The first part of this chapter discussed various views on the relative advantages and disadvantages of isometric versus elastic devices. The two experiments with 6 DOF docking and tracking tasks in this chapter found that the elastic controller, when optimised to accommodate the conflicting requirement of enriching proprioceptive cues and maintaining compatibility with rate control, was indeed superior to the isometric controller. However, the magnitude of this advantage was strongly affected by learning. When enough practice was given to a particular task, subjects performed equally well with both controllers.
The findings in this chapter are in fact in general agreement with some of the recent motor learning theories and empirical studies which compromise earlier more extreme centralist versus peripheralist views. One such example is the schema theory in motor behaviour (Schmidt, 1975, 1988), which states that the human motor control system comprises two types of schemata, recall and recognition schema, similar to the recall and recognition processes found in memory schema research. The recall schema, corresponding to central resources and information outflow, form the relationship between initial conditions, response specifications and response outcome. In contrast, the recognition schema, corresponding to feedback and information inflow, form the relationships among initial conditions, sensory feedback and response outcome. Both recall schema and recognition schema play important roles in motor movement. Their relative contribution depends on the pace of the task and the subjectsÌ experience with the task.
The fact that as learning progresses human motor strategies shift from closed loop behaviour towards open loop behaviour, typically with decreasing importance of visual feedback, has been demonstrated by many researchers. In studying the organisational structure of human motor skills, Pew (1966) , for example, found that motor skills were initially based on feedback but progress towards a hierarchical structure that is more centrally based. Pew reviewed many other motor control theorists' views and asserted: "The underlying themes of these proposals is the hierarchical nature of the control of skilled acts which develop with practice beginning with strict closed-loop control and reaching levels of highly automatized action with occasional 'executive' monitoring". Kohl and Shea (1992) recently replicated Pew (1966) and confirmed Pew's findings.
Based on tracking skill research, Krendel and McRuer (1960; also see Jagacinski and Hah, 1988 for a recent review) proposed their so called "successive organisation of perception (SOP)" theory. Krendel and McRuer identified three modes of tracking behaviour: error nulling, input reconstruction and precognitive behaviour. In the error nulling mode, subjects rely primarily on visual, exteroceptive information to minimise tracking error. In input reconstruction mode, subjects utilise additional proprioceptive information to form control actions. In precognitive mode, subjects depend on open-loop tracking patterns reproduced from memory, while exteroceptive and proprioceptive feedback become less important. With practice, in other words, subjects' behaviour progresses from the error nulling mode to the input reconstruction mode to the precognitive mode, while the source of information used shifts respectively from the visual exteroceptive to the proprioceptive and then to internal memory.
The results of the present experiments appear to support the above theories and findings. In the early learning stage, when control behaviour was dominated by closed-loop inflow processes, the richer proprioceptive feedback from the elastic controller provided an advantage to the subjects in the elastic group relative to the subjects in the isometric group who had less rich proprioceptive feedback. As learning progressed, information feedback became less important and internal open loop mechanisms (motor programs) began to play a more important role, i.e., the motor control behaviour became more open-loop. Similar performance for the elastic and the isometric rate control conditions was therefore found in the later stages of the experiment.
The practical implications of the results can be interpreted a
number of ways. First, the elastic controller is indeed a more
advantageous device for 6 DOF manipulation, in comparison with
the isometric device. In 6 DOF manipulation, the widely presumed
rapidness of isometric devices can hardly be utilised, due to
the complexity of spatially and temporally co-ordinating all six
degrees of freedom. On the other hand, isometric devices are limited
relative to elastic devices in offering less rich proprioceptive
cues, which can have a more pronounced effect in 6 DOF tasks.
Second, with enough practice, performance with isometric devices
can catch up with that of elastic devices but the time required
might be much longer than indicated in the particular experimental
task performed here (i.e. 20 to 40 minutes). In the experiment,
subjects allocated full attention to the task and were coached
thoroughly. In reality, especially in practical computer applications,
where the users might not be trained operators as in aircraft
piloting and teleoperation, users might take a long time to reach
stable performance with isometric devices. Third, equal performance
does not mean equal effort , hence the differences between
the two controllers may reappear when the user has higher workload
or under stress conditions. In the progression-regression theory
of human motor skills, Fuchs (1962) and Jagacinski and Hah (1988
suggest that when under stress, subjects may return to early behaviours.
In the current context, users might therefore perform better with
an elastic device when facing stress. Other authors, such as McKinnon
et al. (1987) and McKinnon and King (1988) , have also pointed
out the probable disadvantages of isometric devices in stressful
teleoperation tasks.