SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2004.14990 · NeurIPS · 2020

Reinforcement Learning with Augmented Data

U Berkeley, Pieter Abbeel, Lerrel Pinto, Michael Laskin, Kimin Lee, Adam Stooke, Aravind Srinivas

arXiv · PDF · Open in the Atlas

Code that ran

21 of 24 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (8 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 12 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/2004.14990.json · badge: /api/ran/2004.14990.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
MishaLaskin/rad named in the paper
Papers with Code archive
18 ran · 3 no recorded run
KarlXing/RL-Visual-Continuous-Control community
Papers with Code archive
3 ran · 0 no recorded run
FunctionStatusWhere it lives
Actor Ran
ran on a synthesized input · constructed an object; no result was computed
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("5aaa7c97de18af3a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

CURL Ran
ran on a synthesized input · constructed an object; no result was computed
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("a551a31aced2e8ed")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

ColorJitterLayer Ran
ran on a synthesized input · constructed an object; no result was computed
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("7eeb6c2c86ebd37d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

Critic Ran
ran on a synthesized input · constructed an object; no result was computed
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("faa8d44dfeaada9d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

IdentityEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("8b06f5d9a377410d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

PixelEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("75ea19a7c991ff1d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

RAD Ran
ran on a synthesized input · constructed an object; no result was computed
KarlXing/RL-Visual-Continuous-Control/src/agent/rad.py
code served (permissive licence) · get_code("4ef6619d0e3853e4")
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 an instance of its own class, RAD: the run constructed the object and did not call it, so no result was computed. 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
import numpy as np

torch.manual_seed(0)

# Define a minimal model with actor and critic components
class MinimalActor(nn.Module):
    def __init__(self, action_dim):
        super().__init__()
        self.action_dim = action_dim
        self.fc = nn.Linear(3, 16)
        self.mu_head = nn.Linear(16, action_dim)
        self.log_std_head = nn.Linear(16, action_dim)
    
    def forward(self, obs, compute_pi=True, compute_log_pi=True, detach=False):
        h = torch.relu(self.fc(obs))
        mu = torch.tanh(self.mu_head(h))
        log_std = self.log_std_head(h)
        log_std = torch.clamp(log_std, -20, 2)
        std = log_std.exp()
        
        if compute_pi:
            pi = mu + std * torch.randn_like(std)
        else:
            pi = mu
        
        if compute_log_pi:
            log_pi = -0.5 * ((pi - mu) ** 2 / std ** 2).sum(dim=-1, keepdim=True)
        else:
            log_pi = torch.zeros(mu.shape[0], 1)
        
        return mu, pi, log_pi, log_std

class MinimalCritic(nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = nn.Linear(3, 16)
        self.q1_head = nn.Linear(16 + 2, 1)
        self.q2_head = nn.Linear(16 + 2, 1)
    
    def forward(self, obs, action, detach=False):
        h = torch.relu(self.encoder(obs))
        if detach:
            h = h.detach()
        x = torch.cat([h, action], dim=-1)
        q1 = self.q1_head(x)
        q2 = self.q2_head(x)
        return q1, q2

class MinimalModel(nn.Module):
    def __init__(self, action_dim):
        super().__init__()
        self.actor = MinimalActor(action_dim)
        self.critic = MinimalCritic()
        self.critic_target = MinimalCritic()
        # Copy initial weights
        for p, target_p in zip(self.critic.parameters(), self.critic_target.parameters()):
            target_p.data.copy_(p.data)
    
    def soft_update_params(self, critic_tau, encoder_tau):
        for p, target_p in zip(self.critic.parameters(), self.critic_target.parameters()):
            target_p.data.copy_(target_p.data * (1 - critic_tau) + p.data * critic_tau)

# Create model
action_shape = (2,)
model = MinimalModel(action_dim=action_shape[0])

# Create a minimal logger
class MinimalLogger:
    def log(self, key, value, step):
        pass

# Create minimal args object
class Args:
    def __init__(self):
        self.actor_update_freq = 2
        self.critic_target_update_freq = 2
        self.critic_tau = 0.01
        self.critic_encoder_tau = 0.01
        self.agent_image_size = 84
        self.log_interval = 1
        self.discount = 0.99
        self.detach_encoder = True
        self.init_temperature = 0.1
        self.actor_lr = 1e-3
        self.actor_beta = 0.9
        self.critic_lr = 1e-3
        self.critic_beta = 0.9
        self.alpha_lr = 1e-3
        self.alpha_beta = 0.9

args = Args()
device = torch.device('cpu')

