SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.15421 · ICML · 2024

Planning in a recurrent neural network that plays Sokoban

Adam Gleave, Adrià Garriga-Alonso, Chris Cundy, Philip Quirke, Mohammad Taufeeque, Maximilian Li, Aaron Tucker

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 2 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2407.15421.json · badge: /api/ran/2407.15421.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
alignmentresearch/learned-planner canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
alignmentresearch/train-learned-planner canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
downsample Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
alignmentresearch/learned-planner/learned_planner/interp/collect_dataset.py
code served (permissive licence) · get_code("c65d2058bbfc8234")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 4, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch as th

th.manual_seed(0)

# Create a small observation tensor with shape (..., H, W)
# Using shape (2, 4, 8, 6) to represent batch=2, channels=4, height=8, width=6
obs = th.randn(2, 4, 8, 6)

# Target downsampling dimensions
h = 4
w = 3

# Boolean flag for average pooling vs max pooling
avg = False

ARGS = {
    "obs": obs,
    "h": h,
    "w": w,
    "avg": avg,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 4, 4, 3) — shape (2, 4, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_probe_preds Ran
ran, honoured the contract we drafted
alignmentresearch/learned-planner/plot/behavior_analysis.py
code served (permissive licence) · get_code("282910eafb91ddc1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import torch
import torch.nn as nn

# Define a minimal probe class that has a predict method
class MinimalProbe:
    def predict(self, X):
        # X shape: (batch, features)
        # Return predictions of shape (batch,)
        return np.zeros(X.shape[0])

# Create a mock LSTM state object
class MockLSTMState:
    def __init__(self, h, c):
        self.h = h
        self.c = c

# Build the probe
probe = MinimalProbe()

# Build the carry (list of LSTM states)
# Each state has h and c tensors
# Using distinct sizes: h is (2, 4) and c is (2, 3) for clarity
h_tensor = torch.randn(2, 4)
c_tensor = torch.randn(2, 3)
state1 = MockLSTMState(h_tensor, c_tensor)

h_tensor2 = torch.randn(2, 4)
c_tensor2 = torch.randn(2, 3)
state2 = MockLSTMState(h_tensor2, c_tensor2)

carry = [state1, state2]

ARGS = {
    "probe": probe,
    "carry": carry
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
is_grid_wise_dataset Ran
ran, violated the contract we drafted
alignmentresearch/learned-planner/learned_planner/interp/train_probes.py
code served (permissive licence) · get_code("8065cb438e9a52d2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    'dataset_name': 'agents_future_position_map'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
unreplicate Ran
ran; our contract draft was wrong, not the code
alignmentresearch/train-learned-planner/cleanba/cleanba_impala.py
pointer only (licence: NOASSERTION) · get_code("dec2d4521ca0eadc")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, alignmentresearch/train-learned-planner (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Planning is essential for solving complex tasks, yet the internal mechanisms underlying planning in neural networks remain poorly understood. Building on prior work, we analyze a recurrent neural network (RNN) trained on Sokoban, a challenging puzzle requiring sequential, irreversible decisions. We find that the RNN has a causal plan representation which predicts its future actions about 50 steps in advance. The quality and length of the represented plan increases over the first few steps. We uncover a surprising behavior: the RNN "paces" in cycles to give itself extra computation at the start of a level, and show that this behavior is incentivized by training. Leveraging these insights, we extend the trained RNN to significantly larger, out-of-distribution Sokoban puzzles, demonstrating robust representations beyond the training regime. We open-source our model and code, and believe the neural network's interesting behavior makes it an excellent model organism to deepen our understanding of learned planning.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2407.15421")
get_code_for_paper("2407.15421")
have("2407.15421")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2407.15421.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2407.15421.svg)](https://syntology.ai/paper/2407.15421)

Connect an agent — have() is free.