LSTM-force-estimation

Force estimation of continuum(hyper-redundant) manipulator using LSTM.

Summary

image processing

curve extrapolation & curve fit

find joint position and angle of segments

result

Prerequisition

# pip module for create virtual environment
sudo apt install python3.xx-venv  # e.g. python3.10-venv (humble)

# make virtual env with --symlink
python3 -m venv {env_name} --system-site-packages --symlinks

source {env_name}/bin/activate

# requirements
pip install -r requirements.txt

Process

cd scripts
python postprocess.py

LSTM model

---

Post-process of custom data (provided)

https://drive.google.com/file/d/1_4A-9XO9phco3tF5dqtNVLMGjcR7xW1V/view?usp=drive_link

mkdir data
wget https://drive.google.com/file/d/1_4A-9XO9phco3tF5dqtNVLMGjcR7xW1V/view?usp=drive_link
cd data
unzip 2024-08-08 experiment.zip

Then, enter each subdirectories (e.g. 2024-08-08-13-46-11 upper_init-X) and take out data_LPF_*, results/curve_fit_result-joint_angle_* to {PROJECT_DIR}/datasets/train (Please check the tree of datasets directory on git)

---

Train

! Check
source {env_name}/bin/activate

# train
cd scripts
python LSTM_train.py

# show
cd scripts
python LSTM_train.tensorboard.py

# predict (test)
cd scripts
python LSTM_predict.py

Run ROS2 node

Segment(Joint) Angle Publisher

deactivate virtual_env (for build using ROS python core only)

deactivate

build & source

# If the error `no module named 'em'`, please remove dirctories named 'build', 'install' and 'log'.
colcon build --symlink-install
. install/setup.bash

# activate virtual_env
source {env_name}/bin/activate

# run
ros2 run estimation_pkg segment_angle_estimator

# launch
ros2 launch estimation_pkg _launch.py

External Force Publisher (TBD)