SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2003.11881 · 2020

An empirical investigation of the challenges of real-world reinforcement learning

arXiv · PDF · Open in the Atlas

Code that ran

11 of 13 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; 0 honoured the contract we drafted, 0 violated it and 11 have no contract verdict; 0 failures of our instrument; 2 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/2003.11881.json · badge: /api/ran/2003.11881.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
google-research/realworldrl_suite canonical (classifier label)
classifier basis: census
11 ran · 2 no recorded run
FunctionStatusWhere it lives
action_roc_constraint Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/realworld_env.py
code served (permissive licence) · get_code("9eca1ba5f0fccf44")
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 numpy as np

class MockEnv:
    def __init__(self):
        self._last_action = np.array([0.1, 0.2, 0.3])
        self.limits = {
            'action_roc_constraint': np.array([0.5, 0.5, 0.5])
        }

env = MockEnv()

safety_vars = {
    'actions': np.array([0.15, 0.25, 0.35])
}

ARGS = {
    'env': env,
    'safety_vars': safety_vars
}
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 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
delay_buffer_spec Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/realworld_env.py
code served (permissive licence) · get_code("7edce8371870a10d")
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 torch
import numpy as np

torch.manual_seed(0)

# The function expects:
# - delay_spec: a dict-like object that supports .get(delay_name, default)
# - delay_name: a string key to look up in the dict

delay_spec = {
    "delay_1": 5,
    "delay_2": 10,
    "delay_3": 3
}

delay_name = "delay_1"

ARGS = {
    "delay_spec": delay_spec,
    "delay_name": delay_name
}
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 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
delayed_buffer_item Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/realworld_env.py
code served (permissive licence) · get_code("fe74fdf94d5ba805")
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. 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 copy

torch.manual_seed(0)

# Analyze the function:
# - buffer_item: can be None or a list (used with append, pop(0), and copy.copy)
# - buffer_item_len: an integer (used to create a list of length buffer_item_len)
# - item: something that can be copy.copy'd (could be a tensor, dict, or simple object)

# Create a simple item that can be copied
item = torch.randn(2, 4)

# Set buffer_item to None initially (function will initialize it)
buffer_item = None

# Set buffer_item_len to control the delay buffer size
buffer_item_len = 3

ARGS = {
    "buffer_item": buffer_item,
    "buffer_item_len": buffer_item_len,
    "item": item,
}
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 tuple — a tuple output has no array shape of its own; none is 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_normalized_regret Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/utils/evaluators.py
code served (permissive licence) · get_code("0141ec69754e453e")
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 collections
import numpy as np

class MockEvaluator:
    """Mock evaluator with task_name and get_normalized_regret method."""
    def __init__(self, task_name, regret_values):
        self.task_name = task_name
        self.regret_values = regret_values
        self._call_count = 0
    
    def get_normalized_regret(self):
        """Return regret value, cycling through the list on repeated calls."""
        val = self.regret_values[self._call_count % len(self.regret_values)]
        self._call_count += 1
        return val

# Create a list of evaluators with different task names and regret values
evaluator_list = [
    MockEvaluator("task_1", [0.1, 0.15, 0.12]),
    MockEvaluator("task_1", [0.11, 0.14, 0.13]),
    MockEvaluator("task_2", [0.2, 0.25, 0.22]),
    MockEvaluator("task_2", [0.21, 0.24, 0.23]),
]

ARGS = {
    "evaluator_list": evaluator_list,
}
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 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
get_regret_plot Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/utils/evaluators.py
code served (permissive licence) · get_code("36ba0420471457fb")
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 Figure. 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 collections
import numpy as np

class MockEvaluator:
    def __init__(self, task_name, regret_value):
        self.task_name = task_name
        self.regret_value = regret_value
    
    def get_normalized_regret(self):
        return self.regret_value

evaluator_list = [
    MockEvaluator("task_a", 0.15),
    MockEvaluator("task_a", 0.18),
    MockEvaluator("task_b", 0.25),
    MockEvaluator("task_b", 0.22),
    MockEvaluator("task_c", 0.10),
    MockEvaluator("task_c", 0.12),
]

ARGS = {
    "evaluator_list": evaluator_list
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Figure — a Figure output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Figure 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_return_plot Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/utils/evaluators.py
code served (permissive licence) · get_code("c51722c106544461")
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 Figure. 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 collections
import numpy as np
import matplotlib.pyplot as plt

class MockEvaluator:
    def __init__(self, task_name, num_episodes):
        self.task_name = task_name
        self.stats = {
            'return_stats': {
                'episode_totals': np.random.randn(num_episodes)
            }
        }

evaluator_1_task_a = MockEvaluator('task_a', 20)
evaluator_2_task_a = MockEvaluator('task_a', 20)
evaluator_1_task_b = MockEvaluator('task_b', 20)
evaluator_2_task_b = MockEvaluator('task_b', 20)

evaluator_list = [
    evaluator_1_task_a,
    evaluator_2_task_a,
    evaluator_1_task_b,
    evaluator_2_task_b,
]

