SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.01497 · 2023

Teach a Robot to FISH: Versatile Imitation from One Minute of Demonstrations

arXiv · PDF · Open in the Atlas

Code that ran

12 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; 3 honoured the contract we drafted, 1 violated it and 5 have no contract verdict; 3 failures of our instrument; 1 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/2303.01497.json · badge: /api/ran/2303.01497.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
siddhanthaldar/FISH canonical (classifier label)
classifier basis not recorded on the link
12 ran · 1 no recorded run
FunctionStatusWhere it lives
compute_gradient_penalty Ran
ran on a synthesized input
siddhanthaldar/FISH/FISH/agent/dac.py
code served (permissive licence) · get_code("52cbdb60c855f08b")
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
import torch.nn as nn
from torch import autograd

torch.manual_seed(0)

# Define a simple discriminator model
class SimpleDiscriminator(nn.Module):
    def __init__(self, input_dim):
        super().__init__()
        self.linear1 = nn.Linear(input_dim, 8)
        self.linear2 = nn.Linear(8, 1)
    
    def forward(self, x):
        x = torch.relu(self.linear1(x))
        x = self.linear2(x)
        return x

# Create discriminator
discriminator = SimpleDiscriminator(input_dim=4)

# Create small tensors with batch_size=2, feature_dim=4
# expert_data and policy_data need to require gradients for autograd.grad to work
expert_data = torch.randn(2, 4, requires_grad=True)
policy_data = torch.randn(2, 4, requires_grad=True)

ARGS = {
    "discriminator": discriminator,
    "expert_data": expert_data,
    "policy_data": policy_data
}
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
cosine_distance Ran
ran, violated the contract we drafted
siddhanthaldar/FISH/FISH/rewarder.py
pointer only (licence: MIT); the stored copy get_code serves, from dwjshift/IL_ADS, is pointer only (licence: NONE) · get_code("3201fb6cb44890b3")
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 dwjshift/IL_ADS, 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.

episode_len Ran
ran, honoured the contract we drafted
siddhanthaldar/FISH/FISH/replay_buffer_hand.py
code served (permissive licence) · get_code("9f840a22d31a89bc")
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 int, ndim 0, dtype family int, 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)

# The function accesses episode.values() and calls .shape[0] on the result
# So episode should be a dict where values are tensors (or array-like with shape)
# The function returns shape[0] - 1, so we need at least 1 element in dimension 0

episode = {
    "observations": torch.randn(5, 4),
    "actions": torch.randn(5, 3),
    "rewards": torch.randn(5, 2),
}

ARGS = {
    "episode": episode
}
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 int, ndim 0, dtype family int, 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
euclidean_distance Ran
ran, honoured the contract we drafted
siddhanthaldar/FISH/FISH/rewarder.py
pointer only (licence: MIT); the stored copy get_code serves, from dwjshift/IL_ADS, is pointer only (licence: NONE) · get_code("5ddff5153e2b8494")
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 dwjshift/IL_ADS, 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.

load_episode Ran
ran; our contract draft was wrong, not the code
siddhanthaldar/FISH/FISH/replay_buffer_hand.py
code served (permissive licence) · get_code("184876e5f4ab6562")
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 numpy as np
from pathlib import Path
import tempfile
import os

# Create a temporary file to serve as the episode file
temp_dir = tempfile.mkdtemp()
temp_file_path = Path(temp_dir) / "episode.npz"

# Create a sample episode dictionary with numpy arrays
episode_data = {
    'observations': np.random.rand(5, 4).astype(np.float32),
    'actions': np.random.randint(0, 3, size=(5,)).astype(np.int32),
    'rewards': np.random.rand(5).astype(np.float32),
    'dones': np.random.randint(0, 2, size=(5,)).astype(bool),
}

# Save as npz file
np.savez(temp_file_path, **episode_data)

# Define ARGS with the file path
ARGS = {
    'fn': temp_file_path
}
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 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
make_agent Ran
ran on a synthesized input
siddhanthaldar/FISH/FISH/train_hand.py
code served (permissive licence) · get_code("b6a479ef98b42e5f")
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 torch.nn as nn
from dataclasses import dataclass, field
from typing import Dict, Any

torch.manual_seed(0)

# Create a minimal config class with nested structure
@dataclass
class AgentConfig:
    obs_shape: tuple = None
    action_shape: tuple = None
    env_name: str = None
    _target_: str = "builtins.dict"

@dataclass
class SuiteConfig:
    task_name: str = "test_task"

@dataclass
class MainConfig:
    obs_type: str = "image"
    agent: AgentConfig = field(default_factory=AgentConfig)
    suite: SuiteConfig = field(default_factory=SuiteConfig)

