Delay turns closed-loop control into predictive control

What it is

Communication latency turns teleoperation from direct closed-loop control into what is effectively predictive control: after issuing a command, the operator must wait out the round-trip delay before seeing its visual feedback, and the longer that window gets, the less the operator can rely on real-time feedback to fine-tune the next move, leaving prediction of the remote system's behavior as the only basis for the following command.

Why it happens

The threshold for this shift connects directly to a classic result from supervisory control theory — early work on long-delay teleoperation of vehicles such as lunar rovers found that once round-trip delay exceeds roughly the order of magnitude of human closed-loop reaction time (degradation becomes marked somewhere in the range of a few hundred milliseconds to one or two seconds), operators spontaneously switch from continuous control to a move-and-wait strategy: move a bit, wait for feedback, move again. This strategy is effectively the operator giving up on real-time feedback and falling back to open-loop, predictive operation to avoid the overcorrection and oscillation that delayed feedback produces — continuing to correct in closed-loop fashion means that by the time a correction command arrives, the feedback for the next deviation the operator sees already reflects an even older state, so each correction makes things worse rather than better. The switch is not operator error; it is a rational adaptation — move-and-wait trades speed for realigning each action with its matching feedback as a single judgeable unit, avoiding a spiral of corrections stacked on stale error.

Studying it

A common design systematically manipulates round-trip communication delay (0, 0.5, 1, 2, 5 seconds, for example), measuring task completion time, the number of discrete moves (whether move-and-wait is adopted more often), and rates of overshoot and oscillation. Other work specifically calibrates the delay threshold at which operators spontaneously switch from continuous control to move-and-wait, finding that this threshold shifts with task precision requirements and operator experience — the higher the precision demand, the lower the delay at which the switch happens. Command granularity — the displacement associated with a single action — is another common manipulated variable: when operators can set their own step size, larger delay leads them to spontaneously shrink and moderate that step size, which is itself a continuous-dial version of move-and-wait. Studies also need to manipulate pure round-trip delay and delay variability separately, since conflating the two removes the reference point needed to say what delay actually triggers the switch.

Where it stops holding

How much delay matters depends heavily on the nature of the task — tasks that need continuous closed-loop fine-tuning (fine assembly, high-speed obstacle avoidance) are extremely delay-sensitive, while discretized, high-level command tasks (setting a waypoint, selecting a path) tolerate much longer delay. This does not apply to supervisory-control scenarios that are already designed around discrete commands, since operators there never depended on a continuous feedback loop in the first place, so the marginal effect of delay is small. Tolerable delay also scales with how reversible the consequences are: a deep-space probe's actions can usually be redone slowly, whereas tasks involving living tissue or fragile material push operators toward move-and-wait earlier and more conservatively at the very same delay, so no single delay threshold transfers across domains.

Applying it

When designing the task flow, evaluate whether the actual communication delay calls for redesigning a continuous-control task into a discrete-command one (turning "real-time driving" into "mark a waypoint, then autonomous navigation") rather than forcing the operator to grind through delay with move-and-wait. Where delay cannot be removed, provide an explicit switch between rate control and position control so operators can deliberately drop into a more conservative mode at high delay, instead of discovering move-and-wait on their own inside a continuous-control interface. Validate by measuring, at the target delay, whether operators have already spontaneously shifted to move-and-wait — if so, the current continuous-control design no longer fits that delay level and the interaction needs to be redesigned rather than left for operators to adapt to through experience.

Related

• Same group: X7.02.2 Compensating delay requires a predictive display or adjustable autonomy · X7.02.3 Variable latency is harder to adapt to than constant latency
• Nearby: X7.03 Force and haptic feedback · X4.04 Shared control
• Search terms: teleoperation · supervisory control · move-and-wait strategy · latency compensation