AAAI paper: probabilistic hierarchical goal network planning with UCT, compressed search wins at scale
xuanalogue · x · 2026-09-16
A University of Maryland / U.S. Naval Research Laboratory paper at AAAI, "Probabilistic Hierarchical Goal Network Planning with UCT," extends hierarchical goal networks (HGNs) to stochastic settings:
- Formalizes probabilistic HGN planning with action-insertion semantics, letting probabilistic planners exploit domain knowledge from goal decomposition.
- Proposes two UCT-based solvers: an asymptotically optimal one and a compressed, shared-value approach that optimizes each goal separately within the goal-subgoal hierarchy.
- On modified FOND HTN benchmarks, the compressed search converges faster and outperforms the asymptotically optimal search on larger problems, suggesting compression helps in anytime settings with stochastic action outcomes.
The thread also references a mini-paper on goal-based hierarchical RL (combining ACGVF with hierarchical HMMs, no experiments yet); both share the assumption that goals can be chosen as agent-internal actions.
More from Research
- Combining Continued Pretraining with RAG: Teaching Qwen 3.5 4B a Fictional Subway Map — funJS · 2026-09-16
- Therna Bio opens Chronos platform, releases gene expression datasets with millions of measurements — AllThingsApx · 2026-09-16
- GlossoGen: open-source platform shows AI agents spontaneously inventing new languages under pressure — EliasEskin · 2026-09-16
- Paper: upsampling alignment discourse in pretraining cuts misalignment from 45% to 9% — TuhinChakr · 2026-09-16
- One researcher taught world model Odyssey-3 to drive in India with just 20 hours of data — soleio · 2026-09-16
- Continual Learning in Fruit Fly Brain Decoded, Framed as Missing Piece for AGI — skolnaja · 2026-09-16