SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2304.04782 · 2023

Reinforcement Learning from Passive Data via Latent Intentions

arXiv · PDF · Open in the Atlas

Code that ran

11 of 15 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; 3 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 0 failures of our instrument; 4 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/2304.04782.json · badge: /api/ran/2304.04782.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
dibyaghosh/icvf_release canonical (classifier label)
classifier basis: census
11 ran · 4 no recorded run
FunctionStatusWhere it lives
capitalize_game_name Ran
ran on a synthesized input
dibyaghosh/icvf_release/icvf_envs/atari/atari_envs.py
code served (permissive licence) · get_code("25a1b02d068ed008")
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 str. 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

# Analyze the function:
# - Takes one parameter: 'game' (a string)
# - Replaces '-' with '_'
# - Splits by '_'
# - Capitalizes each part
# - Joins them back together

# The function expects a string representing a game name with hyphens or underscores

ARGS = {
    'game': 'pac-man'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
distance_from_beginning Ran
ran on a synthesized input
dibyaghosh/icvf_release/icvf_envs/atari/offline_env.py
code served (permissive licence) · get_code("239fb49d34d3d55a")
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 int, 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 numpy as np

# Analyze the function:
# - terminals parameter is indexed with [i-1] and len() is called on it
# - It should be a sequence (list, tuple, or array) of boolean-like values
# - The function checks if terminals[i-1] is truthy
# - Returns np.array of distances

# Create a simple terminals sequence
# terminals[i-1] = True means we reset the distance counter to 1
# terminals[i-1] = False means we increment the distance counter
terminals = [False, False, True, False]

ARGS = {
    "terminals": terminals
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — 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 int, 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
expectile_loss Ran
ran on a synthesized input
dibyaghosh/icvf_release/src/icvf_learner.py
code served (permissive licence) · get_code("97edc824282c1fed")
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 2, dtype family float, shape (2, 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.numpy as jnp
import numpy as np

# Create small tensors for the expectile_loss function
# The function takes:
# - adv: advantage values (can be positive or negative)
# - diff: differences (typically TD errors or similar)
# - expectile: scalar parameter (default 0.8)

batch_size = 2
dim = 4

# Create advantage tensor with mixed positive and negative values
adv = jnp.array([
    [0.5, -0.3, 0.2, -0.1],
    [-0.4, 0.6, -0.2, 0.3]
], dtype=jnp.float32)

# Create difference tensor
diff = jnp.array([
    [0.1, 0.2, 0.15, 0.25],
    [0.3, 0.05, 0.1, 0.2]
], dtype=jnp.float32)

# Expectile parameter
expectile = 0.8

ARGS = {
    "adv": adv,
    "diff": diff,
    "expectile": expectile
}
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 2, dtype family float, shape (2, 4) — shape (2, 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
get_canvas_image Ran
ran, honoured the contract we drafted
dibyaghosh/icvf_release/icvf_envs/antmaze/d4rl_ant.py
pointer only (licence: MIT); the stored copy get_code serves, from seohongpark/HILP, is pointer only (licence: NONE) · get_code("2d17f9689108389d")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from seohongpark/HILP, and GitHub detected no licence in that repository when it was checked. 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.

get_dir_path Ran
ran on a synthesized input
dibyaghosh/icvf_release/icvf_envs/atari/offline_env.py
code served (permissive licence) · get_code("0041222c7cf86bf6")
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 str. 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 os
from os.path import expanduser

BASE_DIR = os.environ.get('ATARI_DATASET_DIR', os.path.join(expanduser('~'), 'nfs', 'atari'))

ARGS = {
    'env': 'Pong',
    'index': 0,
    'epoch': 5,
    'base_dir': BASE_DIR
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
get_inner_env Ran
ran on a synthesized input
dibyaghosh/icvf_release/icvf_envs/antmaze/d4rl_ant.py
code served (permissive licence) · get_code("70db1a94e7d40b3c")
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 MazeEnv. 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 torch.nn as nn

class MazeEnv:
    def __init__(self, has_maze_size_scaling=False):
        if has_maze_size_scaling:
            self._maze_size_scaling = 1.0

class WrappedEnv:
    def __init__(self, inner_env):
        self.wrapped_env = inner_env

class EnvWrapper:
    def __init__(self, inner_env):
        self.env = inner_env

