SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.19396 · 2024

NAVIX: Scaling MiniGrid Environments with JAX

arXiv · PDF · Open in the Atlas

Code that ran

9 of 9 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; 2 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 6 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/2407.19396.json · badge: /api/ran/2407.19396.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
epignatelli/navix canonical (classifier label)
classifier basis: census
9 ran · 0 no recorded run
FunctionStatusWhere it lives
categorical_kl Ran
ran, honoured the contract we drafted
epignatelli/navix/navix/agents/models.py
code served (permissive licence) · get_code("0cb886a5040816ee")
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 ArrayImpl, 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 jax
import jax.numpy as jnp
import distrax

# Set seed for reproducibility
jax.config.update('jax_platform_name', 'cpu')

# Create small categorical distributions
# num_classes = 4, batch_size = 2
num_classes = 4
batch_size = 2

# Create logits for posterior and prior distributions
# Shape: (batch_size, num_classes)
post_logits = jnp.array([
    [1.0, 2.0, 0.5, 1.5],
    [0.8, 1.2, 2.0, 0.5]
])

prior_logits = jnp.array([
    [1.5, 1.0, 1.2, 0.8],
    [1.0, 1.5, 0.9, 1.1]
])

# Create distrax.Categorical distributions
post = distrax.Categorical(logits=post_logits)
prior = distrax.Categorical(logits=prior_logits)

ARGS = {
    'post': post,
    'prior': prior
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
distrax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, 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
coordinates Ran
ran, honoured the contract we drafted
epignatelli/navix/navix/grid.py
code served (permissive licence) · get_code("08d7716969571f07")
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 tuple, ndim 3, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 jax.numpy as jnp
import jax

# Set seed for reproducibility
jax.config.update('jax_platform_name', 'cpu')

# Create a small 2D grid array with shape (height=4, width=3)
grid = jnp.array([[1, 2, 3],
                   [4, 5, 6],
                   [7, 8, 9],
                   [10, 11, 12]], dtype=jnp.int32)

ARGS = {
    "grid": grid
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 3, dtype family int — a tuple output has no array shape of its own; none is stated
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
coordinates_from_idx Ran
ran, violated the contract we drafted
epignatelli/navix/navix/grid.py
code served (permissive licence) · get_code("8baa2d1b3a965700")
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 ArrayImpl, ndim 1, dtype family int, 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 jax.numpy as jnp
import jax

# Set random seed for reproducibility
jax.config.update('jax_platform_name', 'cpu')

# Create a small 2D grid
# grid.shape[1] is used to get the width for divmod operation
grid = jnp.ones((5, 4), dtype=jnp.int32)  # height=5, width=4

# Create a flat index
# Valid indices should be in range [0, height*width) = [0, 20)
idx = jnp.asarray(7, dtype=jnp.int32)  # flat index

ARGS = {
    "grid": grid,
    "idx": idx,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family int, 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
field Ran
ran; our contract draft was wrong, not the code
epignatelli/navix/navix/components.py
code served (permissive licence) · get_code("78338016bc57d3e0")
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 Field. The image it ran in was not recorded. If your copy returns the same type 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
import dataclasses
from typing import Tuple

torch.manual_seed(0)

# The function 'field' takes:
# 1. shape: Tuple[int, ...] - a tuple of integers representing the shape
# 2. **kwargs: additional keyword arguments to pass to dataclasses.field

# Looking at the function body:
# - It calls dataclasses.field() with metadata={'shape': shape} and **kwargs
# - The shape parameter is a tuple of ints
# - kwargs are forwarded to dataclasses.field, which accepts things like:
#   default_factory, default, init, repr, hash, compare, metadata, kw_only, etc.

# Construct valid arguments
shape_value = (2, 3)  # A tuple of ints representing shape (e.g., 2D position)

kwargs_dict = {
    'default_factory': list  # A common kwarg for dataclasses.field
}

ARGS = {
    'shape': shape_value,
    'default_factory': kwargs_dict['default_factory']
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Field — a Field output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Field output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
idx_from_coordinates Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
epignatelli/navix/navix/grid.py
code served (permissive licence) · get_code("a6a3563b64257584")
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 ArrayImpl, ndim 1, dtype family int, 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 jax
import jax.numpy as jnp
import numpy as np

# Create a small 2D grid (4 height x 3 width)
grid = jnp.array([[1, 2, 3],
                   [4, 5, 6],
                   [7, 8, 9],
                   [10, 11, 12]], dtype=jnp.int32)

# Create a batch of 2D coordinates (batch_size=2, 2 coordinates per point)
# Each coordinate is [row, col]
coordinates_batch = jnp.array([[0, 1],
                                [2, 2]], dtype=jnp.int32)

ARGS = {
    "grid": grid,
    "coordinates": coordinates_batch
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family int, 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
probe_hparam_field_stats Ran
ran; our contract draft was wrong, not the code
epignatelli/navix/navix/es.py
code served (permissive licence) · get_code("babff6f579982c24")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 jax
import jax.numpy as jnp
import distrax

