Some questions about the bootstrapped λ-returns in Dremerv3

[SEU-zxj]

Hello! Thank you for your amazing work and for sharing this implementation 🙏

I have a small question regarding the Critic Learning section in the DreamerV3 paper.

In this section, the λ-return is computed using the following formula:

However, in the DreamerV2 paper, the λ-return is computed differently, as shown here:

My question is about the use of the value term $v_t$ in formula (5) of DreamerV3. In DreamerV2, it is $v_{t+1}$ that appears. I’m wondering why DreamerV3 uses $v_t$ instead?

From my understanding, the reward $r_t$ corresponds to time step $t$, so using $v_t$ in the same step might double-count the same information. Could you clarify the motivation or reasoning behind this change?

Looking forward to your response, and thank you again for this great work!