This code is a supplementary material to the paper "TLIO: Tight Learned Inertial Odometry". To use the code here requires the user to generate its own dataset and retrain. For more information about the paper and the video materials, please refer to our website.

Installation

Dependencies tree can be retrieved from pyproject.toml. It is written for the poetry tool. All dependencies can thus be installed at once in a new virtual environment with:

cd src
poetry install

Then the virtual environment is accessible with:

poetry shell

Alternatively, the dependencies are also specified and can be installed through requirements.txt. First create a virtual environment with python3 interpreter, then run

pip install -r requirements.txt

Next commands should be run from this environment.

Dataset

A dataset is needed in the format of hdf5 to run with this code. The dataset tree structure looks like this under root directory Dataset:

Dataset
├── test.txt
├── train.txt
├── val.txt
├── seq1
│   ├── atttitude.txt
│   ├── calib_state.txt
│   ├── evolving_state.txt
│   └── data.hdf5
├── seq22
│   ├── atttitude.txt
│   ├── calib_state.txt
│   ├── evolving_state.txt
│   └── data.hdf5
...

data.hdf5 contains raw and calibrated IMU data and processed ground truth data. It is used for both the network and the filter. calib_state.txt contains calibration states from VIO and is used for filter initialization. atttitude.txt and evolving_state.txt are the outputs from AHRS attitude filter and VIO pose estimates. These are not used by the filter, but loaded for comparison / debug purposes.

The generation of data.hdf5 is specified in gen_fb_data.py, which requires interpolated stamped IMU measurement files and time-aligned VIO states files. The user can generate his/her own dataset with a different procedure to obtain the same fields to be used for network training and filter inputs.

File formats

Dataset lists

test.txt, train.txt and val.txt are list files specifying the split with one sequence name per row in the testing, training and validation datasets respectively. The name should be the same of the sequence directory for example seq1 and seq22 as above.

Used to generate data.hdf5

Timestamps (t) are in microseconds (us). Each row corresponds to data in a single timestamp. All data is delimited by commas.

• my_timestamps_p.txt VIO timestamps.
  • [ t ]
  • Note: single column, skipped first 20 frames
• imu_measurements.txt IMU data, raw and calibrated using VIO calibration
  • [t, acc_raw (3), acc_cal (3), gyr_raw (3), gyr_cal (3), has_vio]
  • Note: has been interpolated evenly between images. Every timestamp in my_timestamps_p.txt will have a corresponding timestamp in this file (has_vio==1)
• evolving_state.txt VIO evolving states at IMU rate.
  • [t, q_wxyz (4), p (3), v (3)]
  • Note: not interpolated, raw IMU timestamps. Every timestamp in my_timestamps_p.txt will have a corresponding timestamp in this file
• calib_state.txt VIO calibration states at image rate (used in data_io.py)
  • [t, acc_scale_inv (9), gyr_scale_inv (9), gyro_g_sense (9), b_acc (3), b_gyr (3)]
  • Note: Changing calibration states from VIO.
• atttitude.txt AHRS attitude from IMU
  • [ t, qw, qx, qy, qz ]

Network training and evaluation

For training or evaluation of one model

There are three different modes for the network part.--mode parameter defines the behaviour of main_net.py. Select between train, test and eval.
train: training a network model with training and validation dataset.
test: running an existing network model on testing dataset to obtain concatenated trajectories and metrics.
eval: running an exising network model and save all statistics of data samples for network performance evaluation.

1. Training:

Parameters:

--root_dir: dataset root directory. Each subfolder of root directory is a dataset.
--train_list: directory of the txt file with a list of training datasets. It should contain name of subfolder in root.
--val_list: directory of the txt file with a list of validation datasets. \It should contain name of subfolder in root.
--out_dir: training output directory, where checkpoints and logs folders will be created to store trained models and tensorboard logs respectively. A parameters.json file will also be saved.

