UC Berkeley

This dissertation investigates adaptive control strategies for nonprehensile gripper contact intwo distinct embodied systems: squirrels landing on branches and a suction cup haptically searching surfaces. Central to this work is the concept of embodied dexterity - the integration of morphological computation with adaptive control. Morphological computation leverages the body’s design to assist in task execution, while adaptive control uses sensory input for real-time adjustments. Embodied dexterity is what enables agents to effectively interact with the physical world, where effective contact is crucial and varies with grasp type. Prehensile grasps for example, where the gripper wraps around the substrate, can rely on form closure, reducing the need for friction. However, nonprehensile grasps, necessary when the substrate is larger than the gripper, depend on high squeeze or suction forces for force closure, posing challenges in achieving contact stability.

The following studies highlight how positioning errors in grasping tasks can be dynamicallycompensated for by leveraging the physical design and sensory feedback of embodied agents, whether biological or robotic. I first present work on the biomechanics of squirrels landing on branches, examining their adaptive landing strategies. This includes their rapid forelimb dynamics to manage landing forces and torques. I also briefly introduce a squirrel-inspired gripper design. Next, I introduce the Smart Suction Cup and demonstrate how two tasks were accomplished with two different control algorithms. The first algorithm enhances grasping success on challenging surfaces by leveraging haptic signals, while the second enables contour- following.