The exponential growth of global data demands unprecedented reliability and efficiency of data center operations. Traditional manual inspection methods remain inefficient and error-prone. This article presents an autonomous mobile robot enhanced by large language models and hierarchical three-dimensional scene graphs for intelligent operation and maintenance. The core innovation lies in enabling semantic-aware navigation, allowing the robot to interpret high-level instructions like “inspect servers with the red alert light on in cold aisle B” by leveraging the three-dimensional scene graph for spatial-semantic reasoning and the large language model for task parsing. Deployed in multiple ultra-large data centers, the system has improved inspection efficiency by over 50%, autonomously patrolled over 10,000 km, and identified more than 1200 equipment anomalies. This work demonstrates a significant step towards fully autonomous, intelligent infrastructure management.
Cheng et al. (2026) studied this question.