# Create a simple PRNG key
key = jax.random.PRNGKey(0)

# Create a dictionary of distributions for hyperparameters
hparams_distr = {
    "learning_rate": distrax.Uniform(low=jnp.array(0.001), high=jnp.array(0.1)),
    "gae_lambda": distrax.Uniform(low=jnp.array(0.0), high=jnp.array(1.0)),
    "entropy_coeff": distrax.Uniform(low=jnp.array(0.0), high=jnp.array(0.01)),
}

n_probe = 32

ARGS = {
    "hparams_distr": hparams_distr,
    "n_probe": n_probe,
    "key": key,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
distrax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
sample_antithetic_candidates Ran
ran; our contract draft was wrong, not the code
epignatelli/navix/navix/es.py
code served (permissive licence) · get_code("bf7f4b56213cfb45")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 jax
import jax.numpy as jnp

# Initialize JAX PRNG key
key = jax.random.PRNGKey(0)

# Define theta (current per-field mean) - dict with scalar arrays
theta = {
    'param_a': jnp.array(0.5),
    'param_b': jnp.array(1.2),
    'param_c': jnp.array(-0.3),
}

# Define scale (per-field noise scale) - dict with scalar arrays
scale = {
    'param_a': jnp.array(0.1),
    'param_b': jnp.array(0.2),
    'param_c': jnp.array(0.15),
}

# Define lo (per-field lower bound) - dict with scalar arrays
lo = {
    'param_a': jnp.array(0.0),
    'param_b': jnp.array(0.0),
    'param_c': jnp.array(-1.0),
}

# Define hi (per-field upper bound) - dict with scalar arrays
hi = {
    'param_a': jnp.array(1.0),
    'param_b': jnp.array(2.0),
    'param_c': jnp.array(1.0),
}

# Population size (must be even)
pop_size = 4

# Noise scale
sigma = 0.5

ARGS = {
    'theta': theta,
    'scale': scale,
    'lo': lo,
    'hi': hi,
    'pop_size': pop_size,
    'sigma': sigma,
    'key': key,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
straight_through_sample Ran
ran; our contract draft was wrong, not the code
epignatelli/navix/navix/agents/models.py
code served (permissive licence) · get_code("27056902b3134a1a")
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 ArrayImpl, ndim 1, dtype family float, shape (4,) 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 jax
import jax.numpy as jnp
import distrax

# Set JAX to CPU mode
jax.config.update('jax_platform_name', 'cpu')

# Create a random number generator key
rng = jax.random.PRNGKey(0)

# Create a Categorical distribution with 4 categories
# We need a distribution with probs attribute and sample method
probs = jnp.array([0.1, 0.2, 0.3, 0.4])
dist = distrax.Categorical(probs=probs)

ARGS = {
    "dist": dist,
    "rng": rng,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
distrax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family float, shape (4,) — shape (4,) 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
unimix_categorical Ran
ran; our contract draft was wrong, not the code
epignatelli/navix/navix/agents/models.py
code served (permissive licence) · get_code("bbd549a9780617e8")
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 Categorical. The image it ran in was not recorded. If your copy returns the same type 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 jax
import jax.numpy as jnp
import distrax

# Set seed for reproducibility
jax.config.update('jax_platform_name', 'cpu')

# Create small test logits tensor
# Shape: (batch_size=2, num_classes=4)
logits = jnp.array([
    [1.0, 2.0, 0.5, -1.0],
    [-0.5, 1.5, 2.5, 0.0]
], dtype=jnp.float32)

# Unimix parameter (fraction of uniform probability mass to mix in)
unimix_value = 0.01

ARGS = {
    "logits": logits,
    "unimix": unimix_value,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
distrax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Categorical — a Categorical output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Categorical output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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 9 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

As Deep Reinforcement Learning (Deep RL) research moves towards solving large-scale worlds, efficient environment simulations become crucial for rapid experimentation. However, most existing environments struggle to scale to high throughput, setting back meaningful progress. Interactions are typically computed on the CPU, limiting training speed and throughput, due to slower computation and communication overhead when distributing the task across multiple machines. Ultimately, Deep RL training is CPU-bound, and developing batched, fast, and scalable environments has become a frontier for progress. Among the most used Reinforcement Learning (RL) environments, MiniGrid is at the foundation of several studies on exploration, curriculum learning, representation learning, diversity, meta-learning, credit assignment, and language-conditioned RL, and still suffers from the limitations described above. In this work, we introduce NAVIX, a re-implementation of MiniGrid in JAX. NAVIX achieves over 200 000x speed improvements in batch mode, supporting up to 2048 agents in parallel on a single Nvidia A100 80 GB. This reduces experiment times from one week to 15 minutes, promoting faster design iterations and more scalable RL model development.

For agents

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

get_harvested_code_for_paper("2407.19396")
get_code_for_paper("2407.19396")
have("2407.19396")

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

curl https://syntology.ai/api/ran/2407.19396.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.19396.svg)](https://syntology.ai/paper/2407.19396)

Connect an agent — have() is free.