question about line 3 of Algorithm 2 and the code

[Jushan-Randy-Chen]

Hi,

Thanks for sharing this interesting work!

While trying to understand the code in mbd_planner.py, I have a question regarding line 105 and I hope you could clarify my concerns!

In line 3 of Algorithm 2 of the paper, the variance of the Gaussian distribution is (1/alpha_bar[i-1]) - 1, or equivalently (1-alpha_bar[i-1]/alpha_bar[i-1]. However, in line 105 of mbd_planner.py, we have:

Y0s = eps_u * sigmas[i] + Ybar_i

where sigmas is defined as:

sigmas = jnp.sqrt(1 - alphas_bar)

Therefore I think there is a mismatch and we should downscale sigmas by a factor of \sqrt(alpha_bar[i-1]:

Y0s = eps_u * sigmas[i]/sqrt(alpha_bar[i-1]) + Ybar_i

I wonder if you could please clarify this? thanks!

[purewater0901]

Hello. I am not the author, but I think your suggestion is correct. The sigma in the code does not match the equation in the paper.