SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1910.01465 · 2019

Reducing Overestimation Bias in Multi-Agent Domains Using Double Centralized Critics

arXiv · PDF · Open in the Atlas

Code that ran

10 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 7 have no contract verdict; 3 failures of our instrument; 3 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/1910.01465.json · badge: /api/ran/1910.01465.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
fdcl-gwu/gym-rotor community (classifier label)
classifier basis not recorded on the link
10 ran · 3 no recorded run
FunctionStatusWhere it lives
approx_spectral_norm Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/algos/spectral_norm_regularization.py
code served (permissive licence) · get_code("75aa7fa60710d8bc")
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 Tensor, 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)

# Create a small weight matrix W
# The function uses W.shape[1] for n, and performs matrix operations W.T @ W @ x and W @ v1
# So W should be a 2D tensor. Let's make it 4x3 (distinct dimensions)
W = torch.randn(4, 3)

# Device - CPU only
device = torch.device('cpu')

ARGS = {
    'W': W,
    'device': device,
}
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 Tensor, 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
batched_projection Ran
ran; our contract draft was wrong, not the code
fdcl-gwu/gym-rotor/algos/emlp_torch/interface.py
code served (permissive licence) · get_code("9293b19c0e12cbb2")
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 Tensor, 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 torch
import torch.nn as nn

torch.manual_seed(0)

# Create small batched tensors for projection
# Shape: (batch_size, feature_dim)
# bu: vector to project onto
# bv: vector being projected
bu = torch.randn(2, 4)  # batch=2, dim=4
bv = torch.randn(2, 4)  # batch=2, dim=4

ARGS = {
    "bu": bu,
    "bv": bv,
}
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 Tensor, 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
matrix_power_simple Ran
ran; our contract draft was wrong, not the code
fdcl-gwu/gym-rotor/algos/emlp_torch/groups.py
code served (permissive licence) · get_code("9071e08c34a5feb3")
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 Tensor, ndim 2, dtype family float, shape (4, 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 torch

torch.manual_seed(0)

# Construct a small invertible matrix (4x4) for testing
# Use a simple invertible matrix by adding identity scaled
M = torch.eye(4) + 0.1 * torch.randn(4, 4)

# Create a scalar tensor for the power n
n = torch.tensor(3)

ARGS = {
    "M": M,
    "n": n
}
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 Tensor, ndim 2, dtype family float, shape (4, 4) — shape (4, 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
obs1_decomposition Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/gym_rotor/wrappers/wrapper_utils.py
code served (permissive licence) · get_code("694023469e7d948b")
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 2, 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 obs1_decomposition expects a single parameter 'obs'
# that supports slicing with obs[start:end]
# It extracts 5 components from indices 0-3, 3-6, 6-9, 9-12, 12-15
# Total length needed: 15 elements

# Create a tensor with shape (15,) to satisfy the slicing operations
obs = torch.randn(15)

ARGS = {
    "obs": obs
}
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 2, 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
obs2_decomposition Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/gym_rotor/wrappers/wrapper_utils.py
code served (permissive licence) · get_code("e8bc3a8105ef63b7")
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 obs2_decomposition expects a single parameter 'obs' 
# which is indexed with [0], [1], and [2], so it must be a sequence/list
# with at least 3 elements containing norms (scalars or tensors)

eb1_norm = torch.tensor(0.5)
eIb1_norm = torch.tensor(1.2)
eW3_norm = torch.tensor(0.8)

obs = [eb1_norm, eIb1_norm, eW3_norm]

ARGS = {
    "obs": obs
}
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
obs_decomposition Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/gym_rotor/wrappers/wrapper_utils.py
code served (permissive licence) · get_code("769fbcf64434b175")
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

torch.manual_seed(0)

# The function expects obs to be indexable (like a tensor or list)
# and slices it at various positions to extract components
# obs[0:3], obs[3:6], obs[6:9], obs[9:18], obs[18], obs[19], obs[20:23]
# Total elements needed: up to index 23 (so at least 23 elements)

# Create a tensor with 23 elements to satisfy all slicing operations
obs = torch.randn(23)

ARGS = {
    "obs": obs
}
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
rel_err Ran
ran; our contract draft was wrong, not the code
fdcl-gwu/gym-rotor/algos/emlp_torch/groups.py
code served (permissive licence) · get_code("71640ef09311bdf6")
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 Tensor, 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)

# Create two small tensors with the same shape
A = torch.randn(2, 4)
B = torch.randn(2, 4)

ARGS = {
    "A": A,
    "B": B,
}
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 Tensor, 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
state_de_normalization Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/gym_rotor/envs/quad_utils.py
code served (permissive licence) · get_code("3650d0ed097001ac")
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 numpy as np

# Create valid arguments for state_de_normalization function
state = np.concatenate([
    np.random.randn(3) * 0.5,      # x: normalized position (3,)
    np.random.randn(3) * 0.5,      # v: normalized velocity (3,)
    np.eye(3).flatten('F') * 0.9,  # R_vec: flattened rotation matrix (9,)
    np.zeros(3),                    # padding to reach index 15 (3,)
    np.random.randn(3) * 0.5       # W: normalized angular velocity (3,)
])

x_lim = 10.0
v_lim = 5.0
W_lim = 2.0

ARGS = {
    'state': state,
    'x_lim': x_lim,
    'v_lim': v_lim,
    'W_lim': W_lim
}
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 — 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
state_decomposition Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/gym_rotor/envs/quad_utils.py
code served (permissive licence) · get_code("58499f313fc35c5b")
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 numpy as np