ARGS = {
    'model': model,
    'device': device,
    'action_shape': action_shape,
    'args': args
}
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 RAD — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a RAD 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
SAC Ran
ran on a synthesized input · constructed an object; no result was computed
KarlXing/RL-Visual-Continuous-Control/src/agent/rad.py
code served (permissive licence) · get_code("bf3db7327a139e28")
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 an instance of its own class, SAC: the run constructed the object and did not call it, so no result was computed. 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
import numpy as np

torch.manual_seed(0)

class DummyActor(nn.Module):
    def __init__(self, action_dim=2):
        super().__init__()
        self.action_dim = action_dim
        self.fc = nn.Linear(64, 128)
        self.mu_head = nn.Linear(128, action_dim)
        self.log_std_head = nn.Linear(128, action_dim)
    
    def forward(self, obs, compute_pi=True, compute_log_pi=True, detach=False):
        x = self.fc(obs)
        mu = self.mu_head(x)
        log_std = self.log_std_head(x)
        log_std = torch.clamp(log_std, min=-20, max=2)
        std = log_std.exp()
        
        pi = None
        log_pi = None
        
        if compute_pi:
            pi = mu + std * torch.randn_like(std)
        
        if compute_log_pi:
            log_pi = -0.5 * (log_std.sum(dim=-1, keepdim=True) + 
                            ((pi - mu) ** 2 / std ** 2).sum(dim=-1, keepdim=True))
        
        if pi is None:
            pi = mu
        if log_pi is None:
            log_pi = torch.zeros(obs.shape[0], 1, device=obs.device)
        
        return mu, pi, log_pi, log_std


class DummyCritic(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(66, 128)
        self.q1_head = nn.Linear(128, 1)
        self.q2_head = nn.Linear(128, 1)
    
    def forward(self, obs, action, detach=False):
        x = torch.cat([obs, action], dim=-1)
        x = self.fc(x)
        q1 = self.q1_head(x)
        q2 = self.q2_head(x)
        return q1, q2


class DummyModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.actor = DummyActor(action_dim=2)
        self.critic = DummyCritic()
        self.critic_target = DummyCritic()
        for p in self.critic_target.parameters():
            p.requires_grad = False
    
    def soft_update_params(self, critic_tau, encoder_tau):
        pass


class DummyArgs:
    def __init__(self):
        self.actor_update_freq = 2
        self.critic_target_update_freq = 2
        self.critic_tau = 0.005
        self.critic_encoder_tau = 0.001
        self.agent_image_size = 84
        self.log_interval = 1
        self.discount = 0.99
        self.detach_encoder = True
        self.init_temperature = 0.1
        self.actor_lr = 1e-3
        self.actor_beta = 0.9
        self.critic_lr = 1e-3
        self.critic_beta = 0.9
        self.alpha_lr = 1e-3
        self.alpha_beta = 0.9


ARGS = {
    'model': DummyModel(),
    'device': torch.device('cpu'),
    'action_shape': (2,),
    'args': DummyArgs(),
}
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 SAC — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a SAC 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
center_crop_image Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
KarlXing/RL-Visual-Continuous-Control/src/agent/rad.py
code served (permissive licence) · get_code("31e6960f8b942b0d")
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 3, dtype family float, shape (2, 2, 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 torch

torch.manual_seed(0)

# The function expects:
# - image: a tensor with shape (C, H, W) where it accesses image.shape[1:]
# - output_size: an integer for the crop size

# Create a small image tensor with shape (batch, height, width)
# Using distinct dimensions: batch=2, height=4, width=3
image = torch.randn(2, 4, 3)

# output_size should be smaller than the image dimensions
output_size = 2

ARGS = {
    "image": image,
    "output_size": output_size,
}
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 3, dtype family float, shape (2, 2, 2) — shape (2, 2, 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
grayscale Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("1bb28d88e7ae1e71")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

hsv2rgb Ran
ran; our contract draft was wrong, not the code
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("718e1902f65f1be2")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

make_encoder Ran
ran; our contract draft was wrong, not the code
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("3ff1a7cb98f67025")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

no_aug Ran
ran, honoured the contract we drafted
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("b1cb15a87e0863be")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_convolution Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("3b7275db7d30b2b9")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_crop Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("cc205ddaca3bfbbc")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_cutout Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("417e34ed0818a71b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_cutout_color Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("d4841127aa83b755")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_flip Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("0c0e4ec30cabf972")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_grayscale Ran
ran; our contract draft was wrong, not the code
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("ef56bb771298e9ef")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

random_translate Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("77241bfa75e23abd")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

rgb2hsv Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("1c24bb2fb72c4f9d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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.

RadSacAgent Not yet run
unverified (harvested, no recorded run)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("62d57e42506581cb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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. No invocation recorded.

random_color_jitter Not yet run
unverified (harvested, no recorded run)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("5a5ab3e704f9a4cd")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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. No invocation recorded.

random_rotation Not yet run
unverified (harvested, no recorded run)
MishaLaskin/rad/curl_sac.py
pointer only (licence: NONE) · get_code("c0364ede43f86490")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MishaLaskin/rad 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. No invocation recorded.

