#China #Shanghai #TARS #SAIL Award #AWE Model

TARS Premieres Innovative AI at WAIC 2026

TARS, a leading name in artificial intelligence, proudly introduced its vision for "Trustworthy Physical AI" during the World Artificial Intelligence Conference (WAIC 2026), which occurred from July 17 to July 20 at the Shanghai World Expo Exhibition and Convention Center. This remarkable event saw TARS's AWE (AI World Engine) embodied foundation model clinching the esteemed SAIL (Superior AI Leader) Award, in recognition of its technical innovation and industrial potential.

During the main forum at WAIC, TARS Founder and CEO Dr. Chen Yilun presented AWE 3.5—an advanced embodiment of the company's foundation model. This latest iteration builds upon extensive real-world data, having been trained on over one million hours of human-centric data verified in industrial environments. The AWE 3.5 model integrates various facets of AI, including action, perception, geometry, and tactile sensing, all within a comprehensive framework. Comparison with its predecessor, Pi 0.5, reveals that this updated model nearly doubles task-execution efficiency, enhances performance in complex tasks, and preserves closed-loop interactions for prolonged scenarios.

A crucial innovation of the AWE 3.5 is its pioneering "pre-training + post-training" approach. This paradigm is tailored not only for reproducibility but also supports scalability and ongoing iteration, a testament to TARS's commitment to advancing AI technology.

At the conference, TARS showcased AWE-powered robots performing practical tasks such as phone packing, organizing backpacks, and sorting screws with precision. These demonstrations illustrated how the model can translate physical understanding into feasible actions, even generalizing to unfamiliar objects and situations. TARS aims to expand its pre-training dataset from one million to a staggering ten million hours by the end of 2026, a move anticipated to boost generalization, facilitate long-horizon reasoning, and enhance task execution in real-world applications.

The innovation did not stop there, as TARS also recreated a full-scale circular automotive wiring harness production line operated by multiple A1 robots. These machines deftly grasped, routed, connected, and assembled flexible wiring harnesses, highlighting the reliability and value that this technology brings to advanced manufacturing. This scenario was featured in WAIC's official embodied robotics exhibition zone, dubbed "Model Era • Partner City," within the category of the "Smart Manufacturing Hub." TARS is actively collaborating with the Jiading District in Shanghai along with various industry partners to validate this technology at a larger scale and to realize China's first large-scale industrial embodied-robot cluster, targeting units in the thousands.

Another memorable moment at WAIC 2026 was the live performance of DexHand, mounted on an A1 robot. In a captivating demonstration alongside magician Deng Nanzi, DexHand exhibited an array of skills including card spreading, shuffling, handwriting, and Rubik's Cube solving. This display of real-time perception, high-precision control, and dependable human-robot collaboration captivated the audience and showcased the potential of embodied intelligence in practical applications.

TARS's trio of innovations—AWE 3.5, the industrial production line, and DexHand—represent the company's comprehensive vision for physical AI. By understanding complex environments, delivering substantial value on production lines, and executing high-precision tasks, TARS aims to democratize access to these cutting-edge emotions of AI, paving the way for widespread application in various sectors. The future of embodied intelligence is bright, with TARS committed to transitioning from impressive demonstrations to real-world implementations that promise transformative impacts in industry and society alike.