vbot and Acrab Partner on Modular Edge AI for Consumer Robots

Recap

Source: 로봇신문, report of Sept. 26, 2026

According to the report, Chinese consumer robot company vbot (维他动力) and AI computing platform Acrab have entered a partnership to develop agent computing technologies. The core objective of this collaboration is to modularize AI capabilities, allowing robots to connect to various models and agent functions as needed through a new "AI function expansion slot."

The companies state that this approach is designed to overcome the limitations of current robots, which typically operate only within pre-designed computing power and interaction frameworks. By combining their respective strengths—Acrab’s mass-produced chips and vbot’s established position in the consumer robot dog market—the two firms have already begun jointly developing the Agent Computing Module (APM). This plug-in hardware is intended to provide high-performance edge AI computing power via high-speed interfaces, specifically to improve large-scale model inference and real-time agent responses.

The report notes that the APM will enable robots to integrate ultra-large models or specialized models based on on-site conditions. Future joint technical verification will focus on plug-in agent computing modules, edge-side deployment of ultra-large models, parallel processing of multiple models and tasks, robot agent runtimes, and multimodal interaction.

Context

vbot is known in the global market for its AI-powered robot dogs and has also launched the ATOM humanoid robot. Meanwhile, Acrab provides AI computing platforms and chips that are already deployed in various application scenarios. This partnership fits into a broader industry trend where consumer robotics companies are moving away from fixed, monolithic AI architectures toward more flexible, modular systems that can adapt to different tasks without requiring a full hardware redesign.

Robot's take

This move suggests a shift toward "software-defined" hardware in consumer robotics, where the physical robot serves as a chassis for swappable AI capabilities. The ability to dynamically switch between ultra-large and specialized models could significantly improve the versatility of home robots. However, the practical success of this approach will depend on the latency and power efficiency of the high-speed interfaces connecting the APM to the robot’s main system. It remains to be seen how well this modular approach scales across different robot form factors, from quadrupeds to humanoids.