Knowing When to Stop Thinking: A Dissection of Expected Free Energy as a Test-Time Compute Controller for Latent World-Model Planning
A planner using a learned world model can improve its predicted outcome while choosing a worse action.
ProofPaper ↗
Key points
- We study when expected free energy (EFE) can help decide how much computation to spend on a plan.
- An EFE-inspired objective combines goal distance with disagreement among stochastic predictions.
- We test candidate scoring, stopping, and online predictor updates separately.
- Stopping when elite cost stalls reduces refinement on tasks whose success plateaus with depth, with mean success close to fixed-depth references.
Sources (1)
- [1]Knowing When to Stop Thinking: A Dissection of Expected Free Energy as a Test-Time Compute Controller for Latent World-Model PlanningarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 10:56 AM
A planner using a learned world model can improve its predicted outcome while choosing a worse action.
We study when expected free energy (EFE) can help decide how much computation to spend on a plan.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026Learning the 4D World through Point Trajectory Forecasting
- Oct 19, 2026VPFlow: Value-Predictive Planning with Imperfect World Models for Offline Goal Reaching
- Oct 19, 2026Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
- Oct 19, 2026Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
- Oct 19, 2026State World Models: Execute the Rules, Generate the Pixels
- Oct 19, 2026AnyGoal: On World Model Generalization Beyond Expert Rollouts