DIDO:Deep Inertial Quadrotor Dynamical Odometry

Dataset

Each folder contains a ‘’data.hdf5‘’ containing：

（1）time ts

（2）Ground truth value of IMU frame in the gravity frame：acceleration gt_acc，position gt_p，linear velocity gt_v，attitude gt_q (in w,x,y,z form)

（3）Ground truth value of IMU frame in the IMU frame：angular velocity gt_gyr，

（4）IMU measurements：acceleration acc，angular velocity gyr

gen_list.txt：contains all the name of data
train.txtx：training data
val.txt：validation data
test：testing data

NOTE: The content in the existed train.txt, val.txt and test.txt are just examples. You can choose your own training, validation and testing data.

De_bias_acc

The neural network to reduce bias of the accelerator measurments

Execute De_bias_acc_main.py for training


cd De_bias_acc/src
python3 De_bias_acc_main.py \
--mode train \
--root_dir ../../dataset \
--train_list ../../dataset/train.txt \
--val_list ../../dataset/val.txt \
--out_dir ../train_outputs

--root_dir ../../dataset \ # root directory of data
--train_list ../../dataset/train.txt \ # the list of training data
--val_list ../../dataset/val.txt \ # the list of validation data
--out_dir ../train_outputs \ # output directory

Execute De_bias_acc_main.py for testing

cd De_bias_acc/src
python3 De_bias_acc_main.py \
--mode test \
--root_dir ../../dataset \
--test_list ../../dataset/test.txt \
--model_path ../train_outputs/checkpoints/checkpoint_*.pt \
--out_dir ../test_outputs

--root_dir ../../dataset \ # root directory of data
--test_list ../../dataset/test.txt \ # the list of testing data
--model_path ../trainoutputs/checkpoints/checkpoint*.pt \ # saving path of de_acc_bias_net
--out_dir test_outputs # output directory (there are the velocity figures obtained by acceleration integration using the true attitude in this directory)

De_bias_gyr

The neural network to reduce bias of the gyroscope measurments

Execute the De_bias_gyr_main.py for training

cd De_bias_gyr/src
python3 De_bias_gyr_main.py \
--mode train \
--root_dir ../../dataset \
--train_list ../../dataset/train.txt \
--val_list ../../dataset/val.txt \
--out_dir ../train_outputs

--root_dir ../../dataset \ # root directory of data
--train_list ../../dataset/train.txt \ # the list of training data
--val_list ../../dataset/val.txt \ # the list of validation data
--out_dir ../train_outputs \ # output directory

Rotation_ekf

First, using the De_bias_net to estimate the acceleration and angular velocity.

cd Rotation_ekf/src
python3 generate_net_acc_net_gyr.py \
--root_dir ../../dataset \
--network_acc_path ../../De_bias_acc/train_outputs/checkpoints/checkpoint_*.pt \
--network_gyr_path ../../De_bias_gyr/train_outputs/checkpoints/checkpoint_*.pt \
--test_list ../../dataset/gen_list.txt \
--out_dir ../output

--root_dir ../../dataset \ # root directory of data
--network_acc_path ../../De_bias_acc/trainoutputs/checkpoints/checkpoint*.pt # saving path of de_acc_bias_net
--network_gyr_path ../../De_bias_acc/trainoutputs/checkpoints/checkpoint*.pt # saving path of de_gyr_bias_net
--test_list ../../dataset/gen_list.txt # the list of all the data in the dataset
--out_dir ../output # output directory

NOTE: The produced acceleration and angular velocity will be saved in directory “../output/net_acc” and “../output/net_gyr”

Second, obtain the attitude after gravity alignment update.

cd Rotation_ekf/src
python3 Rotation_stage.py \
--root_dir ../../dataset \
--network_acc_out_path ../output/net_acc/ \
--network_gyr_out_path ../output/net_gyr/ \
--test_list ../../dataset/test.txt \
--out_dir ../output

--root_dir ../../dataset \ # root directory of data
--network_acc_out_path ../output/net_acc/ \ # saving path of acceleration after using de_bias_net
--network_gyr_out_path ../output/net_gyr/ # saving path of angular velocity after using de_bias_net
--test_list ../../dataset/test.txt \ # the testing data
--out_dir ../output # output directory

NOTE: There are some results in the output directory

ekf_q: attitude estimated by rotation ekf
euler: figures of the attitude estimation (euler_pred: the euler angle estimated by directly new angular velocity integration; euler_gt_euler: the ground truth euler angle; ekf_euler: euler angle estimated by rotation ekf)
euler_pred_error: figures of the attitude estimation error (euler_pred_error: error of euler_pred; ekf_euler_error: error of ekf_euler)

