Topic

Embodied Intelligence Data Engineering: Technical Exploration and Practice

With the rapid development of artificial intelligence technology, Embodied AI (AI) has gradually become a research hotspot. Embodied AI emphasizes that intelligences learn and understand the world through interaction with the physical environment, and its core lies in the combination of perception, cognition and action. However, the development of embodied intelligence faces the great challenge of data scarcity. High-quality and diverse data is the key to training embodied intelligence models, but there are still many bottlenecks in current data production techniques. This talk will delve into the value of embodied intelligence data, the data bottlenecks currently faced, and how to break through these bottlenecks through innovative data engineering techniques. In particular, we will introduce AIRSPEED, a data production platform for embodied intelligence developed by our team, and how it can provide powerful support for embodied intelligence research by integrating real-world data collection and simulation data generation. Finally, we will discuss the problems in current data production techniques and provide an outlook on future directions. Outline: 1. the value of data for embodied intelligence 2. Data bottlenecks in embodied intelligence 3. data engineering for embodied intelligence 4. AIRSPEED, a data production platform for embodied intelligence 5. Problems and Prospects