Background: Modern automation and robotics rely on large-scale data for training, testing, and development. However, in specialized environments such as warehouses, collecting and annotating real-world data is extremely costly. Synthetic data offers a promising alternative by significantly reducing the reliance on real data. Currently, the design of 3D scenes still mainly depends on manual work by domain experts.

Problem: The major barrier to scaling high-quality, designed 3D scenes lies in the vast diversity of objects and the complexity of their spatial arrangements. In this thesis, the student will investigate how to describe warehouses using a scene graph, and how to generate diverse synthetic scenes by randomizing scene graph parameters.

Tasks: Your tasks include:

• Research methods of using a scene graph to represent 3D environments
• Design a warehouses scene graph structure and define rules for randomized generation
• Collect or create CAD models of warehouse-related objects
• Implement an automatic pipeline to generate synthetic warehouses within the ISAAC SIM simulation environment

Requirements:

• Programming experience in Python
• Familiarity with CAD modeling and Isaac Sim is beneficial