Supervised vs. Unsupervised Learning for Operator State Modeling in Unmanned Vehicle Settings

Abstract

In this paper, we model operator states using hidden Markov models applied to human supervisory control behaviors. More specifically, we model the behavior of an operator of multiple heterogeneous unmanned vehicle systems. The hidden Markov model framework allows the inference of higher operator states from observable operator interaction with a computer interface. For example, a sequence of operator actions can be used to compute a probability distribution of possible operator states. Such models are capable of detecting deviations from expected operator behavior as learned by the model. The difficulty with parametric inference models such as hidden Markov models is that a large number of parameters must either be specified by hand or learned from example data. We compare the behavioral models obtained with two different supervised learning techniques and an unsupervised hidden Markov model training technique. The results suggest that the best models of human supervisory control behavior are obtained through unsupervised learning. We conclude by presenting further extensions to this work.

Survey of RPAS Autonomous Control Systems Using Artificial Intelligence

“…There are two types of supervised learning algorithms -regression and classification. Supervised learn- [85] used a classic supervised learning method for autonomous path planning. For each hidden state in the learning paradigm, an emission probability function was analyzed to determine the set of most likely observables.…”

“…It is helpful in scenarios where we do not have access to labelled or purely standard data and let the machine work independently to learn. The first step to the method is the selection of a model that will determine the number of hidden states it should maintain [85]. The model is made to find its structure from the input data and discover groups with similar examples within the data.…”

The Impact of Different Levels of Autonomy and Training on Operators’ Drone Control Strategies

“…Human operator behavior models are important tools for behavior analysis in unmanned aerial system (UAS) settings. Through these models, we can determine if observed behavior patterns match experimenters' expectations, investigate operators' strategies to identify points of inefficiency or error, study both endogenous and exogenous factors that impact operator behavior patterns, and study how automation can improve operators' performance and success rate in task performance [3,4]. Models for operators undergoing different training can help us learn the effect of training factors, including how different technologies influence people's ability to master skills.…”

The Impact of Increasing Autonomy on Training Requirements in a UAV Supervisory Control Task

“…Previous research has indicated that the application of hidden Markov models (HMMs), a machine learning modeling approach, can provide greater insight into operator strategies, particularly in terms of supervisory control environments like RESCHU. Indeed, a previous HMM analysis of a similar RESCHU environment was able to determine that, despite having three different types of vehicles under their control, operators tended to cluster the vehicles into two categories, thus reducing their cognitive complexity and workload (Boussemart, Cummings, Las Fargeas, & Roy, 2011).…”