# Create obs_spec as a dict-like object with different observation types
class ObsSpec:
    def __init__(self):
        self.image = type('obj', (object,), {'shape': (4, 84, 84)})()
        self.vector = type('obj', (object,), {'shape': (12,)})()
    
    def __getitem__(self, key):
        return getattr(self, key)

# Create action_spec with shape attribute
class ActionSpec:
    def __init__(self):
        self.shape = (6,)

# Build the arguments
obs_spec = ObsSpec()
action_spec = ActionSpec()
cfg = MainConfig()

# Update agent config to use hydra instantiate target
cfg.agent._target_ = "builtins.dict"

ARGS = {
    "obs_spec": obs_spec,
    "action_spec": action_spec,
    "cfg": cfg,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
hydra, 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
make_agent Ran
ran on a synthesized input
siddhanthaldar/FISH/FISH/train_robot.py
code served (permissive licence) · get_code("552d3925ad4f9b07")
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 Identity. 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
from hydra.utils import instantiate
from omegaconf import DictConfig

torch.manual_seed(0)

# Create a simple stand-in class for specs
class ObsSpec:
    def __init__(self, name, shape):
        self.name = name
        self.shape = shape

class ActionSpec:
    def __init__(self, shape):
        self.shape = shape

# Create obs_spec as a dict-like object with obs types
obs_spec = {
    'image': ObsSpec('image', (4, 3, 32, 32)),
    'state': ObsSpec('state', (4, 10)),
    'vector': ObsSpec('vector', (4, 8))
}

# Create action_spec
action_spec = ActionSpec((4, 6))

# Create a minimal config that can be instantiated
# The config should have a _target_ to specify what to instantiate
cfg = DictConfig({
    '_target_': 'torch.nn.Identity',
    'obs_type': 'image'
})

ARGS = {
    'obs_spec': obs_spec,
    'action_spec': action_spec,
    'cfg': cfg
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
hydra, omegaconf, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Identity — a Identity output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Identity 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
make_replay_loader Ran
ran; our contract draft was wrong, not the code
siddhanthaldar/FISH/FISH/replay_buffer_hand.py
code served (permissive licence) · get_code("cc4c5cba4f6f2757")
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 DataLoader. 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 tempfile
import numpy as np
import torch
from pathlib import Path

torch.manual_seed(0)
np.random.seed(0)

# Create a temporary directory to serve as replay_dir
temp_dir = tempfile.TemporaryDirectory()
replay_dir = Path(temp_dir.name)

# Create a minimal episode file in the replay directory
# Episodes are stored as .npz files with naming convention: episode_<eps_idx>_<eps_len>.npz
episode_data = {
    'observation': np.random.randn(20, 4).astype(np.float32),  # 20 timesteps, 4-dim observations
    'action': np.random.randn(20, 3).astype(np.float32),  # 20 timesteps, 3-dim actions
    'vinn_action': np.random.randn(20, 3).astype(np.float32),  # 20 timesteps, 3-dim vinn_action
    'reward': np.random.randn(20, 1).astype(np.float32),  # 20 timesteps, scalar rewards
    'discount': np.ones((20, 1), dtype=np.float32),  # 20 timesteps, scalar discounts
}

# episode_len() returns shape[0] - 1, so with 20 timesteps, episode_len = 19
episode_path = replay_dir / 'episode_0_19.npz'
np.savez(episode_path, **episode_data)

# Define ARGS for make_replay_loader
ARGS = {
    'replay_dir': replay_dir,
    'max_size': 1000,
    'batch_size': 2,
    'num_workers': 0,
    'save_snapshot': False,
    'nstep': 2,
    'discount': 0.99,
}
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 DataLoader — a DataLoader output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataLoader 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
optimal_transport_plan Ran
ran on a synthesized input
siddhanthaldar/FISH/FISH/rewarder.py
code served (permissive licence) · get_code("d1fbf3515dde771e")
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, 3) 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
import torch
import ot

torch.manual_seed(0)

# Create small test tensors
X = torch.randn(2, 4, requires_grad=False)  # Shape: (2, 4)
Y = torch.randn(3, 4, requires_grad=False)  # Shape: (3, 4)

# Create cost matrix - must be compatible with X and Y shapes
# Cost matrix should have shape (X.shape[0], Y.shape[0]) = (2, 3)
cost_matrix = torch.randn(2, 3, requires_grad=True)

ARGS = {
    'X': X,
    'Y': Y,
    'cost_matrix': cost_matrix,
    'method': 'sinkhorn_gpu',
    'niter': 500,
    'epsilon': 0.01
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, ot, 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, 3) — shape (2, 3) 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
schedule Ran
ran, honoured the contract we drafted
siddhanthaldar/FISH/FISH/utils.py
code served (permissive licence) · get_code("ae234269debed3f4")
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 float, 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 re
import numpy as np

