PlanFM Research Demos
Interactive Research Demonstrations
Focused demonstrations that expose methods, metrics, and generalization behavior for planning-like tasks.
Each card below links to a full interactive walkthrough.
Demo 01
On the Choice of Tokenizers for Generalized Planning via Learned Transition Models
Systematic evaluation of WL, Shortest Path, GraphBPE, and SimHash across Blocks, Gripper, Logistics, and
Visitall with LSTM and XGBoost in delta/state prediction modes.
WL
Shortest Path
GraphBPE
SimHash
LSTM
XGBoost
Demo 02
State-Transition Learning
End-to-end walkthrough of the generalized planning pipeline: symbolic plan generation, state trajectory
reconstruction, ILG construction, WL color refinement, fixed-dimensional embeddings, and neuro-symbolic
transition inference.
Blocksworld
ILG
WL Refinement
State Embeddings
LSTM
XGBoost
