Event
PRISM: Extracting Action Priors from Frozen World Model Representations to Guide Planning
1 reports · 1 sources
Yuhai Wang and other authors propose the PRISM method: a lightweight action head is trained on features from a frozen latent world model encoder to obtain a goal-conditioned action prior; the mean and predictive uncertainty of this prior are then used to guide action search, and the world model evaluates the consequences of actions. The authors report that experiments on four simulated control tasks and two real robot platforms show that this lightweight action head improves planning performance; the same fusion interface can also incorporate a diffusion prior based on the same frozen features without modification.
Report timeline
- PRISM: Prior-Guided Imagination Sampling in World Model Representations
The authors propose PRISM, which trains a lightweight action head on features from a frozen latent world model encoder to obtain a goal-conditioned action prior, then uses its mean and predicted uncertainty to guide action search, with the world model evaluating the consequences of actions. Experiments on four simulated control tasks and two real-robot platforms show that the lightweight action head improves planning performance, and that the same fusion interface can accept a diffusion prior over the same frozen features without modification.
arXiv cs.RO ·