Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
episodes: int64
successful_episodes: int64
decision_ticks: int64
tasks: list<item: string>
child 0, item: string
control_hz: double
obs_dim: int64
action_dim: int64
world_seed: int64
policy: string
ticks: list<item: struct<t: double, obs: list<item: double>, action: struct<vx: double, vy: double, vz: dou (... 95 chars omitted)
child 0, item: struct<t: double, obs: list<item: double>, action: struct<vx: double, vy: double, vz: double, yaw_ra (... 83 chars omitted)
child 0, t: double
child 1, obs: list<item: double>
child 0, item: double
child 2, action: struct<vx: double, vy: double, vz: double, yaw_rate: double, mission_state: int64>
child 0, vx: double
child 1, vy: double
child 2, vz: double
child 3, yaw_rate: double
child 4, mission_state: int64
child 3, pos: list<item: double>
child 0, item: double
child 4, yaw: double
child 5, gps: bool
outcome: struct<success: bool, reason: string, time_s: double, safety_violations: int64, battery_end: double, (... 20 chars omitted)
child 0, success: bool
child 1, reason: string
child 2, time_s: double
child 3, safety_violations: int64
child 4, battery_end: double
child 5, path_len_m: double
task: string
world: struct<structure: struct<cx: double, cy: double, w: int64, d: int64, h: int64>, fire_xy: list<item: (... 124 chars omitted)
child 0, structure: struct<cx: double, cy: double, w: int64, d: int64, h: int64>
child 0, cx: double
child 1, cy: double
child 2, w: int64
child 3, d: int64
child 4, h: int64
child 1, fire_xy: list<item: double>
child 0, item: double
child 2, plume_radius: double
child 3, victims: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 4, gps_denied_radius: double
child 5, wind: list<item: double>
child 0, item: double
episode_id: string
instruction: string
to
{'task': Value('string'), 'instruction': Value('string'), 'policy': Value('string'), 'world_seed': Value('int64'), 'world': {'structure': {'cx': Value('float64'), 'cy': Value('float64'), 'w': Value('int64'), 'd': Value('int64'), 'h': Value('int64')}, 'fire_xy': List(Value('float64')), 'plume_radius': Value('float64'), 'victims': List(List(Value('float64'))), 'gps_denied_radius': Value('float64'), 'wind': List(Value('float64'))}, 'outcome': {'success': Value('bool'), 'reason': Value('string'), 'time_s': Value('float64'), 'safety_violations': Value('int64'), 'battery_end': Value('float64'), 'path_len_m': Value('float64')}, 'ticks': List({'t': Value('float64'), 'obs': List(Value('float64')), 'action': {'vx': Value('float64'), 'vy': Value('float64'), 'vz': Value('float64'), 'yaw_rate': Value('float64'), 'mission_state': Value('int64')}, 'pos': List(Value('float64')), 'yaw': Value('float64'), 'gps': Value('bool')}), 'episode_id': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1816, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
episodes: int64
successful_episodes: int64
decision_ticks: int64
tasks: list<item: string>
child 0, item: string
control_hz: double
obs_dim: int64
action_dim: int64
world_seed: int64
policy: string
ticks: list<item: struct<t: double, obs: list<item: double>, action: struct<vx: double, vy: double, vz: dou (... 95 chars omitted)
child 0, item: struct<t: double, obs: list<item: double>, action: struct<vx: double, vy: double, vz: double, yaw_ra (... 83 chars omitted)
child 0, t: double
child 1, obs: list<item: double>
child 0, item: double
child 2, action: struct<vx: double, vy: double, vz: double, yaw_rate: double, mission_state: int64>
child 0, vx: double
child 1, vy: double
child 2, vz: double
child 3, yaw_rate: double
child 4, mission_state: int64
child 3, pos: list<item: double>
child 0, item: double
child 4, yaw: double
child 5, gps: bool
outcome: struct<success: bool, reason: string, time_s: double, safety_violations: int64, battery_end: double, (... 20 chars omitted)
child 0, success: bool
child 1, reason: string
child 2, time_s: double
child 3, safety_violations: int64
child 4, battery_end: double
child 5, path_len_m: double
task: string
world: struct<structure: struct<cx: double, cy: double, w: int64, d: int64, h: int64>, fire_xy: list<item: (... 124 chars omitted)
child 0, structure: struct<cx: double, cy: double, w: int64, d: int64, h: int64>
child 0, cx: double
child 1, cy: double
child 2, w: int64
child 3, d: int64
child 4, h: int64
child 1, fire_xy: list<item: double>
child 0, item: double
child 2, plume_radius: double
child 3, victims: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 4, gps_denied_radius: double
child 5, wind: list<item: double>
child 0, item: double
episode_id: string
instruction: string
to
