Tactile sensing for ground classification

This paper describes the ground classification procedure for a six-legged walking robot. The terrain is classified according to the information gathered with the 6-DOF torque-force sensors. The statistical description of the obtained signals allow to obtain compact representation of the contact type for each 6-DOF component. Namely, the 4-element vector [variance, skewness, kurtosis, fifth moment] was used. Subsequently Discriminant Analysis was applied separately for classification of each component of the 6-DOF vector. The signal which gives the best classifica tion rate was established. The obtained result provides the information for the design of the new contact sensors for the robot feet.