TARS Launches at WAIC 2026 with AWE Foundation Model Winning the Prestigious SAIL Award
TARS Launches at WAIC 2026
TARS recently made its debut at the World Artificial Intelligence Conference (WAIC) 2026, unveiling its groundbreaking vision for a "trustworthy physical AI". Held from July 17 to 20 at the Shanghai World Expo Exhibition and Convention Center, the event highlighted TARS's latest foundation model, AWE (AI World Engine), which clinched the prestigious SAIL (Superior AI Leader) award for its technological innovation.
During the main forum of WAIC, Dr. Chen Yilun, the founder and CEO of TARS, introduced AWE 3.5, the newest version of the company's embodied-native foundation model. Drawing on over a million hours of human-centered real-world data, AWE 3.5 incorporates action, perception, geometry, and tactile sensing into a cohesive framework. In comparison to its predecessor, Pi 0.5, this model approximately doubles task execution efficiency, enhances performance in complex tasks, and maintains a closed-loop interaction during tasks for several minutes. Moreover, AWE 3.5 represents the first embodiment of a "pre-training + post-training" paradigm designed for reproducible, scalable, and continuous iterations.
TARS showcased robots equipped with the AWE technology tackling tasks such as phone packaging, backpack organization, and precise screw sorting. These demonstrations illustrated how the model translates physical understanding into actionable objectives and adapts to unknown objects and scenarios. Following a milestone of over a million hours of high-quality human-centric data, TARS aims to expand its pre-training dataset to 10 million hours by the end of 2026 to further strengthen generalization, long-term reasoning, and real-world task execution.
In addition, TARS recreated a full-scale circular production line for automotive wire harnesses, utilizing multiple A1 robots to grasp, route, connect, and assemble soft wire harnesses. This scenario was selected for the official WAIC exhibition area dedicated to embodied robotics, aptly named "Model Era • Partner City", showcased within the "Smart Manufacturing Hub". This presentation showcased the stability, reliability, and added value of this technology within the domain of advanced manufacturing. Collaborating with the Jiading district in Shanghai and industry partners, TARS is working to validate its technology on a larger scale, aiming to develop the first Chinese industrial hub for embodied robots with a capacity of one thousand units.
Another highlight was the debut appearance of DexHand, mounted on an A1 robot, which performed live with magician Deng Nanzi. DexHand executed card manipulations (spreading and shuffling), handwritten notes, and even solved a Rubik's Cube, demonstrating real-time perception, precise control, and reliable human-robot collaboration.
Together, AWE 3.5, the industrial production line, and DexHand epitomize TARS's three-fold vision for a trustworthy physical AI: to understand complex environments, add value in real production lines, and accomplish high-precision tasks. TARS is committed to advancing embodied intelligence from demonstration to implementation, accelerating its large-scale adoption.