On July 27th， according to media reports， Tencent’s Robotics X Lab， in collaboration with the Futian Laboratory， recently launched Tairos， China’s first modular embodied intelligence open platform. This platform， powered by large models， integrates development tools and data services， offering a plug-and-play solution for the robotics industry. (Official website: [insert website if available])
Zhang Zhengyou， Chief Scientist at Tencent and director of both laboratories， explained that Tairos essentially equips various robots with a “brain，” enabling them to perceive their environment， plan tasks， and make autonomous decisions. This advancement is crucial for evolving robots from mere instruction-following machines into intelligent entities capable of actively adapting to reality.
As one of the few domestic teams with full-stack robotics capabilities， Tencent’s Robotics X Lab has previously introduced robots like Max and Ollie， as well as core components such as the TRX-Hand dexterous hand and the TRX-Arm robotic arm. The release of the Tairos platform represents the latest achievement from the lab’s long-term dedication to robotics and AI， leveraging large model technology to drive industry upgrades.
At the WAIC 2025 Tencent Forum， the initial list of platform partners was unveiled， including prominent companies such as UBTECH Robotics， Unitree Robotics， Exosapien Technology， Lean Robotics， and Q-Link Robotics. Moving forward， Tairos will be open for collaboration not only with all robot hardware companies but also support the integration of algorithms and models from various fields， encouraging developers to freely combine and innovate.
Prior to its official release， Tairos underwent extensive trials and testing， with several robot enterprises exploring its application in various industries， including industrial automation， automotive， and home appliances. At the WAIC 2025 Tencent exhibition， Unitree’s G1 humanoid robot， powered by the Tairos embodied intelligence open platform， served as a guide for attendees. The “Little Five” household robot demonstrated its ability to understand human language and engage in physical interactions like hugs. Furthermore， the Exosapien X-Trainer robotic arm， integrated with Tairos， showcased impressive capabilities， such as preparing meals upon simple verbal commands， highlighting the platform’s versatility.
Specifically， the Tairos platform comprises two main components: model algorithms and cloud services. The model layer includes a multimodal perception model， a planning large model， and a perception-action joint large model. The cloud service platform encompasses a simulation platform， a data platform， and development tools. The platform provides services through standardized interface calls and Software Development Kits (SDKs).
The multimodal perception model can be likened to the right brain of a human， enabling robots to truly understand their environment. Through senses like “vision” and “touch，” robots can perceive their surroundings， organizing environmental data into a three-dimensional hierarchical map in real-time. This information can then be accessed via natural language queries， providing ample data for robot interaction and task planning.
The planning large model， analogous to the left brain， empowers robots to comprehend their roles and capabilities， define task objectives， and break them down into executable strategic steps. In practical applications， Tairos’ large-scale models equip embodied intelligent agents with the ability to reason， cognize， reflect， and utilize external tools. In complex， long-sequence human-robot interaction tasks， the platform has achieved an 80% success rate， surpassing leading closed-source agent systems like GPT-4o and O1.
The perception-action joint large model functions like the cerebellum， bridging the critical gap between “seeing” and “doing.” It not only allows robots to understand their environment but also translates this understanding into safe and accurate action commands. Through three core technologies—spatio-temporal physical perception， language-vision pixel-level mapping， and zero-shot generalization—it significantly enhances robots’ environmental comprehension， task planning， and decision-making execution capabilities in complex scenarios.
In conjunction with Tencent Games’ R&D product department’s Genesis-Scene 3D scene asset generation technology， the Tairos cloud simulation environment has generated multiple scene assets for different models， precisely replicating real-world environments. The platform also includes 20 interactive actions and 62 operable objects， supporting advanced features such as high-precision displacement， object decomposition， and basic navigation.
Moving forward， the Tairos platform will integrate cloud computing resources to provide developers with a one-stop R&D solution. This will encompass robot本体 (benti – physical body) data collection and labeling， algorithm training and model validation， and one-click model deployment to robots， thereby accelerating the practical implementation of embodied intelligence in the industry.
