Robotic Tactile Sensors For Changing Contact Conditions PDF Download

In recent years, robots have increasingly operated in a range of relatively unstructured environments, from outdoor agricultural operations to a cluttered kitchen in the home. As robots operate in these environments, they interact through continuously changing contact conditions between their hands and feet and the surfaces they touch. Toward allowing robots to respond to changing contact conditions, this thesis presents new tactile sensors for three particularly challenging scenarios: small running robots that need to sense changing contact conditions at their feet; grippers that employ gecko-inspired adhesion and need to sense how the adhesion is changing; and frictional grippers that use controlled sliding for manipulation. In each case, the sensing solution is informed by models of the contacts and how they can change. The first application focuses on leg-ground contacts for small running robots. Although legs are more complicated than wheels, legged robots are gradually growing in popularity due to their agility and versatility on various outdoor terrains. For best performance in terms of speed, efficiency and robust operation, legged robots should be equipped with sensors on their feet to monitor ground reaction forces and contact locations, so that they can account for how these affect running dynamics. However, it has been challenging to implement force sensors on the legs of small running robots because of the scale and geometry. To tackle this challenge, I developed a flexible capacitive force sensor array that measures distributed normal forces and a shear force. The sensor is mounted on the compliant C-shaped feet of a small hexapod robot and provides information about the ground reaction forces, contact locations, and overall gait smoothness and stability. Using the sensor information, I demonstrate two adaptive gait control methods that achieve improved running in terrain transitions and that reduce trajectory disturbances arising from obstacle contacts. Secondly, this thesis addresses robots that rely on adhesion, especially gecko-inspired adhesion. Grippers with astrictive force capabilities, such as suction or adhesion, adhere to an object surface even in with the negative grasp forces, allowing to them handle challenging objects such as large flat tiles and large curved objects that they cannot enclose. Among the various astrictive forces, gecko-inspired adhesion enjoys recent attention for its controllability: it is activated simply by applying a shear force and releases when the shear force is relaxed. However, measuring the adhesion is difficult because it depends on the area of contact formed by microscopic fibrillar structures and a surface. To tackle this challenge, I devised two direct contact area sensors for a gecko-adhesive gripper by using guided Lamb wave sensing and capacitive near-field proximity sensing. The former is relatively insensitive to the material of the adherend surface; the latter provides a high spatial resolution, which is useful for small grippers. In both approaches, I show that the sensor response matches the real contact area of the microscopic fibrillar structures sticking to a surface. Using these sensors, the robot can monitor contact area changes during a grasping process and evaluate the gripping quality before a failure occurs. Lastly, this thesis considers tactile sensing for in-hand manipulation with sliding. In this type of contact, multimodal sensors are necessary to simultaneously monitor steady force interactions and dynamic contact events. This information is useful both for stable gripping under varying load and for manipulation with respect to a hand. However, it has been challenging to build a compact multimodal sensor with a large taxel array that can be sampled rapidly for detecting directional dynamic events such as linear or rotational sliding. To address this challenge, I devised a capacitive nib array sensor that measures local stresses as well as directional sliding motions. The sensor rapidly samples the tactile array by dynamically clustering the sensing electrodes into groups that are selectively sensitive to certain types of directional sliding. Using this sensor, I demonstrate an in-hand sliding manipulation that measures changing sliding contacts and controls the grasp force to pivot an object lying on a table to an upright pose.