Mohsen Parsa

  Abstract

  Research on the legged robots has been started since 1970. Because of the intrinsic instability and walking stability constraint, legged robot motion control, especially in bipeds, is more complicated than wheeled robots or manipulators. However, tolerating these difficulties, legged robots have more moving abilities than wheeled robots encountering uneven and discontinuous surfaces. In this thesis, nonlinear model predictive control (NMPC) of a five links biped robot is considered. Gait generation and motion control phases are done by the NMPC simultaneously and real-time. Based on this idea, biped control is not based on trajectory following and is trajectory free. Letting the NMPC to decide about both the walking pattern and control of the biped has advantages such as reaching to more optimal walking while considering the actuator and physical constraints of the biped. But the NMPC is a model based control which suffers from uncertainty and disturbances. In order to robust the controller, nonlinear disturbance observer is applied to the control loop. Thus, having the NMPC trajectory free advantages, the controller has sufficient robustness to unavoidable uncertainty and disturbances. In contrast to the NMPC advantages, computation time of it is a big deal which restricts the NMPC efficiency. In order to reduce the computation time, the main NMPC responsibilities are divided to three smaller NMPCs which are solved in series. Although these NMPCs are solved in series, the computation time is reduced.

  Defense Time: Jun 29 2010 at 10:00

  Location: Seminar Room