# Test case 1: Direct environment with _maze_size_scaling attribute
env_with_scaling = MazeEnv(has_maze_size_scaling=True)

# Test case 2: Environment wrapped with 'env' attribute
inner_env = MazeEnv(has_maze_size_scaling=True)
env_with_env_attr = EnvWrapper(inner_env)

# Test case 3: Environment wrapped with 'wrapped_env' attribute
inner_env2 = MazeEnv(has_maze_size_scaling=True)
env_with_wrapped_env_attr = WrappedEnv(inner_env2)

# Test case 4: Plain environment without special attributes
plain_env = MazeEnv(has_maze_size_scaling=False)

ARGS = {
    'env': env_with_scaling
}
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 MazeEnv — a MazeEnv output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MazeEnv 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
get_normalization Ran
ran, honoured the contract we drafted
dibyaghosh/icvf_release/icvf_envs/antmaze/d4rl_utils.py
pointer only (licence: MIT); the stored copy get_code serves, from seohongpark/HILP, is pointer only (licence: NONE) · get_code("e60bfdd0574ddc5f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from seohongpark/HILP, and GitHub detected no licence in that repository when it was checked. 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.

icvf_loss Ran
ran on a synthesized input
dibyaghosh/icvf_release/src/icvf_learner.py
code served (permissive licence) · get_code("45f3cc2e7c6bba16")
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

# Build mock value functions that take 3 arguments and return 2 values
class MockValueFn:
    def __call__(self, observations, goals, desired_goals):
        batch_size = observations.shape[0]
        v1 = jnp.ones((batch_size,)) * 0.5
        v2 = jnp.ones((batch_size,)) * 0.3
        return v1, v2

value_fn = MockValueFn()
target_value_fn = MockValueFn()

# Build batch with required keys
batch_size = 2
obs_dim = 4
goal_dim = 3
reward_dim = 1

batch = {
    'observations': jnp.ones((batch_size, obs_dim)),
    'next_observations': jnp.ones((batch_size, obs_dim)) * 0.9,
    'goals': jnp.ones((batch_size, goal_dim)),
    'desired_goals': jnp.ones((batch_size, goal_dim)) * 0.8,
    'rewards': jnp.ones((batch_size,)) * 0.1,
    'desired_rewards': jnp.ones((batch_size,)) * 0.2,
    'masks': jnp.ones((batch_size,)),
    'desired_masks': jnp.ones((batch_size,)),
}

# Build config dict with required keys
config = {
    'no_intent': False,
    'min_q': True,
    'expectile': 0.8,
    'discount': 0.99,
}

ARGS = {
    'value_fn': value_fn,
    'target_value_fn': target_value_fn,
    'batch': batch,
    'config': config,
}
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
inspect_dir_path Ran
ran on a synthesized input
dibyaghosh/icvf_release/icvf_envs/atari/offline_env.py
code served (permissive licence) · get_code("9693e418d98ddb87")
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 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 os
import tempfile
import shutil

# Create a temporary directory structure to simulate the ATARI_DATASET_DIR
temp_base_dir = tempfile.mkdtemp()

# Create a valid directory structure with the required .gz files
env_name = "Pong"
index = 0
epoch = 0

env_dir = os.path.join(temp_base_dir, env_name, str(index), str(epoch))
os.makedirs(env_dir, exist_ok=True)

# Create the required .gz files
for name in ['observation', 'action', 'reward', 'terminal']:
    with open(os.path.join(env_dir, name + '.gz'), 'w') as f:
        f.write('')

# Define the arguments dictionary
ARGS = {
    'env': env_name,
    'index': index,
    'epoch': epoch,
    'base_dir': temp_base_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — 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
update_dict Ran
ran on a synthesized input
dibyaghosh/icvf_release/experiments/antmaze/train_icvf.py
code served (permissive licence) · get_code("ffd7f2a1980245a1")
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 dict. 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 numpy as np

torch.manual_seed(0)

