---
license: bsd-3-clause
task_categories:
- robotics
language:
- en
tags:
- deepvl
- underwater
- odometry
---
# DeepVL training dataset
## Introduction
This dataset repository contains the training and testing datasets used in the paper: ["DeepVL: Dynamics and Inertial Measurements-based Deep Velocity Learning for Underwater Odometry"](https://ntnu-arl.github.io/deepvl-deep-velocity-learning/). The dataset was collected by manually pilotting an underwater robot in a pool and in the Trondhiem fjord.
## Dataset details
The training data is located in the `train_full` directory and the test data in `test` directory respectively. The training data directory contains trajectories from `traj1` to `traj12`, and testing data contains from `traj1` to `traj2`. Each trajectory contains files described as follows:
```
trajX/
├── alphasense_imu_data.npy          # IMU data from Alphasense Sensense             | rate: 200Hz
├── biases_data.npy                  # Estimated IMU biases (from ReAqROVIO)         | rate:  20Hz
├── fcu_imu_data.npy                 # IMU data from flight control unit             | rate: 200Hz
├── gravity_b_vec.npy                # Gravity vector in body frame                  | rate:  20Hz
├── motor_commands_data.npy          # Motor command PWM signals for all 8 thrusters | rate: 200Hz
├── orientation_data_Rmat.npy        # Orientation matrices (body to world)          | rate:  20Hz
├── supervision_odom_data.npy        # Ground-truth odometry (from ReAqROVIO)        | rate:  20Hz
├── battery_data.npy                 # (Optional) Battery voltage and current        | rate:  20Hz
```
Each file is in `.npy` format and can be loaded and parsed using numpy. In each numpy file the data is organized as:

```
[data_column_1, data_column_2, ... data_column_N, time_stamp_column]
```

## Contact
For questions or support, contact authors:
* [Mohit Singh](mailto:mohit.singh@ntnu.no)
* [Kostas Alexis](mailto:konstantinos.alexis@ntnu.no)