TARS Unveils Innovative AWE Model at WAIC 2026, Celebrated with SAIL Award

TARS's Innovations at WAIC 2026

At the recent World Artificial Intelligence Conference (WAIC) 2026 held in Shanghai from July 17 to 20, TARS unveiled its latest advancements in robotic technology under the theme of 'Trustworthy Physical AI'. Leading the charge was the AWE (AI World Engine) embodied foundation model, which was honored with the prestigious SAIL (Superior AI Leader) Award for its contributions to technical innovation and industrial potential.

Introducing AWE 3.5

During the conference, TARS Founder and CEO Dr. Chen Yilun showcased AWE 3.5, a greatly advanced version of the company's embodied-native foundation model. This model benefits from over one million hours of human-centric, real-world data collected from industrial applications. The AWE 3.5 model integrates multiple dimensions of functionality, namely action, perception, geometry, and tactile sensing, all within a cohesive framework. Compared to its predecessor model, Pi 0.5, AWE 3.5 boasts nearly double the efficiency in task execution and improved performance in managing complex tasks while ensuring effective closed-loop interactions over longer durations.

New Methods and Enhanced Learning

AWE 3.5 is notable for being the first to successfully implement a 'pre-training + post-training' approach tailored for embodied-native settings, which facilitates reproducible, scalable, and continuously iterative model development. This groundbreaking advancement positions TARS ahead, setting the stage for improved generalization and effective real-task execution across varied environments.

The demonstration included AWE-powered robots successfully completing tasks such as packing phones, organizing backpacks, and sorting screws with precision, effectively illustrating how the model translates theoretical understanding of physical interactions into practical and deployable actions with the ability to generalize to new objects and scenarios.

Expanding the Data Pool

Looking ahead, TARS plans to increase its pre-training dataset from one million hours to ten million hours by the end of 2026. This massive expansion aims to further enhance the model's ability to generalize across varying contexts, improve reasoning over extended time frames, and execute real-world tasks more effectively.

Highlighting Smart Manufacturing

One of the standout exhibits at WAIC was TARS's recreation of a full-scale circular production line specifically designed for automotive wiring harnesses. Leveraging multiple A1 robots, the demonstration showcased sophisticated capabilities in gr ASP and assembly of flexible wiring components. This representation featured as part of WAIC’s dedicated exhibition area focusing on embodied robotics within the 'Smart Manufacturing Hub'. Here, the emphasis was on stability and reliability—key factors for successful integration into advanced manufacturing processes.

Furthermore, TARS is collaborating with the Jiading District in Shanghai alongside industry partners to comprehensively validate the technology, aiming towards large-scale applications within the country's first cluster of one thousand embodied-robot units.

A Live Performance Featuring DexHand

Another exciting moment came from the live performance debut of DexHand, which was mounted on an A1 robot. The performance, conducted with magician Deng Nanzi, showcased both creativity and technical prowess. DexHand executed intricate tasks such as card shuffling, handwriting, and even solving a Rubik's Cube. This live demonstration highlighted the advanced real-time perception and precision control capabilities of TARS’s technology and affirmed the value of effective human-robot collaboration.

Future Outlook

The integration of AWE 3.5, the showcased industrial production line, and the impressive performance of DexHand represents TARS's comprehensive vision aimed at achieving trustworthy physical AI. This encompasses not only a deep understanding of complex environments but also delivering tangible value in real-world production scenarios and executing high-precision tasks.

As TARS moves forward, its commitment to transitioning from demonstration to real-world application underlines its strategic vision to promote the broad adoption of embodied intelligence within various industries. With continuous innovation at its helm, TARS is steering toward an era of transformative AI applications.