A Sokoban-playing recurrent network stores future box moves in identifiable 'path channels' and builds plans by extending activations forward from boxes and backward from goals, with negative activations pruning failed paths.
The heuristic varies by algorithm: • For A*, it is distance(n) +heuristic(n)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Path Channels and Plan Extension Kernels: a Mechanistic Description of Planning in a Sokoban RNN
A Sokoban-playing recurrent network stores future box moves in identifiable 'path channels' and builds plans by extending activations forward from boxes and backward from goals, with negative activations pruning failed paths.