SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2108.05877 · 2021

DexMV: Imitation Learning for Dexterous Manipulation from Human Videos

arXiv · PDF · Open in the Atlas

Code that ran

8 of 12 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 8 have no contract verdict; 0 failures of our instrument; 4 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2108.05877.json · badge: /api/ran/2108.05877.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
yzqin/dexmv-sim canonical (classifier label)
classifier basis: census
8 ran · 4 no recorded run
FunctionStatusWhere it lives
filter_position_sequence Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/joint_utils.py
code served (permissive licence) · get_code("35199fc7a0f39968")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family float, shape (10, 4, 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
from scipy import signal

# Set random seed for reproducibility
np.random.seed(0)

# Create test data for 2D case (time_steps, num_joints)
position_seq_2d = np.random.randn(10, 4)

# Create test data for 3D case (time_steps, num_joints, coordinates)
position_seq_3d = np.random.randn(10, 4, 3)

# Define ARGS dict with parameters for the function
ARGS = {
    "position_seq": position_seq_3d,  # Using 3D array (time_steps=10, joints=4, coords=3)
    "wn": 5,
    "fs": 25
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (10, 4, 3) — shape (10, 4, 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
get_point_cloud_from_depth Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/camera_utils.py
code served (permissive licence) · get_code("1cdd31e80225df5b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family float, shape (4, 3, 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

# Create a small depth image (4x3 for distinct dimensions)
depth_image = np.random.rand(4, 3) * 5 + 0.1  # depth values between 0.1 and 5.1

# Create intrinsic matrix (3x3 camera intrinsics)
intrinsic = np.array([
    [500.0, 0.0, 160.0],
    [0.0, 500.0, 120.0],
    [0.0, 0.0, 1.0]
], dtype=np.float32)

# Create extrinsic matrix (4x4 transformation matrix)
extrinsic = np.array([
    [1.0, 0.0, 0.0, 0.1],
    [0.0, 1.0, 0.0, 0.2],
    [0.0, 0.0, 1.0, 0.3],
    [0.0, 0.0, 0.0, 1.0]
], dtype=np.float32)

ARGS = {
    "depth_image": depth_image,
    "intrinsic": intrinsic,
    "extrinsic": extrinsic
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (4, 3, 3) — shape (4, 3, 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
inverse_pose Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/pose_utils.py
code served (permissive licence) · get_code("8671435fde043531")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 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 numpy as np

# Create a valid 4x4 homogeneous transformation matrix
# This represents a pose with a 3x3 rotation matrix and a 3x1 translation vector
pose = np.array([
    [0.0, -1.0, 0.0, 1.0],
    [1.0, 0.0, 0.0, 2.0],
    [0.0, 0.0, 1.0, 3.0],
    [0.0, 0.0, 0.0, 1.0]
], dtype=np.float32)

ARGS = {
    "pose": pose
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 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
load_hand_object_data Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/data_utils.py
code served (permissive licence) · get_code("e78e725294c4e4ca")
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 tempfile
import os
import numpy as np

# Create temporary directories for hand and object data
temp_dir = tempfile.mkdtemp()
hand_dir = os.path.join(temp_dir, 'hand')
object_dir = os.path.join(temp_dir, 'object')
os.makedirs(hand_dir, exist_ok=True)
os.makedirs(object_dir, exist_ok=True)

# Create sample joint data files
# Joint data shape: (num_joints, 3) for 3D positions
num_samples = 2
num_joints = 5
for i in range(1, num_samples + 1):
    joint_data = np.random.randn(num_joints, 3).astype(np.float32)
    joint_file = os.path.join(hand_dir, f'joints_{i}.npy')
    np.save(joint_file, joint_data)

# Create sample object data files
# Object data shape varies: some valid (3x4 pose matrices), some invalid (< 12 elements)
for i in range(1, num_samples + 1):
    if i == 1:
        # Valid object pose: 3x4 matrix (will be concatenated to 4x4)
        object_data = np.random.randn(3, 4).astype(np.float32)
    else:
        # Invalid object data (size < 12)
        object_data = np.array([1.0, 2.0], dtype=np.float32)
    
    object_file = os.path.join(object_dir, f'{i}.npy')
    np.save(object_file, object_data)

