search

danijar / dreamerv3

Mastering Diverse Domains through World Models
https://danijar.com/dreamerv3
MIT License
1.28k stars 219 forks source link

Question about a potential error in the paper #69

Closed till2 closed 1 year ago

till2 commented 1 year ago

Hi Danijar,

I wanted ask whether there is a negative sign missing in front of the $sg(R^\lambda_t)/\max(1,S)$ in eq. 11 in the D-V3 paper. I think the negative return (or advantage) should be minimized, right?

image

Best, Till

danijar commented 1 year ago

You're right, it'll be fixed in the next update.