1. There is a interesting diversity in input devices with 6 Degree-of-freedom (6 DOF). Many different devices (and potentially more) have been developed for the same purpose of manipulating objects in 3D environments (VR, teleoperation, 3D interactive graphics etc). What concerns the user is the usability (or human performance) of these devices. We want to know which is "good" and which is "bad", or more precisely, which is "good" or "poor" for what functions [1] , [2] .
2. There are potentially endless products out there and it is not too productive to compare every pairs of them. So a useful thing to do is classifying them from user's point of view. There are many dimensions one can classify them, such as sensoring technology, size, range, and so on. I think two of the most interesting human factors dimensions are the physical resistance and transfer (or mapping) function, because they cause different ways (strategies) of using the input device but the performance difference is not obvious. [1] , [2] .
3. When the physical resistance of the input device is infinite. The device does not move. It stays in a constant location. This is called isometric devices in human factors and human motor control literature. With isometric devices, the force and torque that the user applies get converted to object motion. Spaceball is a typical isometric device. When the physical resistance of the input device is zero, the input device moves freely with the user's limb. In this case, the device is isotonic. Gloves and "bats" are examples of isotonic devices. A mouse is a 2D isotonic device. There is a long standing controversy between the relative (human) performance with isometric vs. isotonic devices among engineering psychologiests. The literature goes back (at least) to 1950s (Check Poulton 1974 for review). Of course that was about 1 or 2 DOF devices. There is a continuum between isometric and isotonic devices. Their resistance is neither zero nor infinite. When the resistant force is proportional to the displacement of the handle, it is an "elastic" device.
4. Another important dimension is the transfer (or mapping) function. After certain variable (force or displacement) is sensed from the human limb, there are many ways of mapping the data to object motion. One way is to proportionally map the sensed input variable to the location/orientation of controlled object, either absolutely or relatively and either linearly or non-linearly (e.g. power mouse). That is called position control (or zero order control, as in control theory).
If the input variable is mapped to the velocity of the objects, (i.e. there is a integral in the transfer function), it is rate control (or first order control).
Position control has limited output range. Our hands can move and rotate only to a limited magnitude. Clutch and de-clutch is a way to repeatedly map the same input change to greater output range. With a mouse, one lifts (declutch) and put down to a new location (clutch). With a glove, a button will do it. Clutch and declutch is also called indexing in human factors and telerobotics literature.
Rate control has unlimited output range so it is auto-indexed.
5. Isometric device works better in rate control mode and isotonic device is more compatible with position control mode. This is obvious in some ways. But surprisingly, the early researchers never (to my knowledge) put the resistance dimension and the transfer function dimension in one study. In a 6 DOF docking experiment, we found the two dimensions interact very strongly [2] . In light of this result, we can explain part of the controversy between isometric devices and isotonic devices. Those researchers who found superority with isometric devices often used rate control and other researchers who concluded performance advantage with isotonic devices often used position control.
The key to rate control is self-centering. Isometric devices are self-centred.
6. Elastic devices, such as the Egg , is also self-centred. But it involves both movement and force in its operation. Such devices may afford better "proprioception" to the user and are therefore easier to learn, according to the psychomotor learning literature. Some of my experiments confirmed this hypothsis [3] , [4].
7. Following are some practical and simplified conclusions: [5]
A. Isotonic position control (e.g. glove) is the most "direct" interface. The major problem is fatigue. Limited output range (therefore clutch needed) is also a disadvantage. This type of devices are good for VR games, users can walk up and play without much learning.
B. Isometric rate control is less fatiguing. Major problem is lack of feedback (proprioception). It takes a while for the user to acquire the skill. Once they learned, it works reasonably well for simply tasks (e.g. docking).
C. Isotonic rate and Isometric position are incompatible combinations.
D. An improvement over isometric rate control is ELASTIC rate control. It offers much better proprioception and therefore easier to learn. The key is once again the combination of rate control with the self-centered property of elastic controller. Elastic rate control will be good for applications like CAD and data visualisation in which the user may work for hours.