SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2107.06011 · 2021

Teaching Agents how to Map: Spatial Reasoning for Multi-Object Navigation

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 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 4 have no contract verdict; 2 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/2107.06011.json · badge: /api/ran/2107.06011.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
PierreMarza/teaching_agents_how_to_map canonical (classifier label)
classifier basis: census
2 ran · 0 no recorded run
PierreMarza/autonerf community (classifier label)
classifier basis not recorded on the link
4 ran · 1 no recorded run
FunctionStatusWhere it lives
center_crop Ran
ran on a synthesized input
PierreMarza/teaching_agents_how_to_map/habitat/core/utils.py
code served (permissive licence) · get_code("51e5403cab9bef01")
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, 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 torch
import numpy as np

torch.manual_seed(0)

# Create a sample observation tensor with shape (height, width, channels)
# Using distinct dimensions: 8 height, 6 width, 3 channels
obs = torch.randn(8, 6, 3).numpy()

# Define the new shape for cropping (4 height, 2 width)
new_shape = (4, 2)

ARGS = {
    "obs": obs,
    "new_shape": new_shape,
}
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 ndarray, ndim 3, dtype family float, shape (4, 2, 3) — shape (4, 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
checking_arr_in_list Ran
ran on a synthesized input
PierreMarza/autonerf/policy/nerf_data_saving.py
code served (permissive licence) · get_code("c3c616b74845ca0d")
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 bool, 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 math
import numpy as np

# Construct valid arguments for checking_arr_in_list
# The function checks if an array matches any array in a list based on:
# - norm of first element (arr[:1] vs arr_[:1])
# - absolute difference of third element (arr[2] vs arr_[2])

# Create a test array with at least 3 elements
arr = np.array([1.0, 2.0, 0.5, 4.0])

# Create a list of arrays to search through
# Include one that will match based on the criteria
arr_match = np.array([1.01, 2.5, math.radians(0.5), 4.5])  # Close match
arr_nomatch = np.array([5.0, 6.0, math.radians(10.0), 7.0])  # No match

list_of_arrays = [arr_match, arr_nomatch]

# The 'args' parameter is not used in the function body, but must be provided
args = {}

ARGS = {
    "arr": arr,
    "list": list_of_arrays,
    "args": args
}
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 bool, 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
compute_reward Ran
ran on a synthesized input
PierreMarza/autonerf/policy/rewards.py
code served (permissive licence) · get_code("8916fff81fdefd69")
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 1, dtype family float, shape (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)

class Args:
    def __init__(self):
        self.map_resolution = 100.0
        self.expl_rew_coeff = 0.5
        self.sem_obj_rew_coeff = 0.3
        self.obs_rew_coeff = 0.2
        self.viewpoint_rew_coeff = 0.4

args = Args()
device = torch.device('cpu')
num_scenes = 2

# full_map_prev and full_map need to have at least 21 channels (0-20+)
# Channel 0: obs, Channel 1: expl, Channels 4-19: sem_obj, Channels 20+: viewpoint
full_map_prev = torch.zeros(num_scenes, 25, 4, 3)
full_map = torch.zeros(num_scenes, 25, 4, 3)

# Add some values to ensure nb_voxels_new >= nb_voxels_prev
full_map_prev[0, 1] = torch.ones(4, 3) * 0.5  # exploration channel
full_map[0, 1] = torch.ones(4, 3) * 0.7
full_map_prev[1, 0] = torch.ones(4, 3) * 0.2  # obstacle channel
full_map[1, 0] = torch.ones(4, 3) * 0.4

use_expl_reward = True
use_sem_obj_reward = False
use_obs_reward = False
use_viewpoint_reward = False

ARGS = {
    "args": args,
    "device": device,
    "num_scenes": num_scenes,
    "full_map_prev": full_map_prev,
    "full_map": full_map,
    "use_expl_reward": use_expl_reward,
    "use_sem_obj_reward": use_sem_obj_reward,
    "use_obs_reward": use_obs_reward,
    "use_viewpoint_reward": use_viewpoint_reward,
}
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 1, dtype family float, shape (2,) — shape (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
get_grid Ran
ran; our contract draft was wrong, not the code
PierreMarza/autonerf/policy/utils/model.py
code served (permissive licence) · get_code("3371127a0ac34f26")
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 5, 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 numpy as np

torch.manual_seed(0)

# Define the arguments for get_grid function
device = torch.device('cpu')

# pose: FloatTensor(bs, 3) where bs=2, and 3 represents [x, y, theta]
pose = torch.tensor([
    [0.5, 1.0, 45.0],
    [1.5, 2.0, 90.0]
], dtype=torch.float32)

# grid_size: 4-tuple (bs, _, grid_h, grid_w)
# bs=2, _=4 (arbitrary channel dimension), grid_h=3, grid_w=5
grid_size = (2, 4, 3, 5)

ARGS = {
    'pose': pose,
    'grid_size': grid_size,
    'device': device
}
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 tuple, ndim 5, 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
saving_autonerf_data Ran
ran on a synthesized input
PierreMarza/autonerf/policy/nerf_data_saving.py
code served (permissive licence) · get_code("a25c2700546d6abd")
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 1, dtype family int, shape (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 numpy as np
import torch
import tempfile
import os

# Create a temporary directory for autonerf_dataset_path
temp_dir = tempfile.mkdtemp()

# Build the args object with required attributes
class Args:
    def __init__(self):
        self.camera_height = 1.5
        self.min_depth = 0.1
        self.max_depth = 10.0

args = Args()