# Create extrinsic matrix (4x4 homogeneous transformation)
extrinsic = np.array([
    [1.0, 0.0, 0.0, 0.1],
    [0.0, 1.0, 0.0, 0.2],
    [0.0, 0.0, 1.0, 0.3],
    [0.0, 0.0, 0.0, 1.0]
], dtype=np.float32)

ARGS = {
    'hand_dir': hand_dir,
    'object_dir': object_dir,
    'target_joint_index': 0,
    'extrinsic': extrinsic
}
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
load_hand_object_data_v2 Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/data_utils.py
code served (permissive licence) · get_code("1bdbc11ee90442b7")
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
import os
import tempfile
import shutil

# Create temporary directories for the function to use
temp_hand_dir = tempfile.mkdtemp()
temp_object_dir = tempfile.mkdtemp()

# Create sample joint files
for i in range(1, 3):
    joint_data = np.random.randn(21, 3)
    np.save(os.path.join(temp_hand_dir, f'joints_{i}.npy'), joint_data)
    
    frame_data = np.random.randn(5, 4, 4)
    np.save(os.path.join(temp_hand_dir, f'results_global_{i}.npy'), frame_data)

# Create sample object files
for i in range(1, 3):
    object_data = {
        '10': np.random.randn(4, 4),
        '14': np.random.randn(4, 4),
    }
    np.save(os.path.join(temp_object_dir, f'{i}.npy'), object_data)

# Set up arguments
hand_dir = temp_hand_dir
object_dir = temp_object_dir
target_joint_index = 5
extrinsic = np.array([
    [1, 0, 0, 0.1],
    [0, 1, 0, 0.2],
    [0, 0, 1, 0.3],
    [0, 0, 0, 1]
])
hand_cam_to_object_cam = np.array([
    [1, 0, 0, 0.05],
    [0, 1, 0, 0.05],
    [0, 0, 1, 0.05],
    [0, 0, 0, 1]
])

ARGS = {
    'hand_dir': hand_dir,
    'object_dir': object_dir,
    'target_joint_index': target_joint_index,
    'extrinsic': extrinsic,
    'hand_cam_to_object_cam': hand_cam_to_object_cam
}
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
mjCOST Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/min_jerk.py
code served (permissive licence) · get_code("3a2d0b3130bd15db")
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 float64, ndim 0, 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 numpy as np
import math

# Analyze the function signature and usage:
# mjCOST(t, x, v0, a0, t0)
# - t: time points, used in np.concatenate and np.diff
# - x: positions, shape (N, D) where N = max(x.shape), D = min(x.shape)
# - v0: initial velocities, list/tuple with 2 elements [v0[0], v0[1]], each shape (D,)
# - a0: initial accelerations, list/tuple with 2 elements [a0[0], a0[1]], each shape (D,)
# - t0: time boundaries, list/tuple with 2 elements [t0[0], t0[1]], each shape (1,) or scalar array

# Let's construct small valid arguments
# N = 5 (number of waypoints)
# D = 3 (dimension)

N = 5
D = 3

# x: position waypoints, shape (N, D) = (5, 3)
x = np.array([
    [0.0, 0.0, 0.0],
    [1.0, 1.0, 1.0],
    [2.0, 1.5, 0.5],
    [3.0, 2.0, 1.0],
    [4.0, 2.5, 1.5]
])

# t: intermediate time points (between start and end)
# Should have N-2 = 3 points
t = np.array([0.5, 1.0, 1.5])

# v0: initial velocities at start and end, each shape (D,)
v0 = [np.array([0.0, 0.0, 0.0]), np.array([0.0, 0.0, 0.0])]

# a0: initial accelerations at start and end, each shape (D,)
a0 = [np.array([0.0, 0.0, 0.0]), np.array([0.0, 0.0, 0.0])]

# t0: boundary times, [start_time, end_time]
# Each should be a 1D array or scalar array
t0 = [np.array([0.0]), np.array([2.0])]

ARGS = {
    't': t,
    'x': x,
    'v0': v0,
    'a0': a0,
    't0': t0
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
mjTRJ Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/min_jerk.py
code served (permissive licence) · get_code("f16ddbbbda1d913a")
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 3, 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 numpy as np
import math

# Analyze the function signatures:
# mjTRJ(tx, x, v0, a0, t0, P)
# - tx: array of intermediate time points
# - x: position waypoints, shape (N, D) where N = max(x.shape), D = min(x.shape)
# - v0: initial/final velocities, tuple of 2 arrays each (D,)
# - a0: initial/final accelerations, tuple of 2 arrays each (D,)
# - t0: initial/final time points, tuple of 2 arrays
# - P: number of points to generate (int)