# Create a valid state vector for state_decomposition
# state should be: [x(3), v(3), R_vec(9), W(3)] = 18 elements total
# x: position (3 elements)
x = np.array([1.0, 2.0, 3.0])

# v: velocity (3 elements)
v = np.array([0.1, 0.2, 0.3])

# R_vec: rotation matrix flattened in Fortran order (9 elements)
# Create a valid SO(3) rotation matrix (identity for simplicity)
R_valid = np.eye(3)
R_vec = R_valid.flatten(order='F')

# W: angular velocity (3 elements)
W = np.array([0.01, 0.02, 0.03])

# Construct the full state vector
state = np.concatenate([x, v, R_vec, W])

ARGS = {
    'state': state
}
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 — 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
state_normalization Ran
ran on a synthesized input
fdcl-gwu/gym-rotor/gym_rotor/envs/quad_utils.py
code served (permissive licence) · get_code("b60f2bf99e894bce")
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 numpy as np

# Analyze the function signature and body:
# state_normalization(state, x_lim, v_lim, W_lim)
# - state: numpy array with shape (18,) based on indexing [0:3], [3:6], [6:15], [15:18]
# - x_lim: scalar or array for normalization of state[0:3]
# - v_lim: scalar or array for normalization of state[3:6]
# - W_lim: scalar or array for normalization of state[15:18]

# Create valid SO(3) rotation matrix for state[6:15]
def create_rotation_matrix():
    # Create a valid rotation matrix via QR decomposition
    A = np.random.randn(3, 3)
    Q, R = np.linalg.qr(A)
    # Ensure determinant is +1 (not -1)
    if np.linalg.det(Q) < 0:
        Q[:, 0] *= -1
    return Q

# Build state vector: [x(3), v(3), R_vec(9), W(3)]
rotation_matrix = create_rotation_matrix()
R_vec = rotation_matrix.reshape(9, 1, order='F').flatten()

state = np.concatenate([
    np.array([0.5, 1.0, -0.3]),  # x (3,)
    np.array([0.1, 0.2, 0.05]),   # v (3,)
    R_vec,                          # R_vec (9,)
    np.array([0.01, 0.02, 0.03])   # W (3,)
])

assert state.shape == (18,), f"Expected state shape (18,), got {state.shape}"

ARGS = {
    'state': state,
    'x_lim': np.array([1.0, 1.0, 1.0]),
    'v_lim': np.array([0.5, 0.5, 0.5]),
    'W_lim': np.array([0.1, 0.1, 0.1])
}
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 — 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
batched_gram_schmidt Not yet run
unverified (harvested, no recorded run)
fdcl-gwu/gym-rotor/algos/emlp_torch/interface.py
code served (permissive licence) · get_code("98b3fcdbaa43214e")
How it ran

no invocation recorded.

noise2sample Not yet run
unverified (harvested, no recorded run)
fdcl-gwu/gym-rotor/algos/emlp_torch/groups.py
code served (permissive licence) · get_code("770eda0d8a367d7e")
How it ran

no invocation recorded.

policy_regularization Not yet run
unverified (harvested, no recorded run)
fdcl-gwu/gym-rotor/algos/policy_regularization.py
code served (permissive licence) · get_code("8db8d2dd300b6762")
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, 5 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 5 later) and 8 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
batched_gram_schmidt 1 other paper
1 later
2311.07143 (2023-11)
Learning Symmetrization for Equivariance with Orbit Distance Minimization
batched_projection 1 other paper
1 later
2311.07143 (2023-11)
Learning Symmetrization for Equivariance with Orbit Distance Minimization
matrix_power_simple 1 other paper
1 later
2311.07143 (2023-11)
Learning Symmetrization for Equivariance with Orbit Distance Minimization
noise2sample 1 other paper
1 later
2311.07143 (2023-11)
Learning Symmetrization for Equivariance with Orbit Distance Minimization
rel_err 1 other paper
1 later
2311.07143 (2023-11)
Learning Symmetrization for Equivariance with Orbit Distance Minimization

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

Many real world tasks require multiple agents to work together. Multi-agent reinforcement learning (RL) methods have been proposed in recent years to solve these tasks, but current methods often fail to efficiently learn policies. We thus investigate the presence of a common weakness in single-agent RL, namely value function overestimation bias, in the multi-agent setting. Based on our findings, we propose an approach that reduces this bias by using double centralized critics. We evaluate it on six mixed cooperative-competitive tasks, showing a significant advantage over current methods. Finally, we investigate the application of multi-agent methods to high-dimensional robotic tasks and show that our approach can be used to learn decentralized policies in this domain.

For agents

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

get_harvested_code_for_paper("1910.01465")
get_code_for_paper("1910.01465")
have("1910.01465")

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

curl https://syntology.ai/api/ran/1910.01465.json

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

[![Syntology run record](https://syntology.ai/api/ran/1910.01465.svg)](https://syntology.ai/paper/1910.01465)

Connect an agent — have() is free.