The TongSIM-Asset dataset is an open-source simulated environment and asset dataset for Embodied AI research, released by the Beijing General Artificial Intelligence Research Institute (BIGAI) in 2025. Related research papers include... TongSIM: A General Platform for Simulating Intelligent MachinesIt aims to provide a unified and highly consistent simulation platform to support embodied intelligent agents in perceiving, making decisions, interacting, and collaborating in indoor and outdoor environments that closely resemble the real physical world.

This dataset contains 100 high-quality 3D simulation scenes, covering multi-room indoor environments and complete outdoor urban scenes. All scenes can run stably at over 60 FPS. It also provides over 3,000 interactive 3D objects, covering over 500 object categories, and includes over 10 agent types (such as adults, children, and robots) and over 100 agent animations, used to simulate diverse behaviors and interactions in the real world. The dataset is built on Unreal Engine 5.6 and is primarily used for training, testing, and evaluating embodied agents in simulation environments.