before it reaches the nerve endings responsible for interpreting touch.

What it means for robotics

According to the study’s authors, this two-zone, mechanically tuned architecture offers a direct, ready-made blueprint for engineers designing the next generation of soft robotic systems.

The trunk does not need separate “armor” and “sensor” components bolted together — it achieves both functions through a single continuous tissue with engineered variation in stiffness and subsurface geometry.

The practical implication: by using biomimetic multi-material approaches — fabricating robotic grippers from layered materials with deliberately varied stiffness, much like the trunk’s natural upper-lower dichotomy — engineers could build grippers with passive outer protection that simultaneously amplify faint tactile signals, all without exposing delicate sensors directly to mechanical wear and abrasion.

That combination has been a persistent challenge in soft robotic gripper design: sensors sensitive enough to detect subtle touch are often the same components most vulnerable to physical damage from repeated contact.

The elephant trunk’s solution — concentrating sensitivity beneath a protective, graduated-stiffness shield — offers a structural answer that robotics researchers have been seeking for years.

The study was published in the journal PNAS Nexus.

Chinese automaker Seres has unveiled its first humanoid robot, Xiaosai, marking the company’s entry into the rapidly growing embodied AI sector. The robot is designed with visual recognition, autonomous greeting, and voice interaction capabilities.

The launch highlights the automaker’s expanding focus on robotics, with several additional embodied intelligence robots expected to debut later this year as Seres broadens its presence beyond electric vehicles.

Recently, Hyundai announced plans to deploy over 25,000 Atlas humanoid robots in the US, targeting 30,000 annual units by 2028 with locally manufactured robot components.

Industrial humanoid debut

Chinese automakers are rapidly expanding into humanoid robotics, leveraging expertise in manufacturing, artificial intelligence, supply chains, and automation to develop a new generation of embodied intelligence systems.

The latest entrant is Seres, which has unveiled its first humanoid robot, Xiaosai. The announcement was made by Seres Group director and vice president Kang Bo, who released a video showcasing the robot and confirmed that several additional embodied intelligence products are currently under development and scheduled to make their official debut later this year, reports CNEVPost.

Thought more details are not available, the humanoid is reported to be equipped with visual recognition capabilities, can autonomously greet people, and supports voice-based interaction. Its introduction highlights Seres’ broader efforts to integrate advanced robotics into manufacturing operations.

The company’s factory already employs a range of specialized robots across AI-powered quality inspection, logistics delivery, and production-line collaboration. Among them are Xiaosai 01, which performs quality inspections for chassis assembly, and Xiaosai 02, which checks the exterior configuration of completed vehicles.

According to Seres, the facility’s smart hub coordinates more than 1,600 intelligent devices, while over 3,000 industrial robots operate in synchronization throughout the plant. The factory also incorporates technologies, including the Internet of Things (IoT), big data, digital twins, 5G connectivity, and artificial intelligence, creating a highly automated digital ecosystem, reports CNEVPost.

Automakers go the robot way

Seres’ move into humanoid robotics builds on earlier investments in embodied intelligence. In October 2024, the company signed a cooperation agreement with Volcengine, a subsidiary of ByteDance, in Beijing. The partnership focuses on intelligent robot technologies, including decision-making systems, control technologies, and human-machine augmentation, with an emphasis on multimodal cloud-edge collaboration.

The launch reflects a wider trend among Chinese automakers, many of which view humanoid robots as a natural extension of capabilities developed through electric vehicle production and intelligent manufacturing. Xpeng is among the companies accelerating efforts in the sector. Earlier this month, chairman and CEO He Xiaopeng personally assumed the role of “CEO” of the company’s robotics division to speed up commercialization and development.

BYD has also confirmed that it is developing humanoid robots and is exploring the possibility of selling them through its extensive dealer network in the future. Meanwhile, Chery-backed Aimoga has moved beyond development and into commercialization. The company has begun selling humanoid robots directly to consumers, with a retail price of 285,800 yuan (approximately $42,260).

As Chinese automakers continue investing in embodied AI, humanoid robots are increasingly becoming the next frontier of competition, transforming car manufacturers into major players in the rapidly evolving robotics industry.

💾