Aim: 

• Task planning for rehabilitation robot

Technical Problem: 

• Unknown reference trajectory for bimanual ADL training

Novel Approach:

• Used grammar of action to reconstruct the human-like task planning 

• Applied Two-Level Classification with Mean Square Error (MSE), after performing feature reduction by Linear Discriminant Analysis (LDA)

• Classification model was verified by Cross-Validation method

Preliminary Result

• Accuracy with test set, using 132 features, achieve 70.52%

Robot task and motion planning

Machine learning technique (Classification)