Search and Belief-Space Planning › Key Parameters
Michael BrenndoerferJuly 9, 20261 min read
World Models HandbookAll chapters Part 7 of 12Planning and Agency
Search and Belief-Space Planning › Key ParametersOne section · Sections in this chapter10 / 13 ↓
Sections in this chapter13
Reading tools Definitions & inline links
Choose your expertise level to adjust how many terms are explained. Beginners see more tooltips, experts see fewer to maintain reading flow. Hover over underlined terms for instant definitions.
Article links
Make inline references clickable
The key parameters for learned search guidance are:
- policy head architecture: Size and depth of the network producing PUCT action priors. More capacity can represent more complex state-to-action mappings but costs more to train and evaluate.
- lr: Learning rate for the policy network optimizer. Controls how quickly the prior adapts to training targets.
- epochs: Number of training passes over the supervised targets. Too few can underfit; additional epochs reduce training loss here, but held-out evaluation is needed to diagnose overfitting or overconfidence.
- PUCT coefficient c: Exploration coefficient used when the learned prior drives selection. Sets how strongly the prior is allowed to outweigh the sampled Q values.