Physical Human-Robot Contact Detection

Contact detection dataset is gathered from a robot arm, Franka Emika Panda, when it was executing a repetitive movement.
Each sample contains 140-milisecond time-lapse sensor data, motor torque, external torque, joint position, and velocity, with a 200 Hz sampling rate (28 data points). It comprises five classes which determine the contact state on the last two links of the robot, including their respective type, incidental or intentional. The following condition are considered to collect the data
Noncontact: a robot preforms its task without having any contact to humans.
Intentional Link5: an operator grasps the fifth link of the robot to interact with it.
Intentional Link6: an operator grasps the sixth link of the robot to interact with it.
Collision Link5: the fifth link of the robot collides to a human arm.
Collision Link6: the sixth link of the robot collides to a human arm.
In total, there are 1199 noncontact samples and 1006 contact samples. The distribution of contact data is as follows
1) Collision Link5 = 180 samples
2) Collision Link6 = 179 samples
3) Intentional Link5 = 339 samples
4) Intentional Link6 = 308 samples

70% of the data considered for training dataset and 30% for testing dataset. This rate is changeable in 'merge_contact_noncontact_csv.m' .