In a significant stride for the robotics industry, Tencent recently launched Tairos, an embodied intelligence open platform designed to accelerate the development of intelligent robots through modular software solutions. This platform, the first of its kind in China, offers comprehensive model layers, development tools, and data services, enabling hardware manufacturers to integrate advanced embodied intelligence capabilities seamlessly. Dr. Zhang Zhengyou, Chief Scientist at Tencent and Director of the Robotics X Laboratory and Futian Laboratory, emphasized that Tairos aims to bridge the gap between hardware and software, fostering a new paradigm of “body-mind integration” and “software-hardware synergy” for embodied robots.
During an exclusive interview, Dr. Zhang highlighted that embodied intelligence represents a transformative approach to robotics, where machines perceive, plan, and act in dynamic environments. “The model layer of Tairos includes multi-modal perception models, planning large models, and perception-action joint large models, which function as the right brain, left brain, and cerebellum for embodied robots,” he explained. “This allows robot hardware manufacturers to select components based on their specific needs, enhancing the overall intelligence of embodied systems.” The launch of Tairos underscores Tencent’s commitment to advancing embodied intelligence without directly entering the hardware market, instead focusing on empowering the ecosystem through open software innovations.
Key Insights from Dr. Zhang Zhengyou on Embodied Intelligence and Industry Challenges
- Software-Hardware Disparity: Dr. Zhang noted that while robot hardware capabilities are advancing rapidly, software development for embodied intelligence lags due to higher complexity and investment requirements. Based on surveys of over 60 to 70 Chinese robotics companies, his team found that many startups excel in hardware but struggle with AI integration. Tairos addresses this by providing accessible embodied intelligence software, such as perception and action models, to accelerate industry growth.
- Evolution of Robot Capabilities: He illustrated progress with examples, such as a robot following a “come to my side” command by understanding spatial context and planning a path. However, challenges remain in areas like force and tactile perception, which are crucial for adapting to unpredictable environments like icy surfaces. Dr. Zhang stressed that embodied intelligence must evolve beyond visual sensing to include multi-modal inputs for robust performance.
- Focus on Software Openness: Tencent’s strategy centers on making embodied intelligence software platform-based and open-source, rather than building robots itself. This approach, dubbed “body-mind integration without barriers,” aims to democratize access to advanced AI tools, helping developers overcome hurdles in model training and real-world application.
Dr. Zhang shared anecdotes from Tairos’s demonstration, where an embodied robot from Yushu Robotics, integrated with the platform, not only executed commands but also displayed emergent behaviors, like humorously declining a drink offer by saying it was “too busy working overtime.” These instances highlight the growing sophistication of embodied intelligence, though Dr. Zhang cautioned that the field is still in its infancy. “Embodied intelligence requires robots to interact with the 3D world in ways that go beyond text-based understanding, which large language models excel at,” he said. “Developing accurate 3D world models for embodied robots is a nascent area that demands industry-wide collaboration.”
Tencent’s Robotics X Laboratory: Pioneering Embodied Robot Prototypes and Future Directions
Since founding the Robotics X Laboratory in 2018, Dr. Zhang has led the development of diverse embodied robot prototypes, each exploring different aspects of embodied intelligence. These include:
- Max: A multi-modal quadruped robot demonstrating advanced mobility and perception.
- Ollie: A wheel-legged robot combining legged obstacle traversal with wheeled efficiency.
- TRX-Hand and TRX-Arm: Self-developed dexterous hands and robotic arms enabling precise manipulation.
- Xiao Wu (Little Five): A human-environment robot prototype designed for elderly care scenarios, featuring composite wheel-leg design, large-area tactile skin, multi-fingered dexterous hands, and safe human-robot physical interaction systems.
Dr. Zhang expressed particular interest in complex human environments, such as smart elderly care, where embodied robots must navigate safety and robustness challenges. “Human environments are highly demanding, driving us to push the boundaries of embodied intelligence and core technologies like tactile sensors,” he noted. However, he questioned the predominance of bipedal humanoid forms, suggesting that more efficient morphologies might be better suited for predominantly flat terrains. “Limiting ourselves to bipedal standards restricts imagination; we are exploring optimal forms for embodied robots in human habitats,” he added.
Reflecting on his decade-long vision for the laboratory, Dr. Zhang acknowledged the difficulties in applications like elderly care but remained optimistic. “We have three years left in our initial plan, and I hope to see progress in areas like elderly support. This is why we didn’t dwell on products like robotic dogs—we have broader goals for embodied intelligence,” he said. The laboratory’s work emphasizes research-driven prototyping over productization, as Dr. Zhang believes that rigid deadlines can lead to “hardcoded” solutions that hinder innovation in embodied intelligence.
Industry Landscape and the Role of Public Enthusiasm in Advancing Embodied Intelligence
Dr. Zhang addressed the current state of the embodied intelligence sector, describing it as “just starting out” and in a primary phase. He highlighted several challenges that require collective effort:
- Simulation and Model Refinement: Improving simulation environments to accurately replicate tactile sensations, such as human-like spatial resolution of 1 millimeter, remains a hurdle. Additionally, enhancing model capabilities and data collection processes is critical for embodied robots to perceive and act in real-world settings.
- 3D World Modeling: Unlike text-based models, embodied robots need deep 3D world understanding that isn’t merely about converting images to 3D. Dr. Zhang called for more enterprises and research institutions to collaborate on this frontier, as no single entity can tackle it alone.
- Economic Realities: He acknowledged that some companies might fail due to technological immaturity, but viewed this as a natural part of technological evolution. The openness of platforms like Tairos could help mitigate risks by providing foundational tools.
Despite these challenges, Dr. Zhang welcomed the high public enthusiasm for embodied intelligence and embodied robots. “Investors and industry professionals have a clear understanding of what is feasible and the current development stage. Public expectations for robots in daily life and production are a positive force, offering us more testing grounds and driving higher standards,” he remarked. This optimism, he believes, will fuel innovation and eventually lead to more integrated embodied systems in society.
In conclusion, Dr. Zhang reaffirmed Tencent’s focus on evolving model abilities and simulation environments through Tairos, while avoiding direct competition in hardware. “The most important thing is how we use our current capabilities to contribute and make the embodied intelligence ecosystem more robust,” he said. As the industry progresses, embodied intelligence and embodied robots are poised to redefine human-robot interactions, with Tairos serving as a catalyst for this transformative journey.