GASE: Gaussian Splatting-Based Automated System for Reconstructing Embodied-Simulation Environments
Researchers have developed GASE, an automated system for creating high-fidelity simulation environments for robot learning. GASE uses multi-view video streams and a camera-pose-based strategy to efficiently scan environments and extract foreground objects for reconstruction. The system reportedly outperforms existing Gaussian-based methods in segmentation accuracy and achieves state-of-the-art inpainting quality, significantly reducing the sim-to-real gap for robot training. AI
IMPACT Enables more efficient and accurate simulation environments, potentially accelerating robot learning and reducing the need for real-world training data.