• obstacle detection
• robotic motion planning and control
• tactile sensors

Robots have to work close to humans. Hence, the interaction robot/human must be safe. To this end, the robots could have better awareness of the surrounding environment, moving in a predictable fashion (for humans) as well as to respond properly to humans’ action/intention. To this aim, what is needed is to devise proper human-like planning and retrieve obstacle and contact information for suitable replanning. In recent times, tactile optical sensors have emerged as a promising approach in robotics to retrieve contact information from the images of the sensor captured by a camera embedded in the sensor structure. In this work, a large-scale optical tactile sensor, the TacLink, is used to estimate the contact position of an obstacle colliding with the kinematic chain of the robotic manipulator. It consists of a soft artificial skin, whose inner part is covered by 256 white markers, mounted around a transparent rigid cylinder. Contact position is estimated via cameras relying on machine learning methods.
In my thesis, I integrated the TacLink as an additional link of a robotic manipulators, and within the ROS environment for motion planning and control. I validated the TacLink approach using Intrinsic Tactile Sensing. Based on the contact information, I designed a replanning algorithm, which is based on human functional components and potentials in the cartesian domain, which allows to avoid the detected obstacle. When the TacLink detects a contact, the planner requires the robot to return in the initial pose and then the trajectory is re-planned in order to avoid the obstacle. Additionally, to provide also contact force information about the interaction with the obstacle, I explored the usage of sensor-less methos, based on the estimation of the residual vector, which allows to estimate robot joint torques. In order to calculate the value of residual vector, the control torque and the manipulator dynamic model have to be known. With the residual vector, the position of the contact can be also estimated but only if the contact occurs at a link i=6, otherwise the Jacobian will not have full row rank. Hence, the idea is to use the TacLink in order to estimate the contact position and the residual vector to estimate the contact force.