XPost: comp.ai.genetic, comp.ai.neural-nets, comp.robotics.misc   
      
   Autopilot wrote:   
   >   
   > Say we have a vehicle (a car or a boat or a helicopter or an airplane etc)   
   > which reacts to control inputs from a driver/pilots in a particular way   
   > defined by its current state and control input.   
   >   
   > Assume that we have a car travelling on flat road. Our control input would   
   > be a vector of   
   >    Ctrl_Input = [Steering Angle, Gear, Accelerator Value, Brake Value]   
   > and current state would be a vector of   
   >    Current_State =  [Velocity X, Velocity Y, Position X, Position Y,   
   > Acceleration X, Acceleration Y]   
   > the next state of the vehicle after a small time increment "dt" would be   
   >    Next_State =  [Velocity X, Velocity Y, Position X, Position Y,   
   > Acceleration X, Acceleration Y]   
   >   
   > In another word we have a system with following input and output   
   >   
   >     Inputs:   
   >       Ctrl_Inputs   
   >       Current_State   
   >    Output: (obtained after "dt" time)   
   >        Next_State   
   >   
   > Assume that we have as a set of Ctrl_Inputs, Current_State and Next_State   
   > data obtained in an experiment inwhich a human driver driving the vehicle on   
   > a flat  road.   
   >   
   > Given these data sets we should be able to train a neural network so that it   
   > relate inputs to outputs.   
   > My questions are as follows;   
   >    1- Which neural network topology would be good for learning vehicle   
   > behaviour (dynamic model of the vehicle) from above mentioned set of data ?   
   >     2- Number of neurons and number of layers required?   
   >     3- If we know our current state and our desired state (next state) how   
   > we can use the neural network to find required Control_Input vector?   
   >     4- Is there any tutorial/demo/code of this kind ?   
   >     5- Thanks.   
      
   The EvSail and SailChallenge software packages, downloadable gratis from   
   the URL below, are examples of ANNs "trained" to control a virtual   
   sailboat.  Most of this software is applicable to your problem.  These   
   are written in C, and are fully documented.   
      
   Our ANNS are not actually trained.  Instead, a good one is evolved by   
   simulated evolution.  The topology is an MLP with one hidden layer and   
   about 40 neurons.   
      
   Mitchell Timin   
      
   --   
   "Many are stubborn in pursuit of the path they have chosen, few in   
   pursuit of the goal." - Friedrich Nietzsche   
      
   http://annevolve.sourceforge.net is what I'm into nowadays.   
   Humans may write to me at this address: zenguy at shaw dot ca   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)