Digital Theses Archive


Tesi etd-03072023-143205

Type of thesis
Octopus-inspired Technologies for Blind Grasping and Manipulation of Unknown Objects
Scientific disciplinary sector
Istituto di Biorobotica - PHD IN BIOROBOTICA
  • Suction Cups
  • Blind Grasping
  • Octopus
  • Object Recognition
  • Sensorized Suction Cups
Exam session start date
Moving and performing manipulation tasks in real unstructured environments, such as natural ones, require autonomous intelligent robots with perception and actuation abilities similar to humans and animals. Sensory feedback, achieved by distributed tactile sensors in the robot body, is fundamental to understanding the environment during manipulation tasks. Biological systems have been the source of inspiration for engineers and roboticists for a long time. The main goal was to create devices that replicated the functionality of the biological models that inspired them, implementing the characteristics found in nature that were identified as the best solutions to the problems. <br>The Octopus arm has been adopted as a source of inspiration for the realization of soft manipulators and grippers for several years because of its fully soft body combined with distributed actuation and perception abilities. However, these artificial counterparts normally do not integrate suction cups along the arm and the only examples do not integrate tactile sensors, which limits its ability to perceive the environment and perform autonomous tasks. Several solutions were adopted in the last years to add tactile sensing abilities over suction cups, however, adding sensing without compromising suction cup softness and keeping low the complexity to guarantee the integration in a soft arm is challenging. The development of such a kind of sensorized suction cup allows the development of a soft arm that embeds in a single device grasping, attachment, and sensing abilities and that is able to operate in both air and water to grasp different objects without using vision (blind grasping).<br>The aim of this Ph.D. thesis is to develop the first octopus arm with sensorized suction cups for grasping and manipulation applications. By applying the principles of bioinspiration, an artificial sucker that replicates the morphology and functionality of the Octopus O. Vulgaris natural sucker was designed. To mimic the tactile ability of the natural sucker different sensing strategies were analysed and adopted. The first technique is based on the realization of embedded resistive strain sensors realized by means of microchannels (600 μm) inside of the artificial sucker, filled with conductive materials. Microchannels are made within the suction cup using two different manufacturing techniques. The first technique includes two-step casting to mix materials with different stiffness for increasing the sensitivity of the suction cup, while the second technique is using one-step casting to permit the realization of more accurate channels to improve sensor’s reliability. As a second technology for sensorizing the suction cup, optoelectronic sensors are utilized. Because lacking wiring and rigid components in sensitive areas, optical sensors realized by optoelectronic components have a great deal of application potential as tactile sensors. They are highly reliable because they are made of commercial components, and they can be made extremely small and low power if a convenient reading strategy is adopted. <br>Two actuation strategies were also investigated: pneumatic, and cable driven. In the first one, a new fluidic PneuNet-like soft actuator that can bend and twist clockwise and anticlockwise was designed, manufactured, and tested. In the second one, a cable-driven octopus’ arm was realized. The capability to bend and twist in both directions was achieved by tuning opportunely the position and path of the tendons inside the arm. <br>The cable-driven octopus-inspired arm was adopted for the integration of the sensorized suction cups. The developed octopus arm showed the capability to sense different kinds of objects along its body while is also able to perform grasping and attachment tasks.<br>The proposed sensorized suction cup, combined with machine learning and deep learning techniques, demonstrates the capability to recognize different parameters such as shape, stiffness, inclination, and relative position of the contacted objects, opening the way to the realization of multifunctional sensorized tool able to perform the blind grasping task in multiple environments.