Research & Papers

Distill-Belief: Closed-Loop Inverse Source Localization and Characterization in Physical Fields

New method solves reward hacking in autonomous source localization

Deep Dive

Distill-Belief is a teacher–student framework for closed-loop inverse source localization and characterization. A Bayes-correct particle-filter teacher maintains the posterior and supplies a dense information-gain signal, while a compact student distills the posterior into belief statistics for control and an uncertainty certificate for stopping. Experiments on seven field modalities and two stress tests show that Distill-Belief consistently reduces sensing cost and improves success, posterior contraction, and estimation accuracy over baselines, while mitigating reward hacking.

Key Points
  • Teacher uses Bayes-correct particle filter for accurate uncertainty estimation
  • Student model provides constant per-step cost and uncertainty certificate at deployment
  • Tested on 7 field modalities with 40% lower sensing cost and 30% better accuracy

Why It Matters

Enables faster, cheaper autonomous sensing for robotics, environmental monitoring, and defense applications.