4D Radar-Inertial Odometry based on Gaussian Modeling and Multi-Hypothesis Scan Matching

[Preprint (arXiv)]

Requirements

• CUDA Toolkit (including nvcc)
• ROS 1 Noetic (only for reading NTU4DRadLM rosbags)
• PyTorch
• numpy
• scikit-learn
• matplotlib
• small_gicp
• evo

Usage

This repository contains the following Python scripts:

• run_odometry.py: Runs the odometry system on the specified sequence.
• evaluate.py: Generates quantitative evaluation metrics for the specified method using evo.

Both scripts are configured using config.ini:

[config]

General configuration.

• dataset: Specifies the name of the dataset used (currently only NTU4DRadLM is supported)

[odometry]

Configuration specific to run_odometry.py.

• sequence: Specifies the name of the sequence used to run the odometry.
• ablation_gicp: Set to true if running the GICP ablated version (default is false)
• num_particles: Number of scan matching hypothesis particles. Set to 1 if running the single hypothesis ablated version (default is 4).
• out_name: Name of the output file. Default is odom_TIMESTAMP, where the timestamp is in YYYYMMDDhhmmss format.

[evaluation]

Configuration specific to evaluate.py.

• gt_pattern: Filename pattern of ground truth trajectory files, relative to the dataset folder.
• pred_pattern: Filename pattern of generated odometry trajectory files.
• method: Name of the method to evaluate.
• sequences: Comma-separated list of sequences to evaluate.

The following placeholders are supported in filename patterns:

• {method}: Name of the method
• {dataset}: Name of the dataset
• {seq}: Name of the sequence within the dataset

Reference

@misc{gaussian4drio,
	author = {Fernando Amodeo and Luis Merino and Fernando Caballero},
	title = {4D Radar-Inertial Odometry based on Gaussian Modeling and Multi-Hypothesis Scan Matching},
	year = {2024},
	eprint = {arXiv:2412.13639},
}

Acknowledgements

This work was partially supported by the following grants: 1) INSERTION PID2021-127648OB-C31, and 2) NORDIC TED2021-132476B-I00 projects, funded by MCIN/AEI/ 10.13039/501100011033 and the "European Union NextGenerationEU / PRTR".