V_P_net

The attitude estimated from rotation_ekf is used in validation and testing

1、Training

# training x axis
cd V_P_net/src
python3 V_P_net_main.py \
--mode train \
--root_dir ../../dataset \
--train_list ../../dataset/train.txt \
--val_list ../../dataset/val.txt \
--out_dir ../train_outputs_x \
--train_axis x_axis

# training y axis
cd V_P_net/src
python3 V_P_net_main.py \
--mode train \
--root_dir ../../dataset \
--train_list ../../dataset/train.txt \
--val_list ../../dataset/val.txt \
--out_dir ../train_outputs_y \
--train_axis y_axis

# training z axis
cd V_P_net/src
python3 V_P_net_main.py \
--mode train \
--root_dir ../../dataset \
--train_list ../../dataset/train.txt \
--val_list ../../dataset/val.txt \
--out_dir ../train_outputs_z \
--train_axis z_axis

--root_dir ../../dataset \ # root directory of data
--train_list ../../dataset/train.txt \ # the list of training data
--val_list ../../dataset/val.txt \ # the list of validation data
--out_dir ../train_outputs_x # output directory
--train_axis x_axis # the axis of velocity and position are trained

2、Testing

cd V_P_net/src
python3 V_P_net_main.py \
--mode test \
--root_dir ../../dataset \
--test_list ../../dataset/train.txt \
--out_dir ../test_outputs \
--x_model ../train_outputs_x/checkpoints/checkpoint_*.pt \
--y_model ../train_outputs_y/checkpoints/checkpoint_*.pt \
--z_model ../train_outputs_z/checkpoints/checkpoint_*.pt

--root_dir ../../dataset \ # root directory of data
--test_list ../../dataset/train.txt \ # the list of testing data
--out_dir ../test_outputs # output directory
--x_model ../train_outputs_x/checkpoints/checkpoint*.pt \ # the saving path of v_p_net about x axis
--y_model ../train_outputs_y/checkpoints/checkpoint*.pt \ # the saving path of v_p_net about y axis
--z_model ../train_outputsz/checkpoints/checkpoint*.pt \ # the saving path of v_p_net about z axis

NOTE: There are some results in the output directory

vp：the mean value of estimated linear velocity and position
vp_cov：the covariance value of estimated linear velocity and position
imu_p_cov_in_world_frame：figure of position covariance
imu_p_error_in_world_frame：figure of position estimation error
imu_p_in_world_frame：figure of position estimation
imu_v_cov_in_world_frame：figure of linear velocity covariance
imu_v_error_in_world_frame：figure of linear velocity estimation error
imu_v_in_world_frame：figure of velocity estimation estimation

Res_dynamic

Training


cd Res_dynamic/src
python3 Res_dynamic_main.py \
--mode train \
--root_dir ../../dataset \
--train_list ../../dataset/train.txt \
--val_list ../../dataset/val.txt \
--out_dir ../train_outputs

--root_dir ../../dataset \ # root directory of data
--train_list ../../dataset/train.txt \ # the list of training data
--val_list ../../dataset/val.txt \ # the list of validation data
--out_dir train_outputs \ # output directory

Translation_ekf

cd Translation_ekf/src
python3 Translation_stage.py \
--root_dir ../../dataset \
--test_list ../../dataset/test.txt \
--out_dir ../output \
--network_dyn_path ../../Res_dynamic/train_outputs/checkpoints/checkpoint_*.pt \
--network_v_p_path ../../V_P_net/src/test_outputs/vp/ \
--network_acc_path ../../Rotation_ekf/output/net_acc/ \
--network_gyr_path ../../Rotation_ekf/output/net_gyr/ \
--network_q_path ../../Rotation_ekf/output/ekf_q/ \

--root_dir ../../dataset \ # root directory of data
--test_list ../../dataset/test.txt \ # the list of testing data
--out_dir ../output \ # output directory
--network_dyn_path ../../Res_dynamic/trainoutputs/checkpoints/checkpoint*.pt # the path of res_dynamics_net
--network_v_p_path ../../V_P_net/src/test_outputs/vp/ # the path of the result of v_p_net
--network_acc_path ../../Rotation_ekf/output/net_acc/ # the path of acceleration modified by de_bias net
--network_gyr_path ../../Rotation_ekf/output/net_gyr/ # the path of angular velocity modified by de_bias net
--network_q_path ../../Rotation_ekf/output/ekf_q/ # the path of attitude got from rotation ekf

zhangkunyi / DIDO

readme