ARGS = {
    'evaluator_list': evaluator_list,
    'stride': 500,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Figure — a Figure output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Figure 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
joint_angle_constraint Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/humanoid.py
code served (permissive licence) · get_code("5d779f1a93663ec6")
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 numpy as np

class MockEnv:
    def __init__(self):
        self.limits = {
            'joint_pos_constraint': 0.5
        }

# Create small joint position array (batch_size=2, joint_dim=4)
joint_pos = np.array([
    [0.1, 0.2, 0.15, 0.05],
    [0.3, 0.1, 0.25, 0.2]
])

safety_vars = {
    'joint_pos': joint_pos
}

env = MockEnv()

ARGS = {
    'env': env,
    'safety_vars': safety_vars
}
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 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
joint_angle_constraint Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/manipulator.py
code served (permissive licence) · get_code("7d13da0649838360")
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 numpy as np

class MockEnv:
    def __init__(self):
        self.limits = {
            'joint_angle_constraint': (
                np.array([0.1, 0.2, 0.15, 0.25]),
                np.array([0.9, 0.8, 0.85, 0.75])
            )
        }

env = MockEnv()

safety_vars = {
    'joint_angle': np.array([0.5, 0.5, 0.5, 0.5])
}

ARGS = {
    'env': env,
    'safety_vars': safety_vars
}
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 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
joint_angle_constraint Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/walker.py
code served (permissive licence) · get_code("81f8e9e3075bcd11")
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 numpy as np

class MockEnv:
    def __init__(self):
        self.limits = {
            'joint_pos_constraint': [np.array([0.0, 0.0, 0.0, 0.0]), 
                                      np.array([1.0, 1.0, 1.0, 1.0])]
        }

env = MockEnv()

safety_vars = {
    'joint_pos': np.array([0.5, 0.5, 0.5, 0.5])
}

ARGS = {
    'env': env,
    'safety_vars': safety_vars
}
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 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
joint_velocity_constraint Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/humanoid.py
code served (permissive licence) · get_code("aa89ad35cd55ff30")
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 numpy as np

# Create a minimal environment mock that has the 'limits' attribute
class MockEnv:
    def __init__(self):
        self.limits = {
            'joint_velocity_constraint': 2.5  # threshold value
        }

# Create safety_vars dict with joint velocities
safety_vars = {
    'joint_vels': np.array([0.5, 1.2, 0.8, 1.5])  # 4 joint velocities, all below threshold
}

# Create the environment instance
env = MockEnv()

# Define ARGS mapping all parameters
ARGS = {
    'env': env,
    'safety_vars': safety_vars
}
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 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
slider_pos_constraint Ran
ran on a synthesized input
google-research/realworldrl_suite/realworldrl_suite/environments/cartpole.py
code served (permissive licence) · get_code("3d09d7b3f3d1ab9b")
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 numpy as np

class MockEnv:
    def __init__(self):
        self.limits = {
            'slider_pos_constraint': [0.1, 0.9]
        }

env = MockEnv()

safety_vars = {
    'slider_pos': 0.5
}

ARGS = {
    'env': env,
    'safety_vars': safety_vars
}
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 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
balance_velocity_constraint Not yet run
unverified (harvested, no recorded run)
google-research/realworldrl_suite/realworldrl_suite/environments/cartpole.py
code served (permissive licence) · get_code("d8f32de0f53c3b76")
How it ran

no invocation recorded.

load Not yet run
unverified (harvested, no recorded run)
google-research/realworldrl_suite/realworldrl_suite/environments/cartpole.py
code served (permissive licence) · get_code("8c02170b9137ac33")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 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 13 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

Reinforcement learning (RL) has proven its worth in a series of artificial domains, and is beginning to show some successes in real-world scenarios. However, much of the research advances in RL are hard to leverage in real-world systems due to a series of assumptions that are rarely satisfied in practice. In this work, we identify and formalize a series of independent challenges that embody the difficulties that must be addressed for RL to be commonly deployed in real-world systems. For each challenge, we define it formally in the context of a Markov Decision Process, analyze the effects of the challenge on state-of-the-art learning algorithms, and present some existing attempts at tackling it. We believe that an approach that addresses our set of proposed challenges would be readily deployable in a large number of real world problems. Our proposed challenges are implemented in a suite of continuous control environments called the realworldrl-suite which we propose an as an open-source benchmark.

For agents

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

get_harvested_code_for_paper("2003.11881")
get_code_for_paper("2003.11881")
have("2003.11881")

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

curl https://syntology.ai/api/ran/2003.11881.json

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

[![Syntology run record](https://syntology.ai/api/ran/2003.11881.svg)](https://syntology.ai/paper/2003.11881)

Connect an agent — have() is free.