This example is from Wikipedia and may be reused under a CC BY-SA license.
If the goal state changes significantly, the policy eventually finds itself outside the deadzone and once again must adapt the system.
To avoid continual minor adjustments, a deadzone (+ 10% margin each side of the mean) is used.
In this work, a hydraulic deadzone compensated robust adaptive fuzzy control law has been proposed for locomotion control of hydraulically actuated hexapod demining robot.
The deadzone and adaptation factor concepts are exemplified by the granularity policy.
However, a deadzone can be introduced either side of the goal state such that, once inside this zone, the policy makes no adjustment.
The flattering delay imposed by the deadzone of the proportional electromagnetic valves and the time delay due to flow of oil has also been eliminated to a great extent.
In this paper, a hydraulic deadzone compensated robust adaptive fuzzy control law is proposed for controlling the hydraulically actuated leg joints to achieve an accurate foot trajectory tracking.
While controlling the joints to follow the desired trajectory, delay due to flow of oil and dead zone of the proportional electromagnetic control valve pose a serious problem.
I concede that we need a deadzone for readjustment.
From the
Hansard archive