New Tool Generates Realistic AI Simulations

Researchers have introduced EmbodiedGen, an open-source framework for generating realistic, scalable 3D assets tailored to embodied AI tasks. Developed by teams from Horizon Robotics, the Chinese University of Hong Kong, Shanghai Qi Zhi Institute, and Tsinghua University, EmbodiedGen produces physically accurate, watertight 3D objects in URDF format, complete with metadata for simulation use. 

Its six modular components support transforming images or text into 3D models, creating articulated items, and building complete scenes with realistic layouts. By ensuring accurate scale and geometry, EmbodiedGen aims to address a key challenge in training embodied AI: creating realistic environments without costly manual modeling. 

Compatible with major simulation platforms, it enables efficient development of virtual worlds for tasks such as navigation and manipulation, supporting broader research in robotics, digital twins, and real-to-sim learning while reducing development costs and barriers to scalable simulation.