Achieving Global Shapes Through Local Goals in Reconfigurable Robotics

This project looks at the problem of creating an online distributed control method to solve the reconfiguration problem in reconfigurable robotics. We look at breaking down the global reconfiguration task into local tasks that can be performed robustly by the individual modules in order to achieve the global reconfiguration task. This is done by the modules first learning local reconfiguration skills offline. The modules then apply these skills individually online to achieve local goals in order to achieve the global goal as a system. Ideas are taken from transfer learning, work on local contexts [19], and work on sequential decision making [12]. The method devised promises to be useful in addressing the reconfiguration problem for reconfigurable robots but problems with the task assignment hindered its performance and hence requires further investigation.