Soft-Robotics is an emerging field that holds the potential to answer questions in fields ranging from planetary colonization to invasive surgery. Developing controllers for soft robotic systems is challenging and time consuming, as the malleability of their soft materials introduces significant dynamical complexity into the systems. This dynamical complexity is exemplified through the often complex and unintuitive robotic behaviors that are generated by such systems. Due to their unintuitive nature, most locomotive behaviors for soft robots are developed through the time intensive process of empirical trial and error. This leads to single-gait soft robots that fail to exploit the flexibility of their structure.
We experiment with a Quality Diversity Algorithm (QDA) to efficiently construct a behavioral repertoire given no prior knowledge of our soft tensegrity robot. With negligible human interference in a reasonable timescale, QDAs are able to construct a behavioral repertoire that contains a diverse selection of navigationally-useful behaviors. We then show algorithms that use these behavioral repertoires to automatically generate hands-off controller systems for our soft robot. This will allow the robot to autonomously perform simple locomotive tasks while adapting to a changing environment. This automatic approach toward behavior generation will significantly improve the adaptability of mobile soft robots through a diversification of their locomotive capabilities.