# Let's create small valid arguments
N = 5  # number of waypoints
D = 3  # dimension

# x: position waypoints (N, D)
x = np.array([
    [0.0, 0.0, 0.0],
    [1.0, 1.0, 1.0],
    [2.0, 2.0, 2.0],
    [3.0, 3.0, 3.0],
    [4.0, 4.0, 4.0]
], dtype=np.float64)

# tx: intermediate time points between t0[0] and t0[1]
# Should have N-2 points (excluding endpoints)
tx = np.array([1.0, 2.0, 3.0], dtype=np.float64)

# v0: initial and final velocities (tuple of 2 arrays)
v0 = (
    np.array([0.0, 0.0, 0.0], dtype=np.float64),  # initial velocity
    np.array([0.0, 0.0, 0.0], dtype=np.float64)   # final velocity
)

# a0: initial and final accelerations (tuple of 2 arrays)
a0 = (
    np.array([0.0, 0.0, 0.0], dtype=np.float64),  # initial acceleration
    np.array([0.0, 0.0, 0.0], dtype=np.float64)   # final acceleration
)

# t0: initial and final times (tuple of 2 arrays)
t0 = (
    np.array([0.0], dtype=np.float64),
    np.array([4.0], dtype=np.float64)
)

# P: number of output points
P = 10

ARGS = {
    'tx': tx,
    'x': x,
    'v0': v0,
    'a0': a0,
    't0': t0,
    'P': P
}
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 3, 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
replace_nan Ran
ran on a synthesized input
yzqin/dexmv-sim/hand_imitation/misc/data_utils.py
code served (permissive licence) · get_code("48dff37ab53f3fee")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (5, 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

# Case 1: Single numpy array with some NaN values
array_with_nan = np.array([
    np.array([1.0, 2.0, 3.0]),
    np.array([np.nan, np.nan, np.nan]),
    np.array([4.0, 5.0, 6.0]),
    np.array([np.nan, np.nan, np.nan]),
    np.array([7.0, 8.0, 9.0])
])

ARGS = {
    "array_seq": array_with_nan
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (5, 3) — shape (5, 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
get_checkerboard_pose Not yet run
unverified (harvested, no recorded run)
yzqin/dexmv-sim/hand_imitation/misc/camera_utils.py
code served (permissive licence) · get_code("e297529f671ba34b")
How it ran

no invocation recorded.

make Not yet run
unverified (harvested, no recorded run)
yzqin/dexmv-sim/hand_imitation/env/environments/base.py
code served (permissive licence) · get_code("117267bf6a8ae1dd")
How it ran

no invocation recorded.

min_jerk Not yet run
unverified (harvested, no recorded run)
yzqin/dexmv-sim/hand_imitation/misc/min_jerk.py
code served (permissive licence) · get_code("7cc098f3e069ba4c")
How it ran

no invocation recorded.

np2pcd Not yet run
unverified (harvested, no recorded run)
yzqin/dexmv-sim/hand_imitation/misc/camera_utils.py
code served (permissive licence) · get_code("f42fe722b762bef1")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 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 12 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

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 significant progress has been made on understanding hand-object interactions in computer vision, it is still very challenging for robots to perform complex dexterous manipulation. In this paper, we propose a new platform and pipeline DexMV (Dexterous Manipulation from Videos) for imitation learning. We design a platform with: (i) a simulation system for complex dexterous manipulation tasks with a multi-finger robot hand and (ii) a computer vision system to record large-scale demonstrations of a human hand conducting the same tasks. In our novel pipeline, we extract 3D hand and object poses from videos, and propose a novel demonstration translation method to convert human motion to robot demonstrations. We then apply and benchmark multiple imitation learning algorithms with the demonstrations. We show that the demonstrations can indeed improve robot learning by a large margin and solve the complex tasks which reinforcement learning alone cannot solve. More details can be found in the project page: https://yzqin.github.io/dexmv

For agents

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

get_harvested_code_for_paper("2108.05877")
get_code_for_paper("2108.05877")
have("2108.05877")

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

curl https://syntology.ai/api/ran/2108.05877.json

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

[![Syntology run record](https://syntology.ai/api/ran/2108.05877.svg)](https://syntology.ai/paper/2108.05877)

Connect an agent — have() is free.