INVESTIGATING COORDINATION IN MULTIDEGREE OF FREEDOM CONTROL I: TIME-ON-TARGET ANALYSIS OF 6 DOF TRACKING

INVESTIGATING COORDINATION IN MULTIDEGREE OF FREEDOM CONTROL I:TIME-ON-TARGET ANALYSIS OF 6 DOF TRACKINGShumin ZhaiIBM Almaden Research CenterSan Jose, California, USAJohn W. SendersUniversity of TorontoToronto, Ontario, CanadaIn these two companion papers, methods developed in a series of studies in the 1940’s and 1950’s are applied to theanalysis of 6 DOF control devices used in modern human machine systems such as teleoperation and virtualenvironments. Contrary to the early studies, the current work showed that the simultaneous time-on-target inmultidegree of freedom tracking was higher than the product of component time on target scores. The distribution oflinear correlation coefficients between the tracking errors of different degrees of freedom tended to be skewed towardsthe positive values. These results suggested that subjects’ discoordination in early multidegree of freedom trackingstudies was likely due to the limitation of human machine interfaces at that time. With well designed interfaces,subjects exhibited more coordinated trials than discoordinated trials in multidegree of freedom tracking.INTRODUCTIONIt has always been interesting to read about early effortsin the exploration of a new science. The late 1940's and early1950's was a time of intense interest in human control ofsystems. This interest was driven in part by developments inControl Theory and Information Theory and in part by thedemands of the sponsoring agencies, chiefly military.Behavioral theorists saw in the emerging engineeringapproaches new ways of formalizing human behavior. Thesponsors' concern was largely one of whether a target wouldbe hit. Ease of computation and high face validity led to theuse of "time-on-target" as a measure of "tracking"performance. It was natural that the study of multipledegree-of-freedom (DOF) tasks would arise (e.g. Ellson,1947; Senders, Christensen, & Sabeh, 1955; Senders,Wallis, & Senders, 1956). Poor performance in multipleDOF control was attributed to lack of "coordination". It wasnot clear whether "coordination" was a purely humancharacteristic or arose from interaction with equipmentdesign.The growing number of applications of teleoperation,virtual environments and other 3D human machine systemshas renewed research interest in coordination. Manualcontrollers have been designed to allow translational androtational manipulation in 3D space with 6 DOF (see Brooks& Bejczy, 1985; Jacobus, Riggs, Jacobus, & Weinstein,1992, and Zhai, 1995 for reviews). How well humanoperators can handle all 6 DOF has not been satisfactorilyresolved especially with respect to control with one hand.Rice, Yorchak and Hartley (1986) observed that controlling6 DOF with one hand is difficult. Some teleoperationsystems, such as the Shuttle Remote Manipulator, alsoknown as the “Canadarm”, require two-handed operation:one hand for rotation control and the other for translationcontrol. O’Hara (1987) contradicted Rice’s observation,however, and found no differences between two 3 DOFcontrollers and one 6 DOF controller. To base the design andselection of multidegree of freedom control interfaces on a firmground, human coordination in using 6 DOF control devices is afundamental problem that must be addressed. A quantitative (and preferably analytic) measure ofcoordination that also satisfies intuitive understanding of theconcept is of critical importance. Ellson (1947) derivedindependence measures of percent time-on-target scores (TOT)on the DOFs (azimuth, elevation, and range) of the PedestalSight Manipulation Test (PSMT). He recorded simultaneousTOT's (STOT) in all pairs of dimensions as well as all three atonce, in addition to TOT's in each of the componentdimensions. He then compared STOT scores with the productsof the component TOT scores. His argument was that if thepercent STOT was equal to the product of the component TOTs,then the components may be considered independent(uncorrelated). If greater, they were positively correlated; ifless, negatively correlated. Ellson found that the tracking ofmost subjects was characterized by a slightly negativerelationship: STOT scores were slightly less than the products ofthe component TOT scores. In other words, there was sometendency for the subjects to be off target in one dimension whenon target in another dimension. Gardner found that whensubjects used a joystick control with a cross-pointer display(Gardner, 1950) the function STOT - (TOTx).(TOT)y was notsignificantly different from zero.Senders (Senders et al., 1956) extended Ellson's approach.He placed the scores on a two DOF task into a 2 x 2 matrix ofSTOT and component TOT's and computed the phi-coefficient,O. O is an approximation of the product-moment correlationcoefficient of two arbitrarily dichotomized continuous variables.Senders and colleagues found that subjects’ tracking in a twoDOF task which required manipulation of two knobs to control apointer on a dial, produced more negatively correlated thanpositively correlated trials and, with continued practice, actuallyproduced larger negative correlations.