Nanonics : a platform for studying human to machine skill transfer in airplanes

Abstract

Complex dynamic systems such as vehicle navigation are controlled by human operators that develop their skills after years of training. In most cases, this skill is sub-cognitive and difficult to reproduce in an algorithmic format. The operator cannot describe the skill but can in most cases demonstrate it, so that an ideal approach to reconstruct or simulate the human skill would involve machine learning techniques using the operator's actions as input. In recent years much work has been done in the area of abstracting computational models of human control strategies. This type of modeling has been used to create autonomous ground vehicles from observation of human drivers, both in simulation and on real roads. One such example is the Intelligent Vehicle High System (IVHS), which is currently being undertaken in America to cut down on traffic congestion. One method of human skill modeling which has proven successful is the use of neural networks to develop a mapping between sensor inputs and human control outputs. Neural networks are nonlinear function approximators, resulting in a nonlinear control system when the behavior exhibited by the human controlling the vehicle is nonlinear. In this project a platform has been developed as a platform for studying human to machine skill transfer in an aerial vehicle using an evolving neural network model called a cascade correlation network. The platform also consists of a simulator that has the capabilities of recording the control actions of a human pilot, together with instrument readings. This data is then used to develop a model of the human pilot's control strategies which will enable the airplane to fly autonomously.