Topic

Synthesizing data to help bring embodied intelligence to life

There is a huge data gap in the field of embodied intelligence, but real data is scarce and expensive to collect as well as the scenarios lack generalization. Simulated synthetic data is becoming a mainstream solution for embodied model training and algorithm validation. It can not only cover rare edge scenarios, easily expand the data scale, collect efficiently, guarantee privacy compliance, but also help algorithm testing and product validation. From the product perspective, we will discuss synthetic data generation technologies (e.g., physical simulation, scene Real2Sim), data quality assessment system design (authenticity assessment, utility assessment, distribution statistics), landing challenges and solutions, taking into account the case practice and process architecture, and building a product-level landing path.