Calibrating point clouds and velocimetry from a FMCW Lidar sensor against synthetic data generated from our rendering software.
Status (Feb 2024): Lidar simulation validated with hardware. Statistically agreements/disagreements between rendered and real-data are reported (results below, paper to follow). Poster presented at AAS GNC 2024 conference, Breckenridge, CO. Available upon request.
Coming up: File cleanup.
Process: We collect FMCW datasets from real-world scenes using Aeva 4D Lidar (example). We also collect the ground truth trajectory information (poses, velocities) using vicon system. 3D assets are generated from dense reconstruction of the real-world scene. We simulate the physics of the FMCW Lidar using a custom ray-tracing software built at LASR laboratory. Ultimately, we compare the statistically similarities between the rendered and the sensor outputs.
Note: We used two datasets (shown below) to validate our results: asteroid wall (landing approach) and spinning satellite (proximity ops). Datasets and rendering softwares are not public. Reach out to me for details.
Camera poses around reconstructed point cloud object in world frame
Reconstructed point cloud and ground truth trajectory from vicon
Reconstructed point cloud and point velocities
Median Absolute Deviations (MAD) of point velocities from NaRPA (red) and Aeva lidar (blue), within 1-sigma of velocities from sensor
If you use our software/approach/analysis in a scientific publication, we would appreciate using the following citations:
@article{eapen2022narpa,
title={NaRPA: Navigation and Rendering Pipeline for Astronautics},
author={Eapen, Roshan Thomas and Bhaskara, Ramchander Rao and Majji, Manoranjan},
journal={arXiv preprint arXiv:2211.01566},
year={2022}
}
This project is licensed under the Attribution-NonCommercial-NoDerivs 3.0 Unported (CC BY-NC-ND 3.0) License.