Imagination Based Policy Optimization With Latent Rollouts

via “imagination-based policy optimization with latent rollouts”

* ⏫ 02/2023: [Grounding Large Language Models in Interactive Environments with Online RL (GLAM)](https://arxiv.org/abs/2302.02662)

Unique: DreamerV3 uses a two-headed value function (critic and target) trained on imagined trajectories with symlog scaling, enabling stable policy optimization without explicit target networks or replay buffers. The imagination rollout is differentiable end-to-end, allowing gradients to flow through the world model during policy updates (though the world model is typically frozen).