Abstract
To date only wheeled robots have been used for extraterrestrial exploration, however these systems are limited in their agility and ability to adapt to changing environments. Theoretically legged robots offer high levels of adaptability and mobility, however in practice their capabilities are constrained by their complexity and the requirement for actuators which are both fast and powerful; both of which are particularly undesirable for planetary rovers. The aim of this research is to begin to address some of the issues which make walking vehicles less suitable for extraterrestrial exploration; specifically the high power and computational requirements associated with the control of a vehicle with many powerful actuators, and the increased likelihood of mechanical failure due to the large number of moving parts. This has been achieved by using biological inspiration combined with conventional engineering techniques to create a walking rover prototype, known as EchinoBot. The design and implementation of EchinoBot has led to a number of novel contributions: the chassis was designed using a combination of evolutionary algorithms, which attempted to blend the open search space used by artificial life researchers with the more focussed and realistic cost criteria used to optimise robot morphology in previous work. This algorithm was also used to propose some ‘design rules’ for designing more efficient legged systems. A combined gait generation and path planning algorithm was implemented based on an abstracted echinoderm nervous system; this allowed omnidirectional movement and facilitated modularity in the rover chassis. Finally the controller was made more robust and adaptable with a tuning algorithm inspired by neuromodulation. This algorithm also contributes to the field of computational neuroscience by suggesting one way in which neuromodulation can be integrated with the ‘Bayesian brain’ model of brain function. This research illustrates how functional biomimesis can improve the adaptability of a system while avoiding the inefficiencies and poor interpretability often associated with biomimetic hardware and software.