int_problem.m

来自「approximate reinforcement learning」· M 代码 · 共 44 行

function out = int_problem(what)
% Integrator problem setup.
%   OUT = DOUBLEINT_PROBLEM(WHAT)
% This function conforms to the specifications established by SAMPLE_PROBLEM.

maxx = 5;
maxu = 2;
gridstep = .5;
xgrids = {-maxx:gridstep:maxx};
ugrids = {-maxu:gridstep:maxu};
x0 = (2*rand) * maxx - maxx;
gamma = 0.98;
switch what
    case 'model'
        phys.maxx = maxx;
        phys.maxu = maxu;
        phys.K = 2;       % command gain
        
        % reward specification
        % config 1: LQR
        goal.Q = .1; goal.R = .05;
        goal.zeroband = 0;
        goal.zeroreward = 0;
        % config 2: min-time
%         goal.Q = 0; goal.R = 0;
%         goal.zeroband = 0.1;
%         goal.zeroreward = 10;

        Ts = 1;
        fun = @int_mdp;
        out = varstostruct('Ts', 'fun', 'phys', 'goal');

    case 'fuzzy'
        cfg.xgrids = xgrids;
        cfg.ugrids = ugrids;
        cfg.gamma = gamma;
        cfg.x0 = x0;
    
        out = cfg;
        
end;


% END doubleint_problem(), RETURNING out ====================================