MATHEMATICAL MODELING AND NEURAL NETWORK CONTROLLER DESIGN FOR BIPED ROBOT’S HUMAN LIKE WALK

Abstract

This thesis is concern with gait generation methods for stable walk of flat footed and toe footed biped robot models. Methods are proposed to generate suitable trajectory, which can easily adapt the changes in the boundary conditions/constraints during walk. Different type of approaches such as polynomials, Feedforward Neural Network(FNN) and Wavelet Neural Network(WNN) are considered for some biped models walking on uneven surfaces and for avoiding obstacles. To avoid stable leg’s knee bending and for more stable walk, lateral upper body motion is considered. Force/torque control for walk is designed by developing the dynamic equations of the biped model. The thesis is divided into 7 chapters which are briefly described below: Chapter 1 is introductory with a brief literature review in the area of robotics and neural networks. We have discussed some biped models and also the inherent challenges associated with the robots’ stable walk. Finally, a summary of the thesis is presented. Chapter 2 gives some basics and preliminaries which are used in subsequent chapters. Chapter 3 focuses on stable FNN trajectory tracking of flat footed biped robot with upper body motion. Trajectories using cubic spline are generated for ankle joints, hip joints and upper body so that the resulting walk is stable. Here, the effects of different lateral upper body motions of the flat footed robot on ZMP stability is analyzed for plane surface walking. The inverse kinematics of the ankle and hip trajectories are solved using FNN. Further, simulations are done using Matlab2010b. Chapter 4 proposes polynomial based trajectory generation algorithm of a robot model for stable human like gait considering the upper body motion, movable foot and active toe. This approach allows the smooth transition between walking phases namely, single support phase and double support phase. ZMP stability is analyzed for plane and uneven surface walking of the toe footed model by taking into account the lateral upper-body movements i along with the planned motion trajectories. Chapter 5 proposes and compares FNN and WNN based approaches for smooth trajectory generation under given constraints. The trajectory generation procedure is derived from semi-supervised NNs for given boundary conditions without assigning any path in advance. The trajectories generated by using proposed approaches can be modified according to the constraints value at any instant of time during tracking. Further, these approaches are used for the gait generation of a 5 DOF flat footed biped to walk on flat terrain in 3-dimensional space. The suitability of the proposed approaches is studied using ZMP stability criteria and simulations have been carried out using Matlab2014a. In Chapter 6, a biped robot model with flat foot is considered. The dynamic equation of this model is derived and a PD controller for stable walk is presented. FNN approach proposed in Chapter 5 is used for smooth and dynamically stable trajectory generation and the results are compared with polynomial approach. Simulation results (using Matlab2014a) show that this model can cross over obstacles of different heights and cross over a ditch by adjusting the step height and step length in ankle trajectory at any instant during tracking. Conclusions, limitations and future work have been outlined