input_ff_rec_robot_nengo_directu_ocl.py 文件源码

def Wdesired(x):
        ''' x is the augmented variable represented in the network 
            it obeys \tau_s x_\alpha = -x_\alpha + Wdesired_\alpha(x) 
            x is related to the original augmented variable \tilde{x} by x_\alpha = varFactors_\alpha \tilde{x}_\alpha 
            where varFactors_alpha = angleFactor | velocityFactor | torqueFactor
            now, original augmented variable obeys \dot{\tilde{x}}=f(\tilde{x})
            so, we have Wdesired_\alpha(x) = \tau_s * varFactor_alpha * f_\alpha(\tilde{x}) + x
        '''
        # \tilde{x} (two zeroes at x[N:N+N//2] are ignored
        xtilde = x/varFactors
        if XY: angles = armAngles(xtilde[:N])
        else: angles = xtilde[:N//2]
        # f(\tilde{x}), \dot{u} part is not needed
        qdot,dqdot = evolveFns(angles,xtilde[Nobs-N//2:Nobs],xtilde[Nobs:],XY,dt)
                                                                        # returns deltaposn if XY else deltaangles
        # \tau_s * varFactors_alpha * f_\alpha(\tilde{x}) + x
        return np.append(np.append(qdot,dqdot),np.zeros(N//2))*varFactors*tau + x
                                                                        # integral on torque u also
                                                                        # VERY IMP to compensate for synaptic decay on torque
        #return np.append( np.append(qdot,dqdot)*tau*varFactors[:Nobs] + x[:Nobs], np.zeros(N//2) )
                                                                        # normal synaptic decay on torque u

    ##### For the reference, choose EITHER robot simulation rateEvolveProbe above
    #####  OR evolve Wdesired inverted / evolveFns using odeint as below -- both should be exactly same