Movement Primitives based control of Pleurobot
This project aims at finding a low dimensional solution to the high dimensional control problem of a Pleurobot. We use movement primitives obtained by doing a Principal Component Analysis on gait data of movement of salamander over several cycles to reproduce new and different movements. These principal components bring down the dimensionality by capturing essential features of the angle trajectories. These are then recombined using a weight matrix obtained by optimizing a target fitness function. We also do a separate analysis of the joint angle trajectories for swing and stance phases of salamander motion as these seem to be independent. This results in more robust and general primitives which can achieve higher fitnesses on given fitness functions as compared to the primitives from the complete motion cycle.