/// <summary>
/// Reset the state of the gym.
/// </summary>
/// <param name="bGetLabel">Not used.</param>
/// <returns>A tuple containing state data, the reward, and the done state is returned.</returns>
public Tuple <State, double, bool> Reset(bool bGetLabel)
{
double dfX = randomUniform(-0.05, 0.05);
double dfXDot = randomUniform(-0.05, 0.05);
double dfTheta = randomUniform(-0.05, 0.05);
double dfThetaDot = randomUniform(-0.05, 0.05);
m_nStepsBeyondDone = null;
m_nSteps = 0;
m_state = new CartPoleState(dfX, dfXDot, dfTheta, dfThetaDot);
return(new Tuple <State, double, bool>(m_state.Clone(), 1, false));
}
/// <summary>
/// Step the gym one step in its simulation.
/// </summary>
/// <param name="nAction">Specifies the action to run on the gym.</param>
/// <param name="bGetLabel">Not used.</param>
/// <param name="propExtra">Optionally, specifies extra parameters.</param>
/// <returns>A tuple containing state data, the reward, and the done state is returned.</returns>
public Tuple <State, double, bool> Step(int nAction, bool bGetLabel, PropertySet propExtra = null)
{
CartPoleState state = new CartPoleState(m_state);
double dfReward = 0;
processAction((ACTION)nAction);
double dfX = state.X;
double dfXDot = state.XDot;
double dfTheta = state.Theta;
double dfThetaDot = state.ThetaDot;
double dfForce = m_state.ForceMag;
double dfCosTheta = Math.Cos(dfTheta);
double dfSinTheta = Math.Sin(dfTheta);
double dfTemp = (dfForce + m_dfPoleMassLength * dfThetaDot * dfThetaDot * dfSinTheta) / m_dfTotalMass;
double dfThetaAcc = (m_dfGravity * dfSinTheta - dfCosTheta * dfTemp) / (m_dfLength * ((4.0 / 3.0) - m_dfMassPole * dfCosTheta * dfCosTheta / m_dfTotalMass));
double dfXAcc = dfTemp - m_dfPoleMassLength * dfThetaAcc * dfCosTheta / m_dfTotalMass;
dfX += m_dfTau * dfXDot;
dfXDot += m_dfTau * dfXAcc;
dfTheta += m_dfTau * dfThetaDot;
dfThetaDot += m_dfTau * dfThetaAcc;
CartPoleState stateOut = m_state;
m_state = new CartPoleState(dfX, dfXDot, dfTheta, dfThetaDot);
bool bDone = false;
if (dfX < -m_dfXThreshold || dfX > m_dfXThreshold ||
dfTheta < -m_dfThetaThreshold || dfTheta > m_dfThetaThreshold)
{
bDone = true;
}
if (!bDone)
{
dfReward = 1.0;
}
else if (!m_nStepsBeyondDone.HasValue)
{
// Pole just fell!
m_nStepsBeyondDone = 0;
dfReward = 1.0;
}
else
{
if (m_nStepsBeyondDone.Value == 0)
{
m_log.WriteLine("WARNING: You are calling 'step()' even though this environment has already returned done = True. You should always call 'reset()'");
}
m_nStepsBeyondDone++;
dfReward = 0.0;
}
m_nSteps++;
m_nMaxSteps = Math.Max(m_nMaxSteps, m_nSteps);
stateOut.Steps = m_nSteps;
return(new Tuple <State, double, bool>(stateOut.Clone(), dfReward, bDone));
}