Example:

python3 src/main_net.py \
--mode train \
--root_dir data/Dataset \
--train_list data/Dataset/train.txt \
--val_list data/Dataset/val.txt \
--out_dir train_outputs

2. Testing:

Parameters:

--test_list: path of the txt file with a list of testing datasets.
--model_path: path of the trained model to test with.
--out_dir: testing output directory, where a folder for each dataset tested will be created containing estimated trajectory as trajectory.txt and plots if specified. metrics.json contains the statistics for each dataset.

Example:

python3 src/main_net.py \
--mode test \
--root_dir data/Dataset \
--test_list data/Dataset/test.txt \
--model_path models/resnet/checkpoint_*.pt \
--out_dir test_outputs

3. Evaluation:

Parameters:

--out_dir: evaluation pickle file output directory.
--sample_freq: the frequency of network input data sample tested in Hz.
--out_name: (optional) output pickle file name.

Example:

python3 src/main_net.py \
--mode eval \
--root_dir data/Dataset \
--test_list data/Dataset/test.txt \
--model_path models/resnet/checkpoint_*.pt \
--out_dir eval_outputs \
--sample_freq 5 \
--out_name resnet.pkl

Please refer to main_net.py for a full list of parameters.

For batch evaluation on multiple models

Batch scripts are under src/batch_analysis module. Execute batch scripts from the src folder.

Testing:

Batch testing tests a list of datasets using multiple models and for each model save the trajectories, plots and metrics into a separate model folder. Output tree structure looks like this:

batch_test_outputs
├── model1
│   ├── seq1
│   │   ├── trajectory.txt
│   │   └── *.png
│   ├── seq2
...
│   └── metrics.json
├── model2
│   ├── seq1
...
│   └── metrics.json
...

Create an output directory and go to the src folder

mkdir batch_test_outputs
cd src

Run batch tests. --model_globbing is the globbing pattern to find all models to test.

python -m batch_runner.net_test_batch \
--root_dir ../data/Dataset \
--data_list ../data/Dataset/test.txt \
--model_globbing "../models/*/checkpoint_*.pt" \
--out_dir ../batch_test_outputs \

To save plots as well, change parameter save_plot to True in main_net.py.

Evaluation:

Batch evaluation runs the eval mode for multiple models, with various perturbation settings. Different perturbations result in a separate pickle file under each model folder. Output tree structure:

net_eval_outputs
├── model1
│   ├── d-bias-0.0-0.025-grav-0.0.pkl
│   ├── d-bias-0.0-0.05-grav-0.0.pkl
│   ├── d-bias-0.0-0.075-grav-0.0.pkl
│   ├── d-bias-0.0-0.0-grav-0.0.pkl
│   ├── d-bias-0.0-0.0-grav-10.0.pkl
│   ├── d-bias-0.0-0.0-grav-2.0.pkl
│   ├── d-bias-0.0-0.0-grav-4.0.pkl
│   ├── d-bias-0.0-0.0-grav-6.0.pkl
│   ├── d-bias-0.0-0.0-grav-8.0.pkl
│   ├── d-bias-0.0-0.1-grav-0.0.pkl
│   ├── d-bias-0.1-0.0-grav-0.0.pkl
│   ├── d-bias-0.2-0.0-grav-0.0.pkl
│   ├── d-bias-0.3-0.0-grav-0.0.pkl
│   ├── d-bias-0.4-0.0-grav-0.0.pkl
│   └── d-bias-0.5-0.0-grav-0.0.pkl
├── model2
│   ├── d-bias-0.0-0.025-grav-0.0.pkl
│   ├── d-bias-0.0-0.05-grav-0.0.pkl
...

In the current script, the following perturbation values are used:
Accelerometer bias perturbation range: [0, 0.1, 0.2, 0.3, 0.4, 0.5] (m/s^2)
Gyroscope bias perturbation range: [0, 0.025, 0.05, 0.075, 0.1] (rad/s)
Gravity direction perturbation range: [0, 0, 2, 4, 6, 8, 10] (degrees)
These can be changed in the script batch_runner/net_eval_batch.py, and for each perturbation range a pkl file will be saved with the range in the filename.

