Question about V2 Curve Interp: does nearest_traj final yaw affect the interpolated trajectory?

[lin-zy23]

Hi, thanks for releasing TrajTok.

I have a question about the V2 token construction in src/smart/tokens/trajtok.py.

In get_trajtok_vocab(), for expanded cells or cells with noisy yaw, the code calls interpolate_curve(grid_end_x, grid_end_y, nearest_traj[-1, 2]). This seems to pass the final yaw of the nearest valid trajectory as theta.

However, in interpolate_curve(), the default values are:

weight_factor0 = 0
weight_factor1 = 0

As a result, both Hermite tangents become zero, so theta appears to have no effect on the interpolated xy curve. Changing theta does not change the generated centerline unless weight_factor1 is non-zero.

Could you clarify whether this is intentional? Specifically:

1. Should V2 Curve Interp in the paper be interpreted as a zero-tangent smoothstep interpolation to the endpoint, or was theta = nearest_traj[-1, 2] intended to constrain the endpoint heading?
2. If the endpoint heading is indeed supposed to matter as described in the paper, what weight_factor0 / weight_factor1 values were used for the reported experiments?

This matters because with the current default implementation, V2 tokens visually look like straight centerlines rather than heading-conditioned curved trajectories.

Thanks!