Neural Oscillator-based Motion Control for Bio-Mimetic Robot Systems

Abstract

In this dissertation, designs of different models of neural oscillators are proposed in order to use them for various applications in the field of bio-mimetic robotics. The rhythmic behavior in natural habitat taken into consideration and they are used to construct the aforementioned neural oscillators. These natural rhythmic behaviors may be central pattern generator (CPG) based motion in animals or the conscious encircling of some animals around a target to catch, observe, or protect (COP) it. Firstly, a neural oscillators based CPG is proposed so that different types of motions can be generated in a snake like robot. An active-wheels and passive-joints type of snake-like robot is constructed by joining together a series of wheeled links (with active wheels) by using passive prismatic and revolute joints. The joints helps in increase or decrease of lengths of the robot. A network of unified coupled neural oscillators is also provide in order to estimate the desired orientation and the distance between the wheeled link for the snake-like robot. The proposed algorithm is used in simulations and in experiments to achieve serpentine and rectilinear types of snake locomotion in the snake robot. Furthermore, a new snake robot is designed and constructed which is without wheels and have active revolute and prismatic joints. The constructed snake robot is a modular snake robot and each module has two revolute joints, these help in side-wise and upward motion of snake robot. Further to these roles, active prismatic joints are provided in modules to achieve linear motion along the body. Secondly, a neural oscillator model, coupled with integrate and fire neuron for a hopping robot is designed. The gait of hopping robot is generated by a combined action of a rotational and linear actuator. The neural oscillator, which also take into consideration the ground impact force on the robot, is designed to drive the rotational actuator. The oscillator is coupled to integrate and fire neuron to drive the linear actuator. This would help in controlling the periodic jumping of robot. The authenticity of proposed algorithm is tested using simulations. As a last application the neural oscillator based limit cycle is used to generate coordinated motion of multiple robots to mimic the encirclement property of animals while catching, observing or protecting a target. The state of target is supposed to be unknown under the assumption that it is stationary or in motion with constant unknown speed along a straight line. The algorithm also has the properties of avoiding static obstacles as well as self-collision. The robots approaches at a desired distance (limit) from the target, while keeping equally distributed angles around it. The verification of the authenticity of the proposed algorithm is achieved through simulations.