Create an output directory and go to the src folder

mkdir batch_eval_outputs
cd src

Run batch evaluation

python -m batch_runner.net_eval_batch \
--root_dir ../data/Dataset \
--data_list ../data/Dataset/test.txt \
--model_globbing "../models/*/checkpoint_*.pt" \
--out_dir ../net_eval_outputs \
--sample_freq 5.0

Running analysis and generating plots

After running testing and evaluation in batches, the statistics are saved in either metrics.json or the generated pickle files. To visualize the results and compare between models, we provide scripts that display the results in an interactive shell through iPython. The scripts are under src/analysis module.

To visualize network testing results from metrics.json including trajectory metrics and testing losses, go to src folder and run

python -m analysis.display_json \
--glob_dataset "../batch_test_output/*/"

This will leave you in an interactive shell with a preloaded panda DataFrame d. You can use it to visualize all metrics with the following helper function:

plot_all_stats_net(d)

To visualize evaluation results from pickle files, run

python -m analysis.display_pickle \
--glob_pickle "../batch_eval_outputs/*/*.pkl"

This gives access to all the sample data 3D displacement gt and errors, sigmas, mse and likelihood losses, 2D norm and angle gt and errors, and mahalanobis distance based on the regressed covariance. To plot sigmas vs. errors for example, run

plot_sigmas(d)

Running EKF with network displacement estimates

Running EKF with one network model

Use src/main_filter.py for running the filter and parsing parameters. The program supports running multiple datasets on one specified network model.

Parameters:

--model_path: path to saved model checkpoint file.
--model_param_path: path to parameter json file for this model.
--out_dir: filter output directory. This will include a parameters.json file with filter parameters, and a folder for each dataset containing the logged states, default to not_vio_state.txt.
--erase_old_log: overwrite old log files. If set to --no-erase_old_log, the program would skip running on the datasets if the output file already exists in the output directory.
--save_as_npy: convert the output txt file to npy file and append file extension (e.g. not_vio_state.txt.npy) to save space.
--initialize_with_offline_calib: initialize with offline calibration of the IMU. If set to --no-initialize_with_offline_calib the initial IMU biases will be initialized to 0.

Example:

python3 src/main_filter.py \
--root_dir data/Dataset \
--data_list data/Dataset/test.txt \
--model_path models/resnet/checkpoint_75.pt \
--model_param_path models/resnet/parameters.json \
--out_dir filter_outputs \
--erase_old_log \
--save_as_npy \
--initialize_with_offline_calib

Please refer to main_filter.py for a full list of parameters.

Batch running filter on multiple models and parameters

Batch script batch_runner/filter_batch provides functionality to run the main file in batch settings. Go to src folder to run the module and you can set the parameters to test within the script (e.g. different update frequencies).

Example:

python -m batch_runner.run_batch \
--root_dir ../data/Dataset \
--data_list ../data/Dataset/test.txt \
--model_globbing "../models/*/checkpoint_*.pt" \
--out_dir ../batch_filter_outputs

Batch running metrics and plot generation

To generate plots of the states of the filter and to generate metrics.json file for both the filter and network concatenation approaches, batch run plot_state.py on the existing filter and network testing outputs.

Parameters:

--runname_globbing: globbing pattern for all the model names to plot. This pattern should match between filter and ronin and exist in both --filter_dir and --ronin_dir.
--no_make_plots: not to save plots. If removed plots will be saved in the filter output folders for each trajectory.

Example:

python -m batch_runner.plot_batch \
--root_dir ../data/Dataset \
--data_list ../data/Dataset/test.txt \
--runname_globbing "*" \
--filter_dir ../batch_filter_outputs \
--ronin_dir ../batch_test_outputs \

Up to now a metrics.json file will be added to each model folder, and the tree structure would look like this:

batch_filter_outputs
├── model1
│   ├── seq1
│   │   ├── *.png
│   │   ├── not_vio_state.txt.npy
│   │   └── vio_states.npy
│   ├── seq1
│   │   ├── *.png
│   │   ├── not_vio_state.txt.npy
│   │   └── vio_states.npy
...
│   ├── metrics.json
│   └── parameters.json
├── model2
...

To generate plots from the metrics:

python -m analysis.display_json \
--glob_dataset "../batch_filter_outputs/*/"