SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.19972 · NeurIPS · 2024

HumanVLA: Towards Vision-Language Directed Object Rearrangement by Physical Humanoid

Yong-Lu Li, Cewu Lu, Lei Han, Xinyu Xu, Yizheng Zhang

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (4 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 0 failures of our instrument; 0 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/2406.19972.json · badge: /api/ran/2406.19972.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
AllenXuuu/HumanVLA canonical (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
FunctionStatusWhere it lives
BaseNetwork Ran
ran on a synthesized input · constructed an object; no result was computed
AllenXuuu/HumanVLA/models/dagger/vla_network.py
code served (permissive licence) · get_code("661a3b9f4cdce825")
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, BaseNetwork: 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 BaseNetwork(nn.Module):

    def build_mlp(self, in_dim, hiddens, activation='relu', last_activation=True, last_bias=True):
        net = nn.Sequential()
        for i, h in enumerate(hiddens):
            bias = not (i == len(hiddens) - 1 and (not last_bias))
            net.append(nn.Linear(in_dim, h, bias=bias))
            net.append(self.build_activation(activation))
            in_dim = h
        if not last_activation:
            net.pop(-1)
        return net

    def build_activation(self, name):
        if name is None:
            return nn.Identity()
        elif name == 'relu':
            return nn.ReLU()
        elif name == 'softmax':
            return nn.Softmax(dim=-1)
        elif name == 'softplus':
            return nn.Softplus()
        elif name == 'sigmoid':
            return nn.Sigmoid()
        raise NotImplementedError(name)

    def neglogp(self, x, mean, std, logstd):
        neglogp = 0.5 * ((x - mean) / std) ** 2 + 0.5 * np.log(2.0 * np.pi) + logstd
        neglogp = neglogp.sum(-1)
        return neglogp

    def sync_stats(self):
        for name, module in self.named_modules():
            if isinstance(module, (RunningMeanStd,)):
                module.sync_stats()


class RunningMeanStd(nn.Module):

    def __init__(self, insize, epsilon=1e-05):
        super(RunningMeanStd, self).__init__()
        self.insize = insize
        if isinstance(self.insize, int):
            self.insize = (self.insize,)
        self.epsilon = epsilon
        self.running_mean = nn.Parameter(torch.zeros(self.insize, dtype=torch.float32), requires_grad=False)
        self.running_var = nn.Parameter(torch.ones(self.insize, dtype=torch.float32), requires_grad=False)
        self.running_count = nn.Parameter(torch.zeros(1, dtype=torch.float32), requires_grad=False)
        self.trainable = True

    def freeze(self):
        self.trainable = False

    def forward(self, input, unnorm=False):
        input_shape = input.shape
        input = input.view((-1,) + self.insize)
        if self.training and self.trainable:
            assert not unnorm
            bz = input.shape[0]
            delta_mean = input.mean(0) - self.running_mean
            new_count = self.running_count + bz
            new_mean = self.running_mean + delta_mean * bz / new_count
            new_var = self.running_var * self.running_count / new_count + input.var(0) * bz / new_count + delta_mean ** 2 * self.running_count * bz / new_count ** 2
            self.running_mean[:] = new_mean
            self.running_var[:] = new_var
            self.running_count[:] = new_count
        current_mean = self.running_mean
        current_var = self.running_var
        if unnorm:
            y = torch.clamp(input, min=-5, max=5)
            y = torch.sqrt(current_var.float() + self.epsilon) * y + current_mean.float()
        else:
            y = (input - current_mean.float()) / torch.sqrt(current_var.float() + self.epsilon)
            y = torch.clamp(y, min=-5.0, max=5.0)
        return y.view(input_shape)

    def sync_stats(self):
        pass


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 BaseNetwork — 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 BaseNetwork 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
ResBlock Ran
ran on a synthesized input · constructed an object; no result was computed
AllenXuuu/HumanVLA/models/dagger/vla_network.py
code served (permissive licence) · get_code("bd2f4055d9632f22")
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, ResBlock: 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

torch.manual_seed(0)

# ResBlock expects a single parameter: dim (integer)
# The forward method expects input tensor x of shape (batch, dim)
# where it applies layers and adds a residual connection

ARGS = {
    'dim': 4
}
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 ResBlock — 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 ResBlock 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
RunningMeanStd Ran
ran on a synthesized input · constructed an object; no result was computed
AllenXuuu/HumanVLA/models/dagger/vla_network.py
code served (permissive licence) · get_code("94e925f5c51f7d0c")
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, RunningMeanStd: 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

torch.manual_seed(0)

# RunningMeanStd expects:
# - insize: int or tuple specifying the size of the input (excluding batch dimension)
# - epsilon: float for numerical stability (optional, defaults to 1e-05)

# For this setup, we'll use a tuple insize of (4, 3) to have distinct dimensions
insize = (4, 3)
epsilon = 1e-05

ARGS = {
    "insize": insize,
    "epsilon": epsilon,
}
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 RunningMeanStd — 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 RunningMeanStd 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
VLANetwork Ran
ran on a synthesized input · constructed an object; no result was computed
AllenXuuu/HumanVLA/models/dagger/vla_network.py
code served (permissive licence) · get_code("494ec3a18baeb896")
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, VLANetwork: 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 Config:
    def __init__(self):
        self.prop_dim = 5
        self.text_dim = 8
        self.normalize_prop = True
        self.image_pre = type('obj', (object,), {'hidden': [1280, 256, 128]})()
        self.text_pre = type('obj', (object,), {'hidden': [8, 64, 128]})()

cfg = Config()

ARGS = {
    'cfg': cfg
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type VLANetwork — 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 VLANetwork 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 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 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.

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

Please push the small brown table next to the bed. I want to work. Place the laptop on the desk. Move the chair in front of the desk. Figure 1: HumanVLA performs various object rearrangement tasks directed by the egocentric vision and natural language instructions.

For agents

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

get_harvested_code_for_paper("2406.19972")
get_code_for_paper("2406.19972")
have("2406.19972")

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

curl https://syntology.ai/api/ran/2406.19972.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.19972.svg)](https://syntology.ai/paper/2406.19972)

Connect an agent — have() is free.