Computational Design Synthesis for the Co-Design of Morphology and Actuation of Soft Robots for Locomotion

Compliant robots, i.e. robots that yield to external forces, have advantages compared to rigid robots in several respects. In direct contact with humans, compliant robots are safer. They can also compensate for imperfect perception of the environment. For example, a compliant gripper can achieve an even force distribution when holding a tomato without knowing the exact shape of the tomato. Soft robots are compliant robots that achieve compliant behavior through the use of compliant materials for their bodies. The field of soft robotic research is relatively young and next to the fabrication of soft robots, the design of soft robot morphology and actuation is still a challenge. Due to the vast design freedom, selecting a good design concept is challenging. Designing soft robots by hand often requires a trial-and-error process that is time-consuming. Thus, automation of the design can be advantageous. Currently, the majority of soft robot concepts are bio-inspired. Soft robots, however, can also be systems that are infeasible with biological systems. There is a need for a method to explore the solution space and enable an informed choice among concepts. This work focuses on the task of locomotion since it is an essential component of common robotic tasks such as search and rescue, exploration and pick and place. The objective of this research is to develop a computational design synthesis method for the conceptual design of the combined morphology and actuation of soft robots for locomotion. A Computational Design Synthesis (CDS) approach is taken in this thesis. It is an iterative process that generates, evaluates and optimizes design candidates. First, a spatial grammar for the generation of morphologies is presented. It can guide the process towards valid designs, since it enables fabrication requirements and constraints to be incorporated directly while still maintaining enough design freedom. The designs are evaluated in simulation using a rigid-body model for the approximation of soft robots and simulated annealing is used to optimize the designs. Both the CDS of soft robot morphology and the CDS of morphology and actuation are presented. Designs generated with the CDS of morphology only are actuated by fixed actuation patterns. To optimize the actuation patterns for every individual design, Parametric Actuation Curves (PAC) optimization is formulated and integrated in the morphology evaluation phase. The resulting CDS method for the co-design of morphology and actuation is shown to be more effective and more efficient than the CDS method for the automated design of morphology only. The designs resulting from the CDS method of morphology and actuation are prototyped successfully using two actuation methods; pneumatic actuation and tendon actuation. The different constraints and requirements of the two types of robots are incorporated directly in the spatial grammar. The prototypes compete with state-of-the-art soft robots with respect iiito speed and outperform them with respect to novelty, as defined by a novelty metric. The designs resulting from the presented co-design CDS method span a wide range of morphologies and gaits, many without biological analogy. Multiple solutions are generated with the same objective function value showing the multi-modality of the solution space. This work shows that the CDS of morphology and actuation enables a wider exploration of the entire solution space compared to bio-inspired design alone