Luxury tech brand BUTTONS collaborates with TUS-LINK to launch the first audio-visual robot equipped with HALI intelligent agent | Latest news

Author | Huang Nan

Editor | Yuan Silai

Recently, the global luxury technology brand BUTTONS launched a new product, the "BUTTONS SOLEMATE Intelligent Agent Super Audio - Visual Robot". This is also its first hardware device equipped with TSL's general intelligent agent HALI. The general intelligent agent HALI was released on November 14, 2024. After continuous iteration and optimization, it has gradually evolved from a highly anthropomorphic intelligent agent to a "life collaborator" with spatial cognition and physical interaction capabilities.

The general intelligent agent HALI and the BUTTONS SOLEMATE Intelligent Agent Super Audio - Visual Robot

TSL's HALI has built a three - dimensional semantic memory model deeply bound to the physical environment, naturally associating information memory such as item locations, event records, and retrieval with spatial coordinates and environmental context. Its advantage lies in not only effectively improving the intuitiveness and accuracy of information retrieval but also enabling AI to understand and use the "world model" for reasoning.

As the intelligent center for user interaction, HALI can perceive and process user requests in real - time. Different from the traditional model that relies on specific wake - up words or application entrances, HALI intelligent agents take the user's position in space, behavioral intention, and even environmental status as the primary context for interaction. This means that intelligent services do not need to be "summoned" but are presented actively and seamlessly at appropriate spatio - temporal nodes based on a deep understanding of the physical space, realizing an active transformation from "users seeking services" to "services approaching users".

Specifically in the operation process, HALI first analyzes the user's intention and disassembles tasks through the edge - side model. For example, at the planning level, HALI can effectively achieve the optimal arrangement of resources, devices, and behaviors in the spatio - temporal continuum. It can understand the structure of the home space, the user's movement trajectory, and environmental changes. Then, it uses the edge - side chip to complete data pre - processing and encryption. Finally, it links the cloud to schedule multiple vertical - domain intelligent agents to jointly achieve complex functions.

In this process, the GPU server cluster of TSL's Xuzhou AIoT Intelligent Computing Center is responsible for executing large - scale collaborative computing, and the hybrid intelligent computing engine on its AIoT Intelligent Computing Cloud Platform supports dynamic task scheduling. At the same time, TSL's cloud - based large model undertakes path planning to ensure that the device can flexibly avoid spatial obstacles and accurately reach the destination. The visual - language model is responsible for identifying the target, the large - language model generates the execution strategy and sends it back, and finally, the intelligent agent implements the decision - making and response.

The BUTTONS SOLEMATE Intelligent Agent Super Audio - Visual Robot

Based on the huge heterogeneous computing power reserve of TSL's Xuzhou Intelligent Computing Center, complex processes can be completed within seconds. Regarding the problem of fusion computing for heterogeneous chips, TSL's AIoT Intelligent Computing Cloud Platform has built a unified abstraction layer composed of operator libraries of various architecture chips between the model and the hardware and provides an end - to - end toolchain compatible with multiple inference and training frameworks. This enables the model to run seamlessly on heterogeneous computing power and significantly improves the efficiency of inference and training.

Dr. Shao Ling, Chief AI Officer and Global President of TSL Technology Group, pointed out that "the capabilities of dedicated AI Agents are strictly limited to their professional fields, lacking the ability to transfer learning across domains or solve new types of tasks... They mainly operate in the digital world without spatial perception or direct interaction with the physical world."

Currently, a key step in the evolution of AI towards generality is to break through the barriers of the digital world to obtain a complete closed - loop of "perception - reasoning - action" capabilities in the physical environment. "That is to say, a truly general intelligent agent must be able to perceive the geometric structure and dynamic changes of the three - dimensional environment, reason about the spatial relationships between objects, physical laws, and the causal chains of actions, and finally take action to safely and effectively perform tasks to influence the real world," said Shao Ling.