AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Path Channels and Plan Extension Kernels: a Mechanistic Description of Planning in a Sokoban RNN

Mohammad Taufeeque, Aaron David Tucker, Adam Gleave, Adrià Garriga-Alonso

ICLR 2026 · Reinforcement learning › Visual and representation RL · approach: RL for combinatorial routing · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
intervene on 59 path channels via single-step cache ablation (ablation)DRC(3,3) · Boxoban mediumsolve rate drop (%): 57.6 ± 2.8Improved
intervene on 37 non-path channels (ablation)DRC(3,3) · Boxoban mediumsolve rate drop (%): 10.5 ± 1.9No clear effect
intervene on random 37 path channels (count-matched) (ablation)DRC(3,3) · Boxoban mediumsolve rate drop (%): 41.3 ± 2.4Improved
edit pooled next action channels causallyDRC(3,3) · 10,000 Boxoban transitionsintervention score (%): 99.7 ± 0.2Improved
use channel-based intervention instead of linear probe (box movement)DRC(3,3) · 10,000 Boxoban transitionsintervention score (%): 82.5 ± 2.5 -> 86.3 ± 2.1Improved
use agent-movement channel intervention instead of probeDRC(3,3) · 10,000 Boxoban transitionsintervention score (%): 20.7 ± 0.7 -> 53.2 ± 2.1Improved
mean-ablate cell state c (ablation)DRC(3,3) · Boxoban mediumsolve rate drop (%): 21.28% ± 0.04%Improved
mean-ablate forget gate f (ablation)DRC(3,3) · Boxoban mediumsolve rate drop (%): 2.66% ± 0.03%No clear effect
mean-ablate i, j, o gates or hidden state h (ablation)DRC(3,3) · Boxoban mediumsolve rate drop (%): 100.00%Improved
zero-ablate kernels between box-down and box-right channels (remove winner-takes-all) (ablation)DRC(3,3) · two equally viable paths levelplan selectionImproved
scale up plan extension kernels by factor 1.4DRC(3,3) · 40x40 levelsolving long pathsImproved
steer W_ch1 and W_ch2 kernels by factor 1.2DRC(3,3) · 16x16 zig-zag levelsolving levelImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy