Adaptive Dynamic Programming (Approximate Dynamic Programming,
Adaptive Critic Designs, Neuro-Dynamic Programming):
- Here I post some Matlab programs for ADP. The programs are done
for the inverted pendulum problem which has been widely considered as a good
benchmark problem for learning control.
There are five programs that you need to download. In order to run these
programs, you must have Matlab and Neural Network Toolbox installed.
The five programs are
- README - Read this file first
- main.m - Main program
- wnet.m - To form a network consting of critic and action networks
- plant.m - The cart-pole system
- atraingd.m - Training program for action network (modified from the toolbox)
I have tried to put as much comments in the program as I can.
Please email your feedbacks to derongliu@gmail.com
Click here to go ADP depository.
Reinforcement Learning:
- If you would like to try the approach described in
Barto, Sutton, and Anderson,
"Neuronlike Adaptive Elements That Can Solve Difficult Learning Control Problems,"
IEEE Trans. Systems, Man, and Cybernetics,
Vol. SMC-13, pp. 834-846, Sept.-Oct. 1983;
please download the following three files: These are MATLAB files.
For C or C++ files, you can find more information in main.m.
Please read the instructions in main.m