Tencent Robotics X Lab Unveils Tairos: China’s First Modular Embodied Intelligence Open Platform

CNBC AI News – July 27 — Tech giant Tencent’s Robotics X Lab and the Futian Lab have unveiled Tairos, a modular embodied AI open platform marking a significant stride in China’s robotics landscape. This platform, centered around large-scale AI models, integrates development tools and data services, offering a plug-and-play approach for the robotics industry.

Zhang Zhengyou, Tencent Chief Scientist and Director of both labs, likened Tairos to equipping robots with a “brain,” empowering them with the ability to perceive their environment, plan tasks, and make autonomous decisions. This, he argues, propels robots beyond mere mechanical execution to intelligent entities capable of adapting to real-world scenarios.

Tencent Robotics X Lab, one of the few teams in China with full-stack robotics capabilities, has previously introduced robots like Max and Ollie, alongside core components such as the TRX-Hand and TRX-Arm. The Tairos platform represents the culmination of the lab’s deep expertise at the intersection of robotics and AI, now amplified by the power of large language models.

At the recent WAIC 2025 Tencent Forum, the platform announced its initial partners – a diverse group including Yuejiang Technology, Leju Robot, Parsini Perception Technology, Keenon Robotics, Zongqing Robotics Technology, and Unitree Robotics. Tairos aims to foster broad collaboration, inviting both robotics hardware companies and algorithm model developers to contribute and innovate on the platform.

Before its public launch, Tairos underwent extensive trials and testing, exploring real-world applications in collaboration with various robotics firms across industries like manufacturing, automotive, and home appliances. At the WAIC 2025 Tencent showcase, a Unitree humanoid robot G1 powered by Tairos served as a “tour guide,” while the “Xiao Wu” home environment robot demonstrated its ability to understand and respond to human language, offering visitors hugs. A Yuejiang robotic arm, integrated with Tairos, even prepared a meal at a verbal command, showcasing the platform’s potential for intuitive human-robot interaction.

Technically, Tairos consists of two main components: model algorithms and cloud services. The model foundation features a multi-modal perception model, a planning large model, and a perception-action joint large model. The cloud service platform includes a simulation platform, data platform, and development tools, all accessible through standardized API calls and software development kits (SDKs).

The multi-modal perception model gives the robot a comprehensive understanding of its surroundings, akin to a human’s right brain hemisphere. By processing inputs from “vision” and “touch,” this model constructs a three-dimensional hierarchical map of the environment in real-time. This information can be accessed through natural language queries, enabling sophisticated interaction and task planning.

The planning large model, analogous to the left brain, allows the robot to understand its capabilities and the goals of a given task, breaking it down into actionable steps. In practice, Tairos’s large language model equips embodied AI agents with reasoning, cognitive, reflective, and external tool-calling abilities. Tencent claims up to 80% success rate in complex human-machine interaction tasks, surpassing leading closed-source agent systems like GPT4o and o1.

The perception-action joint large model bridges the gap between understanding an environment and acting within it. This model, similar to the cerebellum, allows the robot to translate environmental awareness into precise and safe actions. Through core technologies such as spatio-temporal physical perception, language-visual pixel-level mapping, and zero-shot generalization, it enhances the robot’s ability to understand environments, plan tasks, and execute decisions in complex scenarios.

Integrating Tencent Games’ Genesis-Scene 3D scene asset generation technology, the Tairos cloud simulation environment provides a multitude of scenarios tailored for different models. These high-precision reproductions of real-world environments, combined with 20 built-in interaction actions and 62 operable objects, enable advanced displacement, object decomposition, and basic navigation features.

Looking ahead, the Tairos platform will leverage cloud computing resources to offer developers a comprehensive, one-stop solution for embodied AI research and development — from data collection and annotation to algorithm training and model validation, and ultimately, one-click deployment to the robot itself. This ambitious effort is expected to significantly accelerate the real-world application of embodied AI technologies.