CARLA Applet 2
Instructions
This Java applet demonstrates the CARLA (Continuous Action Reinforcement Learning Automata) in non-stationary
environments with an extension for a variable action range. The red line shows the simple cost environment for this
one dimensional demonstration problem. The aim of the CARLA is to determine the location of the function minima -
they are located at -700 and 200. The white line on the
screen is the zero axis line. The probability density function is shown in green
or yellow with the colour of the distribution changing as the
environment changes. The CARLA has a variable action range and expands or contracts in order to centre the distribution
around the maximal value of probability. The current search range and current maximum value are shown in the lower left window with
the number of iterations displayed in the lower right window. The probability distribution snakes about to determine the optimal value.
With no noise on the cost function and a fixed environment the range will continually shrink towards the optimal value however here the
environment changes and so the distribution range does not continually reduce.
For futher information on the CARLA please refer to my publications
or contact me at mnwhowell@gmail.com.