# Test case 1: simple float string
ARGS = {
    "schdl": "0.5",
    "step": 10
}
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 float, 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
to_torch Ran
ran; our contract draft was wrong, not the code
siddhanthaldar/FISH/FISH/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from premiertaco/premier-taco, is pointer only (licence: NONE) · get_code("6244e9922f4e0610")
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 premiertaco/premier-taco, 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.

update_mean_var_count_from_moments Ran
ran on a synthesized input
siddhanthaldar/FISH/FISH/utils.py
code served (permissive licence) · get_code("19a25f21c8bbd6b7")
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)

# Create small tensors for the function parameters
# The function expects tensors that support arithmetic operations and torch.pow
mean = torch.randn(4)
var = torch.randn(4)
count = torch.tensor(10.0)
batch_mean = torch.randn(4)
batch_var = torch.randn(4)
batch_count = torch.tensor(5.0)

ARGS = {
    "mean": mean,
    "var": var,
    "count": count,
    "batch_mean": batch_mean,
    "batch_var": batch_var,
    "batch_count": batch_count,
}
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
make_agent Not yet run
unverified (harvested, no recorded run)
siddhanthaldar/FISH/FISH/eval_robot.py
code served (permissive licence) · get_code("6dcf7786ccfa6051")
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, 9 are also present in other papers whose code we hold (9 with at least one paper dated earlier, 0 the same month, 7 later) and 4 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
to_torch 23 other papers
10 dated earlier, 13 later · 7 only as an assembly brick
2103.04551 (2021-03)
Behavior From the Void: Unsupervised Active Pre-Training
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2202.10324 (2022-02)
VRL3: A Data-Driven Framework for Visual Deep Reinforcement Learning
and 20 more not named here
schedule 17 other papers
10 dated earlier, 7 later · 7 only as an assembly brick
2103.04551 (2021-03)
Behavior From the Void: Unsupervised Active Pre-Training
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2202.10324 (2022-02)
VRL3: A Data-Driven Framework for Visual Deep Reinforcement Learning
and 14 more not named here
episode_len 12 other papers
3 dated earlier, 9 later
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2211.12740 (2022-11)
Masked Autoencoding for Scalable and Generalizable Decision Making
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
and 9 more not named here
load_episode 10 other papers
2 dated earlier, 8 later
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
2305.04477 (2023-05)
Behavior Contrastive Learning for Unsupervised Skill Discovery
and 7 more not named here
make_replay_loader 3 other papers
1 dated earlier, 2 later
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2405.16907 (2024-05)
GTA: Generative Trajectory Augmentation with Guidance for Offline Reinforcement Learning
2405.17416 (2024-05)
A Recipe for Unbounded Data Augmentation in Visual Reinforcement Learning
cosine_distance 2 other papers
1 dated earlier, 1 later
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
2310.07433 (2023-10)
Imitation Learning from Observation with Automatic Discount Scheduling
euclidean_distance 2 other papers
1 dated earlier, 1 later
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
2310.07433 (2023-10)
Imitation Learning from Observation with Automatic Discount Scheduling
optimal_transport_plan 1 other paper
1 dated earlier
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
update_mean_var_count_from_moments 1 other paper
1 dated earlier
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards

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

While imitation learning provides us with an efficient toolkit to train robots, learning skills that are robust to environment variations remains a significant challenge. Current approaches address this challenge by relying either on large amounts of demonstrations that span environment variations or on handcrafted reward functions that require state estimates. Both directions are not scalable to fast imitation. In this work, we present Fast Imitation of Skills from Humans (FISH), a new imitation learning approach that can learn robust visual skills with less than a minute of human demonstrations. Given a weak base-policy trained by offline imitation of demonstrations, FISH computes rewards that correspond to the "match" between the robot's behavior and the demonstrations. These rewards are then used to adaptively update a residual policy that adds on to the base-policy. Across all tasks, FISH requires at most twenty minutes of interactive learning to imitate demonstrations on object configurations that were not seen in the demonstrations. Importantly, FISH is constructed to be versatile, which allows it to be used across robot morphologies (e.g. xArm, Allegro, Stretch) and camera configurations (e.g. third-person, eye-in-hand). Our experimental evaluations on 9 different tasks show that FISH achieves an average success rate of 93%, which is around 3.8x higher than prior state-of-the-art methods.

For agents

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

get_harvested_code_for_paper("2303.01497")
get_code_for_paper("2303.01497")
have("2303.01497")

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

curl https://syntology.ai/api/ran/2303.01497.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.01497.svg)](https://syntology.ai/paper/2303.01497)

Connect an agent — have() is free.