SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.08245 · ICML · 2026

STEP: Warm-Started Visuomotor Policies with Spatiotemporal Consistency Prediction

Guohao Dai, Jinhao Li, Ningyi Xu, Yijia Zhang, Shan Huang, Yingqiao Wang, Yuxuan Cong, Hao Xia

arXiv · PDF · Open in the Atlas

Code that ran

6 of 8 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 2 no recorded run), measured 2026-11-22; 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/2602.08245.json · badge: /api/ran/2602.08245.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
Kimho666/STEP canonical (classifier label)
classifier basis: census
6 ran · 2 no recorded run
FunctionStatusWhere it lives
compute_metrics Ran
ran; our contract draft was wrong, not the code
Kimho666/STEP/multirun_metrics.py
code served (permissive licence) · get_code("3351ad91683faf7f")
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 pandas as pd
import numpy as np

# Create a sample DataFrame with the required columns
np.random.seed(0)

# Create test data with proper structure
n_rows = 20
log_df = pd.DataFrame({
    'train_loss': np.random.uniform(0.1, 1.0, n_rows),
    'val_loss': np.random.uniform(0.1, 1.0, n_rows),
    'accuracy': np.random.uniform(0.5, 0.95, n_rows),
    'f1_score': np.random.uniform(0.4, 0.9, n_rows),
})

# Add some None values to test the isnull handling
log_df.loc[5:7, 'accuracy'] = None

ARGS = {
    'log_df': log_df,
    'key': 'accuracy',
    'end_step': None,
    'k_min_loss': 10,
    'k_around_max': 10,
    'max_k_window': 10,
    'replace_slash': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, scipy — 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
compute_metrics_agg Ran
ran; our contract draft was wrong, not the code
Kimho666/STEP/multirun_metrics.py
code served (permissive licence) · get_code("0cc739967c061073")
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 pandas as pd
import numpy as np

# Create sample log DataFrames
np.random.seed(0)

# First log dataframe
df1 = pd.DataFrame({
    'train_loss': [0.5, 0.4, 0.3, 0.25, 0.2, 0.15, 0.14, 0.13, 0.12, 0.11],
    'val_loss': [0.6, 0.5, 0.4, 0.35, 0.3, 0.25, 0.24, 0.23, 0.22, 0.21],
    'metric_key': [0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.38, 0.36, 0.34],
})

# Second log dataframe
df2 = pd.DataFrame({
    'train_loss': [0.55, 0.45, 0.35, 0.27, 0.22, 0.17, 0.15, 0.14, 0.13, 0.12],
    'val_loss': [0.65, 0.55, 0.45, 0.37, 0.32, 0.27, 0.25, 0.24, 0.23, 0.22],
    'metric_key': [0.12, 0.18, 0.22, 0.28, 0.32, 0.37, 0.42, 0.40, 0.38, 0.35],
})

log_dfs = [df1, df2]
key = 'metric_key'
end_step = 8
k_min_loss = 3
k_around_max = 4
max_k_window = 3
replace_slash = True

ARGS = {
    'log_dfs': log_dfs,
    'key': key,
    'end_step': end_step,
    'k_min_loss': k_min_loss,
    'k_around_max': k_around_max,
    'max_k_window': max_k_window,
    'replace_slash': replace_slash,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, scipy — 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
create_obs_key_mapper Ran
ran; our contract draft was wrong, not the code
Kimho666/STEP/eval_combined_inference.py
code served (permissive licence) · get_code("d24a3503e232501f")
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 function. 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)

# Test case 1: source_keys and target_keys are different
source_keys_1 = ["image", "state", "action"]
target_keys_1 = ["rgb", "agent_state", "agent_action"]

# Test case 2: source_keys and target_keys are the same (no mapping needed)
source_keys_2 = ["obs", "reward", "done"]
target_keys_2 = ["obs", "reward", "done"]

# Use the first test case
ARGS = {
    "source_keys": source_keys_1,
    "target_keys": target_keys_1,
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
detect_policy_type Ran
ran; our contract draft was wrong, not the code
Kimho666/STEP/eval_cfm.py
code served (permissive licence) · get_code("01ac1bfcde3e63b3")
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 str. 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

