From Bioinspiration to Computer Generation: Developments in Autonomous Soft Robot Design

Joshua Pinskier, David Howard
2021 Advanced Intelligent Systems  
Throughout their history, robots have been characterized by their demand for high speed, precision, and repeatability. Guided by these requirements, their designs have converged toward rigidbodied designs with discrete joints. Rigid links have the mechanical stiffness to support large loads, as required in automated assembly lines and other common applications. Discrete joints constrain motion to specified degrees of freedom (DOFs), with controllable finitedimensional kinematics. However, the
more » ... sulting hard and heavy robots require sophisticated sensing and control systems to operate outside of carefully structured environments. The nascent field of soft robotics overturns the paradigm of rigid robotics. Using materials with an elastic modulus in the order of kilopascals to megapascals, soft robots are designed to be lightweight and compliant. Their deformable bodies can safely interact with delicate objects, conform to the shape of unknown objects or terrain, and reconfigure to suit task requirements. [1, 2] Their monolithic designs and absence of moving parts are also ideal for multimaterial additive manufacturing, simplifying their production. Combining hard, soft, elastic, and conductive materials, entire functional soft robots can now be printed including sensors and actuators. [3, 4] While existing research has focused on the applications of soft grasping and soft locomotion, soft robots have the potential to revolutionize robotics operating in uncertain, confined, and fragile environments, for example, fruit picking, human assistance, or search and rescue robots. By using materials' functional properties to control their position and orientation (pose), soft robots embody intelligence, reducing the need for sensing and perception. Sharing control between embodied intelligence (or morphological computation) and digital computation exploits the robot's materials and morphology to enhance task performance, enabling simple soft robots to outperform sophisticated rigid ones when undertaking delicate tasks. Compared with their simplified control, the design of soft robots is complicated by their nonlinear materials and multiphysics coupling. The resulting large deformations, hyperelasticity, and viscoelasticty produce high-dimensional design spaces, which are complex and unintuitive, even for experienced designers. Nevertheless, in the absence of accurate analytical models and efficient simulation tools, soft robot design remains a primarily manual process in which models are heuristically designed and experimentally verified. Recently, efforts to automate their design have centered on parametric optimizations of existing soft robots. Typically, a gradient-based algorithm optimizes the geometry, materials, or actuation within specified bounds. [5] While these create
doi:10.1002/aisy.202100086 fatcat:jlazieflsjdy3gwydrc5u4vq2u