Expected Free Energy as Belief-Dependent Utility for rho-POMDPs
Artificial Intelligence
Summary
The authors study how agents decide when and what to observe when they cannot see everything around them clearly. Traditional methods either value information only for how it affects rewards later or require manual tuning of how much to care about reducing uncertainty. They show that a method called active inference naturally balances exploring to reduce uncertainty without any tuning by connecting it to a special kind of decision problem (ρ-POMDP) with fixed weights. Their theoretical results and experiments on several problem types demonstrate that this approach matches or beats methods that rely on hand-tuned exploration bonuses, making it easier to apply in real-world tasks like fault detection and medical screening.
Authors
Patrick Cooper, Alvaro Velasquez
Abstract
An agent acting under partial observability must decide when to gather information and which observations are worth their cost. Standard POMDPs value information only through its eventual effect on reward. The $ρ$-POMDP framework instead rewards uncertainty reduction directly, through a belief-dependent utility $ρ$, but in practice both the choice of $ρ$ and the weight placed on it are tuned by hand for every task. We show that active inference removes this tuning entirely. Minimizing Expected Free Energy (EFE) is exactly equivalent to solving a $ρ$-POMDP whose utility is expected information gain, and the exploration weight is fixed at $w=1$ because the variational bound expresses pragmatic and epistemic value in the same units (nats). We prove this equivalence for observe-then-commit POMDPs and extend it to factored observation POMDPs, a broader class that covers interleaved observe-act problems such as non-destructive testing and mobile sensing, where gathering information leaves the hidden state unchanged. Experiments support the theory. Across environments ranging from the classic Tiger problem to RockSample and a new Structural Inspection benchmark with over 65,000 states, the untuned weight matches or outperforms reward-only planning at the same horizon, avoids the over-exploration of bonuses tuned per task, and sits near the reward-maximizing knee of the success-reward Pareto frontier. The practical payoff is an exploration objective that works out of the box. In applications such as fault detection and medical screening, where every test has a price and every missed fault has a cost, EFE supplies a belief-dependent utility that is derived rather than tuned.