Multiple BP-FP Pass required until convergence achieved. For a moving target Multiple-DDP runs are mandatory since the target location evolves probabilistically

In RETRO, distribution shift is included through KL div. in Value function and controls are refined through non-iterative update rule