# Create minimal policy classes that the function checks for
class CFMTransformerPolicy(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x):
        return self.linear(x)

class CFMUnetPolicy(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x):
        return self.linear(x)

class DiffusionPolicy(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x):
        return self.linear(x)

class UnknownPolicy(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x):
        return self.linear(x)

torch.manual_seed(0)

# Test with CFMTransformer policy
policy = CFMTransformerPolicy()

ARGS = {
    'policy': policy
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
get_model_obs_keys Ran
ran; our contract draft was wrong, not the code
Kimho666/STEP/eval_combined_inference.py
code served (permissive licence) · get_code("338cabf4cb17d013")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class MockEncoder(nn.Module):
    def __init__(self, key_shape_map):
        super().__init__()
        self.key_shape_map = key_shape_map
    
    def forward(self, x):
        return x

class MockPolicy(nn.Module):
    def __init__(self, encoder=None, obs_encoder=None, policy=None):
        super().__init__()
        if encoder is not None:
            self.encoder = encoder
        if obs_encoder is not None:
            self.obs_encoder = obs_encoder
        if policy is not None:
            self.policy = policy
    
    def forward(self, x):
        return x

# Test case 1: Policy with encoder that has key_shape_map
key_shape_map_1 = {
    'observation': (4,),
    'state': (8,),
    'action': (2,)
}
encoder_1 = MockEncoder(key_shape_map_1)
policy_1 = MockPolicy(encoder=encoder_1)

ARGS = {
    'policy': policy_1
}
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 list, ndim 1, dtype family float — a list 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
worker_fn Ran
ran, honoured the contract we drafted
Kimho666/STEP/ray_exec.py
code served (permissive licence) · get_code("4194cf80b56b4bde")
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 os
import subprocess
import tempfile

# Create a simple test command that will succeed
# Using 'true' command or 'echo' for a safe, harmless subprocess call
ARGS = {
    'command_args': ['echo', 'test'],
    'data_src': None,
    'unbuffer_python': False,
    'use_shell': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — 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
patch_cfm_policy_solver Not yet run
unverified (harvested, no recorded run)
Kimho666/STEP/eval_cfm.py
code served (permissive licence) · get_code("9061a9fc934ec01d")
How it ran

no invocation recorded.

run_single_evaluation Not yet run
unverified (harvested, no recorded run)
Kimho666/STEP/eval_cfm.py
code served (permissive licence) · get_code("c416c26539f06ac1")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 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 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

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

Diffusion policies have recently emerged as a powerful paradigm for visuomotor control in robotic manipulation due to their ability to model the distribution of action sequences and capture multimodality. However, iterative denoising leads to substantial inference latency, limiting control frequency in real-time closed-loop systems. Existing acceleration methods either reduce sampling steps, bypass diffusion through direct prediction, or reuse past actions, but often struggle to jointly preserve action quality and achieve consistently low latency. In this work, we propose STEP, a lightweight spatiotemporal consistency prediction mechanism to construct high-quality warmstart actions that are both distributionally close to the target action and temporally consistent, without compromising the generative capability of the original diffusion policy. Then, we propose a velocity-aware perturbation injection mechanism that adaptively modulates actuation excitation based on temporal action variation to prevent execution stall especially for real-world tasks. We further provide a theoretical analysis showing that the proposed prediction induces a locally contractive mapping, ensuring convergence of action errors during diffusion refinement. We conduct extensive evaluations on nine simulated benchmarks and two real-world tasks. Notably, STEP with 2 steps can achieve an average 21.6% and 27.5% higher success rate than BRIDGER and DDIM on the RoboMimic benchmark and real-world tasks, respectively. These results demonstrate that STEP consistently advances the Pareto frontier of inference latency and success rate over existing methods. The code is publicly available at https://github.com/Kimho666/STEP.

For agents

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

get_harvested_code_for_paper("2602.08245")
get_code_for_paper("2602.08245")
have("2602.08245")

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

curl https://syntology.ai/api/ran/2602.08245.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.08245.svg)](https://syntology.ai/paper/2602.08245)

Connect an agent — have() is free.