In the simple code version (model "BCPolyCode3.cr2"), the translations and the two rotation angles are distributed among two actors: Parent "pivot" moves on the curve and adjusts the yaw orientation. This is angle "beta". So, we have a rotateY channel "yr_beta" on actor "pivot". Child "dummy" follows the translations and y-rotation of "pivot" and adjusts the pitch orientation, that is, the rotateZ channel "zr_alpha" on actor "dummy". Actually, the code sets the zRot channel "zr_alpha" to zr_alpha = deg_alpha for positive dx and zr_alpha = deg_alpha +180 for negative dx to deal with negative dx values. The channel code on "BODY" is: dx_sign = -dx_var * 1e10 {dx_sign: 0, 180} where "dx_var" is a derivative of the curve and channel "dx_sign" will be set to 0 or 180 depending on the sign of dx_var. The "zRot" channel on "dummy" is: zr_alpha = deg_alpha + dx_sign. This setup works with negative dx values, that is, the CP aa can be placed "behind" the first knot A, and CP cc can be placed in front of knot C. However, if dx is negative, there is a special condition when dx changes the sign. The slope of the curve becomes undefined for a short time. This means, the model flips by 180 degrees about the y-axis. I didn't try to fix this glitch, and it's not the worst problem of this model, anyway.