Design, Development, and Human Analogous Control of a Climbing Robot

Abstract

In this thesis, a re-configurable wheeled climbing robot has been introduced. This robot is capable of doing a multitude of tasks that no other single robot could do in the past. It can climb staircases, move inside empty ducts and pipes, climb up ropes and poles of varying cross sections, and even jump over obstacles with proper motion coordination. It can also move inside narrow passageways by reconfiguring itself. The design of a re-configurable robot capable of traversing a wide range of unconventional terrains is the novelty in this invention. A comprehensive dynamic model of the robot is derived for the first time. A real-time simulator to try different control strategies by a human operator using conventional human-machine interfaces has been developed. This simulator can be employed to size the electromechanical actuators and to synthesize different control strategies in a short time. The data obtained can be also used to design a human-analogous autonomous controller. After outline of the theory, background and applications of soft computing techniques for system construction and control including Artificial Neural Networks (ANN), Fuzzy Logic Control (FLC), and the Adaptive Neuro-Fuzzy Inference Systems (ANFIS), a novel human-analogous control strategy based on ANFIS was implemented to control the position of the robot climbing a straight pole against gravity. The design process of the ANFIS-based human-analogous control strategy includes the following steps: First, a human expert tries to control the real system in real time within a human-in-the-loop simulator via a Human-Machine Interface (HMI) using sensory information obtained from visual tracing in the HMI in real time from the real system. The control task is done by using a control interface (i.e., a joystick). Relevant input/output data is stored, filtered, and used offline to tune the parameters of an ANFIS-based controller. The ANFIS controller whose parameters have been optimized is then implemented on the real system autonomously. Based on the information obtained via the HITL simulator system, the controller can extrapolate needed data for untrained cases.