Jump to navigation
Synthesis of nervous systems in hybrid robots utilizing hierarchical Q - Learning and temporal shifting
Waynelovich, John Luke
Frey, TerranceBaljon, ArlettePercus, AllonNadim, Ali
Robot-human interactions will become more important in many parts of the industrialized world as aging populations increasingly need more assistance, however several hurdles need to be overcome before this can be possible. Rigid robotic systems such as those found on factory floors are inherently dangerous to soft tissue and need to be operated from distances that never allow human contact. In addition, while robots are capable of high speed repetitive tasks in unchanging surroundings, operating in unknown and dynamic environments has been a long standing problem. This work describes several attempts to solve these problems by combining design methods and materials found in both traditional and soft robotics. This hybrid biomimetic robot is designed to apply gradual force to its skeletal frame via pneumatic muscles which allows for safe human interaction. The prototype also contains a synthetic nervous system made up of 14 discrete computers and microprocessors to mimic the most primitive parts of the mammalian brain as well as a small portion of the neocortex. The nervous system runs custom written q-learning algorithms which allow the robot to learn unknown environments by interacting with them. This self-optimizing behavior requires the prototype to choose actions within a state space containing on the order of 1012 states. The synthetic neocortex and custom software overcomes this by utilizing the novel approach of creating a virtual twelve dimensional state space by temporally shifting a master matrix of 1000 entries.
Computational Science Research Center
Doctor of Philosophy (Ph.D.) Claremont Graduate University and San Diego State University, 2017
© 2015 SDSU Library & Information Access. All Rights Reserved.