Gasgoo Munich- At the 2026 World Artificial Intelligence Conference (WAIC 2026), Spirit AI debuted its new generation of commercial service robots, the Moz2. The launch marks the company's formal entry into sectors such as supermarkets, hospitality, and office buildings, expanding the practical boundaries of its general embodied technology.
Meanwhile, Spirit v1.6 showcased a newly upgraded, spatially-aware long-horizon task demonstration. It achieved multi-step, coherent autonomous operation under a single instruction—"tidy up the living room"—breaking through the industry's widespread bottlenecks regarding short-term memory and segmented execution.
Image Credit: Spirit AI
As Spirit AI's first exploratory product for the commercial service track, Moz2 was built with an AI-native philosophy from the ground up. It serves as the core carrier extending the company's general embodied brain into commercial service scenarios.
Specifically, Moz2 employs a collaborative, AI-native design integrating the "embodied brain," the robot body, and uDAS data acquisition devices. This achieves architectural isomorphism between the hardware and the data acquisition system. Human operation data can be transferred directly and efficiently to the robot, providing the underlying support needed to rapidly port Moz1's mature long-horizon planning and fine manipulation capabilities—enabling the reuse of general abilities across different products.
The unit boasts 32 degrees of freedom, featuring an omnidirectional steering chassis and a folding leg structure. Paired with a 7-degree-of-freedom bionic arm and a 4-degree-of-freedom three-finger dexterous hand, it balances mobility with operational potential. Additionally, the body incorporates multiple magnetic expansion points. These allow for rapid exterior changes to match different brand styles, or the attachment of functional accessories like delivery trays and inspection modules, ensuring high reusability across diverse commercial needs.
Addressing the high-density foot traffic of public spaces, Moz1 is equipped with a multi-modal perception system combining LiDAR, vision, and ultrasonic sensors. Working in tandem with compliant control algorithms, it achieves high-precision mapping and dynamic obstacle avoidance, ensuring operational reliability in environments where humans and robots coexist.
For now, Moz2 will focus on refining human-machine interaction and validating the user experience in commercial settings. Moving forward, it will leverage Spirit AI's general embodied brain technology foundation to gradually migrate the full-stack intelligent capabilities already proven at scale by Moz1. The goal is to continuously iterate for diverse scenarios, including supermarket stocking, hotel delivery, building inspections, and exhibition hall guidance.
Image Credit: Spirit AI
Beyond Moz2, Spirit AI's self-developed Spirit v1.6 embodied foundation model featured four live demonstrations covering long-horizon composite tasks, dynamic planning, fine manipulation, and general interaction. These displays fully showcased the company's full-stack technical capabilities, validated across real-world industrial scenarios.
The Spirit v1.6 embodied foundation model utilizes a deeply integrated architecture combining VLA and World Models. Unlike the industry's mainstream approach of splicing together separate modules, this architecture connects the entire decision chain—from environmental perception and task understanding to action planning and state prediction. This enables a human-like reaction mode of perceiving, deciding, and adjusting simultaneously, delivering superior generalization and robustness in complex, dynamic environments.
