Federated approach in Global 3D reconstruction and Mapping for Connected Autonomous Vehicles
Federated approach in Global 3D reconstruction and Mapping for Connected Autonomous Vehicles
Enable accurate global 3D mapping by combining LiDAR, and Visual Odometry with Gaussian Splatting.
Employ latest advancements in monocular depth estimation such as SfM (e.g. COLMAP), MVS (e.g. DUSt3R, MASt3R), and VGGT, etc.
Implement incremental learning of implicit signed distance fields and pose estimation using point-to-implicit registration approach and using transformer-based models to calculate virtual points latent features.
Design and develop elastic and compact implicit mapping pipeline, reducing map storage costs and data transmission bandwidth.
Integrate loop closure detection using neural point features, improving long-term map reliability and reducing navigational drift for autonomous systems.
Data acquisition and sensor calibration with 3D LiDAR, Camera, GPS, IMU, Odometry to conduct experiments in large outdoor and indoor environments under varying lighting, occlusion, and motion conditions.