# The function update_dict simply calls d.update(additional) and returns d
# Both parameters should be dictionaries

d = {
    'key1': 'value1',
    'key2': 42,
    'key3': [1, 2, 3]
}

additional = {
    'key4': 'value4',
    'key5': torch.randn(2, 4),
    'key2': 100  # This will override the existing key2
}

ARGS = {
    'd': d,
    'additional': additional
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
valid_goal_sampler Ran
ran, honoured the contract we drafted
dibyaghosh/icvf_release/icvf_envs/antmaze/d4rl_ant.py
pointer only (licence: MIT); the stored copy get_code serves, from seohongpark/HILP, is pointer only (licence: NONE) · get_code("a5514c312d961e08")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from seohongpark/HILP, and GitHub detected no licence in that repository when it was checked. 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.

get_debug_statistics Not yet run
unverified (harvested, no recorded run)
dibyaghosh/icvf_release/experiments/xmagical/train_icvf.py
code served (permissive licence) · get_code("fba01e14dbbb753b")
How it ran

no invocation recorded.

get_distances Not yet run
unverified (harvested, no recorded run)
dibyaghosh/icvf_release/experiments/xmagical/train_icvf.py
code served (permissive licence) · get_code("1e41eedaedbe22c3")
How it ran

no invocation recorded.

get_policy Not yet run
unverified (harvested, no recorded run)
dibyaghosh/icvf_release/experiments/antmaze/train_icvf.py
code served (permissive licence) · get_code("7efb230a126ddd62")
How it ran

no invocation recorded.

get_values Not yet run
unverified (harvested, no recorded run)
dibyaghosh/icvf_release/experiments/antmaze/train_icvf.py
code served (permissive licence) · get_code("a2628498c8976d46")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 3 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 3 later) and 12 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.

FunctionAlso present inEarliest-datedOthers named
get_canvas_image 4 other papers
4 later
2307.11949 (2023-07)
HIQL: Offline Goal-Conditioned RL with Latent States as Actions
2402.15567 (2024-02)
Foundation Policies with Hilbert Representations
2402.17135 (2024-02)
Unsupervised Zero-Shot Reinforcement Learning via Functional Reward Encodings
and 1 more not named here
get_normalization 3 other papers
3 later
2307.11949 (2023-07)
HIQL: Offline Goal-Conditioned RL with Latent States as Actions
2402.15567 (2024-02)
Foundation Policies with Hilbert Representations
2510.15382 (2025-10)
Towards Robust Zero-Shot Reinforcement Learning
valid_goal_sampler 3 other papers
3 later
2307.11949 (2023-07)
HIQL: Offline Goal-Conditioned RL with Latent States as Actions
2402.15567 (2024-02)
Foundation Policies with Hilbert Representations
2510.15382 (2025-10)
Towards Robust Zero-Shot Reinforcement Learning

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

Passive observational data, such as human videos, is abundant and rich in information, yet remains largely untapped by current RL methods. Perhaps surprisingly, we show that passive data, despite not having reward or action labels, can still be used to learn features that accelerate downstream RL. Our approach learns from passive data by modeling intentions: measuring how the likelihood of future outcomes change when the agent acts to achieve a particular task. We propose a temporal difference learning objective to learn about intentions, resulting in an algorithm similar to conventional RL, but which learns entirely from passive data. When optimizing this objective, our agent simultaneously learns representations of states, of policies, and of possible outcomes in an environment, all from raw observational data. Both theoretically and empirically, this scheme learns features amenable for value prediction for downstream tasks, and our experiments demonstrate the ability to learn from many forms of passive data, including cross-embodiment video data and YouTube videos.

For agents

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

get_harvested_code_for_paper("2304.04782")
get_code_for_paper("2304.04782")
have("2304.04782")

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

curl https://syntology.ai/api/ran/2304.04782.json

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

[![Syntology run record](https://syntology.ai/api/ran/2304.04782.svg)](https://syntology.ai/paper/2304.04782)

Connect an agent — have() is free.