Also present in other papers' code

Of 24 samples lifted from this paper's linked repositories, 24 carry a function identity and 0 do not. Of the 24 distinct functions with an identity, 10 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 4 the same month, 9 later) and 14 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
make_encoder
brick of this paper's assembly: helper
5 other papers
1 the same month, 4 later
2004.04136 (2020-04)
CURL: Contrastive Unsupervised Representations for Reinforcement Learning
2010.09163 (2020-10)
D2RL: Deep Dense Architectures in Reinforcement Learning
2205.00943 (2022-05)
CCLF: A Contrastive-Curiosity-Driven Learning Framework for Sample-Efficient Reinforcement Learning
and 2 more not named here
CURL
brick of this paper's assembly: model
3 other papers
1 the same month, 2 later · 2 only as an assembly brick
2004.04136 (2020-04)
CURL: Contrastive Unsupervised Representations for Reinforcement Learning
2007.04309 (2020-07)
Self-Supervised Policy Adaptation during Deployment
2205.00943 (2022-05)
CCLF: A Contrastive-Curiosity-Driven Learning Framework for Sample-Efficient Reinforcement Learning
center_crop_image
brick of this paper's assembly: data_transform, helper
3 other papers
1 the same month, 2 later · 2 only as an assembly brick
2004.04136 (2020-04)
CURL: Contrastive Unsupervised Representations for Reinforcement Learning
2110.04935 (2021-10)
Learning Temporally-Consistent Representations for Data-Efficient Reinforcement Learning
2205.00943 (2022-05)
CCLF: A Contrastive-Curiosity-Driven Learning Framework for Sample-Efficient Reinforcement Learning
hsv2rgb
brick of this paper's assembly: helper
2 other papers
2 later
2006.12862 (2020-06)
Automatic Data Augmentation for Generalization in Deep Reinforcement Learning
2210.07241 (2022-10)
Visual Reinforcement Learning with Self-Supervised 3D Representations
rgb2hsv
brick of this paper's assembly: helper
2 other papers
2 later
2006.12862 (2020-06)
Automatic Data Augmentation for Generalization in Deep Reinforcement Learning
2210.07241 (2022-10)
Visual Reinforcement Learning with Self-Supervised 3D Representations
Actor
brick of this paper's assembly: model
1 other paper
1 later
2205.00943 (2022-05)
CCLF: A Contrastive-Curiosity-Driven Learning Framework for Sample-Efficient Reinforcement Learning
Critic
brick of this paper's assembly: model
1 other paper
1 later
2205.00943 (2022-05)
CCLF: A Contrastive-Curiosity-Driven Learning Framework for Sample-Efficient Reinforcement Learning
IdentityEncoder
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2205.00943 (2022-05)
CCLF: A Contrastive-Curiosity-Driven Learning Framework for Sample-Efficient Reinforcement Learning
SAC
brick of this paper's assembly: helper
1 other paper
1 the same month · 1 only as an assembly brick
2004.04136 (2020-04)
CURL: Contrastive Unsupervised Representations for Reinforcement Learning
random_translate
brick of this paper's assembly: helper
1 other paper
1 later
2110.04935 (2021-10)
Learning Temporally-Consistent Representations for Data-Efficient 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

Learning from visual observations is a fundamental yet challenging problem in Reinforcement Learning (RL). Although algorithmic advances combined with convolutional neural networks have proved to be a recipe for success, current methods are still lacking on two fronts: (a) data-efficiency of learning and (b) generalization to new environments. To this end, we present Reinforcement Learning with Augmented Data (RAD), a simple plug-and-play module that can enhance most RL algorithms. We perform the first extensive study of general data augmentations for RL on both pixel-based and state-based inputs, and introduce two new data augmentations -random translate and random amplitude scale. We show that augmentations such as random translate, crop, color jitter, patch cutout, random convolutions, and amplitude scale can enable simple RL algorithms to outperform complex state-of-the-art methods across common benchmarks. RAD sets a new state-of-the-art in terms of data-efficiency and final performance on the DeepMind Control Suite benchmark for pixel-based control as well as Ope-nAI Gym benchmark for state-based control. We further demonstrate that RAD significantly improves test-time generalization over existing methods on several OpenAI ProcGen benchmarks. Our RAD module and training code are available at https://www.github.com/MishaLaskin/rad.

For agents

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

get_harvested_code_for_paper("2004.14990")
get_code_for_paper("2004.14990")
have("2004.14990")

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

curl https://syntology.ai/api/ran/2004.14990.json

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

[![Syntology run record](https://syntology.ai/api/ran/2004.14990.svg)](https://syntology.ai/paper/2004.14990)

Connect an agent — have() is free.