UC Santa Cruz

The compliant nature of soft robotic components lends itself well to manipulation and contact-rich tasks. The soft structures naturally form around objects where there could be uncertainty in shape or orientation and are inherently safer for fragile objects and humans. However, this compliancy makes the robot’s movement less constrained and less predictable making it difficult to control the position of the soft robotic manipulator without new types of sensing. To address these drawbacks, the combination of curvature, inflation, and contact sensors are added to give the finger the unique capability of somatosensory abilities, enhancing how it interacts with objects and its controllability. In this dissertation, we present the application of various approaches that provide a sense of touch for compliant soft robotic fingers. We rethink and validate the role that sensory feedback can play in the control of soft finger-like actuators with proprioceptive sensing capabilities. In our method of touch detection, instead of using the sensory feedback to control precisely the position of a soft finger, we use the disagreement between the controlled curvature sensor measurement and its reference signal to detect the contact between the soft finger and an object. We first consider the case of a static characteristic relation between the inflation pressure and sensor resistance. A control architecture is presented utilizing both the curvature and force sensors with the aim of providing a firm touch of a soft somatosensitive actuator with an object. The first component of the architecture, a reference tracking curvature controller, sets the finger in motion, which becomes blocked if an object is in its path. The result of such an event is that the finger bending is constrained, and the tracking error of the curvature controller increases. Once the error exceeds a predetermined threshold value, there is a switch from the reference tracking curvature controller to the second component, a force controller, which maintains the finger in contact with the object for a certain pressure using the force sensor measurement. We next consider a method for a kinesthetic touch approach for object detection that does not require the force sensor; therefore, it overcomes the necessity for the co-location between the point of contact and the corresponding sensor. The control architecture uses only the finger’s proprioceptive curvature sensor to detect contact with an object and maintain contact by switching to a different reference to hold at a constant curvature. Lastly, we focus on a method for identifying the dynamic finger curvature model to improve the closed-loop control. The proposed method addresses environmental variations as well as variations in material or human factors during the fabrication process that can have an effect on the finger dynamics. The approach uses the reference tracking error as a measure of the finger stress resulting from a contact with an object. The approach for contact detection has been tested in various tasks, including keeping the finger in a firm touch with an object, detecting the object edges and visualizing the operating space based on the sense of touch. These tasks demonstrate that the error signal contains robust information regarding the finger’s sense of touch and how it interacts with the environment. Findings are demonstrated across both a proxy of a soft finger, a simulated compliant multi-link actuator with flexible joints, and a real soft robotic finger.