Skip to content

AntoineWeber/Project_LASA

Repository files navigation

A fast decode of grasping intention from the muscular activity

The goal of this project was to predict the grasping intention in the early stages of the reaching motion using 8 differential EMG channels. Reach-to-grasp motions were assessed using two different grasp types : the Thumb-2 fingers grasp type (the object is lifted using only the thumb, the index and the middle finger) and the power grasp type (with all the fingers). On the other side, the reach-not-to-grasp motions were assessed using a single grasp type being simply the no-grasp type (the hand reaches the object but do not grasp it).

Methods

Different classifiers were used to predict the grasping intention. The implemented classifiers are an Echo State Network, a Linear Discriminant Analysis, a C-Support Vector Machine and finally a Gaussian Mixture Model.

Structure

The code was written on MATLAB2018a. It can be found in the First_Approach and Second_Approach files. As explained in the report, three different approaches were performed to analyse the data. The first two are implementend in the first folder, the third one on the second folder.

To correctly run the code, one should first run the main file of the specific approach to be tested. It will load the raw data and process it accordingly to the given approach. The directory should always remain the same, being the SemesterProject folder. After the run of the main file, a new .mat file will be saved containing the data to be processed. From here, the different analysis being the ESN, LDA, SVM and GMM functions can be runned directly.

More specific definitions are inside the info file.

Author

  • Antoine Weber

Remarks

The dataset is not given for confidential reasons.

About

Work for my Semester project at LASA

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published