# Create infos - list of dicts with scene_name and sim_location
infos = [
    {
        'scene_name': 'scene_0',
        'sim_location': np.array([1.0, 2.0, 0.5])
    },
    {
        'scene_name': 'scene_1',
        'sim_location': np.array([3.0, 4.0, 1.0])
    }
]

# Create autonerf_data - batch of images with RGB, depth, and semantic channels
# Shape: (num_scenes, height, width, channels) where channels = 3 (RGB) + 1 (depth) + 4 (semantic)
num_scenes = 2
height = 64
width = 64
semantic_channels = 4
autonerf_data = np.random.rand(num_scenes, height, width, 3 + 1 + semantic_channels).astype(np.float32)
autonerf_data[:, :, :, :3] = np.clip(autonerf_data[:, :, :, :3] * 255, 0, 255)  # RGB values
autonerf_data[:, :, :, 3] = np.random.rand(num_scenes, height, width)  # Depth in [0, 1]

# Create frame_ids
frame_ids = np.array([0, 1], dtype=np.int32)

# Create autonerf_dataset_path
autonerf_dataset_path = temp_dir

# Create scene2traj_id mapping
scene2traj_id = {
    'scene_0': 0,
    'scene_1': 1
}

# Create pose_sets - list of lists to store poses
pose_sets = [[] for _ in range(num_scenes)]

# Create floor_levels
floor_levels = np.array([0, 0], dtype=np.int32)

# Create semantic ground truth
sem_gt = np.random.randint(0, 10, size=(num_scenes, height, width), dtype=np.uint8)

# Create instance semantic ground truth
instance_sem_gt = np.random.randint(0, 10, size=(num_scenes, height, width), dtype=np.uint8)

# Create wait_env as a tensor with shape (1,) and value 0
wait_env = np.array([0], dtype=np.int32)

ARGS = {
    'infos': infos,
    'autonerf_data': autonerf_data,
    'num_scenes': num_scenes,
    'args': args,
    'frame_ids': frame_ids,
    'autonerf_dataset_path': autonerf_dataset_path,
    'scene2traj_id': scene2traj_id,
    'pose_sets': pose_sets,
    'floor_levels': floor_levels,
    'sem_gt': sem_gt,
    'instance_sem_gt': instance_sem_gt,
    'wait_env': wait_env
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (2,) — shape (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
tile_images Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
PierreMarza/teaching_agents_how_to_map/habitat/core/utils.py
code served (permissive licence) · get_code("a44f38d8e312d9dc")
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 int, shape (8, 10, 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 list of small sample images
# Each image has shape (height x width x channels)
image1 = np.random.randint(0, 256, (4, 5, 3), dtype=np.uint8)
image2 = np.random.randint(0, 256, (4, 5, 3), dtype=np.uint8)
image3 = np.random.randint(0, 256, (4, 5, 3), dtype=np.uint8)

images = [image1, image2, image3]

ARGS = {
    "images": images
}
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 int, shape (8, 10, 3) — shape (8, 10, 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
copying_frames Not yet run
unverified (harvested, no recorded run)
PierreMarza/autonerf/nerfstudio/convert_autonerf_data_to_nerfstudio_format.py
code served (permissive licence) · get_code("3a95b94c54cbcc58")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 2 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
get_grid 8 other papers
3 dated earlier, 5 later · 1 only as an assembly brick
1912.06321 (2019-12)
Sim2Real Predictivity: Does Evaluation in Simulation Predict Real-World Performance?
2004.05155 (2020-04)
Learning to Explore using Active Neural SLAM
2007.00643 (2020-07)
Object Goal Navigation using Goal-Oriented Semantic Exploration
and 5 more not named here
tile_images 3 other papers
2 dated earlier, 1 later
1904.01201 (2019-04)
Habitat: A Platform for Embodied AI Research
2010.13439 (2020-10)
On Embodied Visual Navigation in Real Environments Through Habitat
2310.13724 (2023-10)
Habitat 3.0: A Co-Habitat for Humans, Avatars and Robots
center_crop 1 other paper
1 dated earlier
2010.13439 (2020-10)
On Embodied Visual Navigation in Real Environments Through Habitat

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

In the context of visual navigation, the capacity to map a novel environment is necessary for an agent to exploit its observation history in the considered place and efficiently reach known goals. This ability can be associated with spatial reasoning, where an agent is able to perceive spatial relationships and regularities, and discover object characteristics. Recent work introduces learnable policies parametrized by deep neural networks and trained with Reinforcement Learning (RL). In classical RL setups, the capacity to map and reason spatially is learned end-to-end, from reward alone. In this setting, we introduce supplementary supervision in the form of auxiliary tasks designed to favor the emergence of spatial perception capabilities in agents trained for a goal-reaching downstream objective. We show that learning to estimate metrics quantifying the spatial relationships between an agent at a given location and a goal to reach has a high positive impact in Multi-Object Navigation settings. Our method significantly improves the performance of different baseline agents, that either build an explicit or implicit representation of the environment, even matching the performance of incomparable oracle agents taking ground-truth maps as input. A learning-based agent from the literature trained with the proposed auxiliary losses was the winning entry to the Multi-Object Navigation Challenge, part of the CVPR 2021 Embodied AI Workshop.

For agents

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

get_harvested_code_for_paper("2107.06011")
get_code_for_paper("2107.06011")
have("2107.06011")

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

curl https://syntology.ai/api/ran/2107.06011.json

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

[![Syntology run record](https://syntology.ai/api/ran/2107.06011.svg)](https://syntology.ai/paper/2107.06011)

Connect an agent — have() is free.