Skip to content

Latest commit

 

History

History
4 lines (3 loc) · 1.49 KB

README.md

File metadata and controls

4 lines (3 loc) · 1.49 KB

Human-robot-interacton

Human robot contact detection by applying machine learning techinques on Contact detection dataset gathered from a robot arm, Franka Emika Panda, when it was executing a repetitive movement.

Three machine learning models are trained using the dataset and implemented hyperparameter tunning. After hyperparameter tuning, significant enhancements in model performance have been observed across all three models: Random Forest, K-Nearest Neighbors (KNN), and Ensemble Bagging Trees. The accuracy of the Random Forest model notably increased to 83.55%, exhibiting improvements in precision, recall, and F1-scores across classes, particularly enhancing predictions for 'Collision_Link5' and 'Collision_Link6' classes. The KNN model achieved an accuracy of 91.5%, showcasing substantial improvements in precision and recall for various classes, especially excelling in predicting the 'Noncontact' class with an F1-score of 0.98. However, the most notable enhancement was seen in the Ensemble Bagging Trees model, reaching an accuracy of 92.42%. It demonstrated exceptional precision, recall, and F1-scores across all classes, particularly excelling in predicting 'Noncontact' instances. Considering the overall performance metrics post-tuning, the Ensemble Bagging Trees model emerges as the most robust and accurate choice for this classification task, closely followed by the KNN model, while the Random Forest model, despite improvement, exhibits comparatively lower performance across various classes.