{'task': Value('string'), 'instruction': Value('string'), 'policy': Value('string'), 'world_seed': Value('int64'), 'world': {'structure': {'cx': Value('float64'), 'cy': Value('float64'), 'w': Value('int64'), 'd': Value('int64'), 'h': Value('int64')}, 'fire_xy': List(Value('float64')), 'plume_radius': Value('float64'), 'victims': List(List(Value('float64'))), 'gps_denied_radius': Value('float64'), 'wind': List(Value('float64'))}, 'outcome': {'success': Value('bool'), 'reason': Value('string'), 'time_s': Value('float64'), 'safety_violations': Value('int64'), 'battery_end': Value('float64'), 'path_len_m': Value('float64')}, 'ticks': List({'t': Value('float64'), 'obs': List(Value('float64')), 'action': {'vx': Value('float64'), 'vy': Value('float64'), 'vz': Value('float64'), 'yaw_rate': Value('float64'), 'mission_state': Value('int64')}, 'pos': List(Value('float64')), 'yaw': Value('float64'), 'gps': Value('bool')}), 'episode_id': Value('string')}
because column names don't match
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
task string | instruction string | policy string | world_seed int64 | world dict | outcome dict | ticks list | episode_id string |
|---|---|---|---|---|---|---|---|
fire_assess | Determine fire origin location from aerial view | scripted | 1,000 | {
"structure": {
"cx": 40.21,
"cy": 31.04,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
35.16,
35.63
],
"plume_radius": 6,
"victims": [
[
46.21,
31.21,
5
],
[
38.36,
26.04,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.44,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 29.4,
"safety_violations": 0,
"battery_end": 84.4,
"path_len_m": 79.6
} | [
{
"t": 0,
"obs": [
0.8043,
0.6208,
0,
0.8599,
0.7031,
0.7126,
0.9243,
0.6242,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00000 |
fire_assess | Evaluate wind effect on fire spread pattern | scripted | 1,001 | {
"structure": {
"cx": 41.13,
"cy": 25.16,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
35.5,
28.44
],
"plume_radius": 6,
"victims": [
[
47.13,
22.61,
5
],
[
39.33,
20.16,
5
]
],
"gps_denied_radius": 4,
"wind": [
0.26,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28.5,
"safety_violations": 0,
"battery_end": 84.8,
"path_len_m": 78.3
} | [
{
"t": 0,
"obs": [
0.8225,
0.5031,
0,
0.81,
0.71,
0.5688,
0.9425,
0.4523,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00001 |
fire_assess | Determine if fire has reached the second floor | scripted | 1,002 | {
"structure": {
"cx": 38.81,
"cy": 28.57,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
34.3,
32.79
],
"plume_radius": 6,
"victims": [
[
44.81,
27.59,
5
],
[
39.14,
23.57,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.63,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 29.8,
"safety_violations": 0,
"battery_end": 84,
"path_len_m": 77.2
} | [
{
"t": 0,
"obs": [
0.7762,
0.5714,
0,
0.808,
0.6861,
0.6559,
0.8962,
0.5519,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00002 |
fire_assess | Assess fire condition of the ground floor | scripted | 1,003 | {
"structure": {
"cx": 36.88,
"cy": 27.57,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
31.87,
31.25
],
"plume_radius": 6,
"victims": [
[
42.88,
29.08,
5
],
[
38.22,
22.57,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.32,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28.2,
"safety_violations": 0,
"battery_end": 85.2,
"path_len_m": 75
} | [
{
"t": 0,
"obs": [
0.7376,
0.5513,
0,
0.7649,
0.6373,
0.6249,
0.8576,
0.5816,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00003 |
fire_assess | Check the roof for fire venting conditions | scripted | 1,004 | {
"structure": {
"cx": 35.01,
"cy": 26.3,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
29.29,
30.54
],
"plume_radius": 6,
"victims": [
[
41.01,
27.03,
5
],
[
37.03,
21.3,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.75,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28.7,
"safety_violations": 0,
"battery_end": 84.9,
"path_len_m": 73.2
} | [
{
"t": 0,
"obs": [
0.7002,
0.526,
0,
0.7198,
0.5857,
0.6108,
0.8202,
0.5405,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00004 |
fire_assess | Assess the structure fire from the east side | scripted | 1,005 | {
"structure": {
"cx": 35.82,
"cy": 34.31,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
30.4,
38.03
],
"plume_radius": 6,
"victims": [
[
41.82,
36.31,
5
],
[
38.68,
29.31,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.68,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 29.4,
"safety_violations": 0,
"battery_end": 84.3,
"path_len_m": 77.5
} | [
{
"t": 0,
"obs": [
0.7164,
0.6861,
0,
0.8362,
0.6079,
0.7606,
0.8364,
0.7263,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00005 |
fire_assess | Evaluate fire conditions on the northwest facade | scripted | 1,006 | {
"structure": {
"cx": 38.37,
"cy": 31.01,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
32.96,
35.53
],
"plume_radius": 6,
"victims": [
[
44.37,
32.81,
5
],
[
37.38,
26.01,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.19,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28.8,
"safety_violations": 0,
"battery_end": 84.6,
"path_len_m": 77.8
} | [
{
"t": 0,
"obs": [
0.7674,
0.6202,
0,
0.8305,
0.6593,
0.7107,
0.8874,
0.6561,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00006 |
fire_assess | Assess the structure fire from the east side | scripted | 1,007 | {
"structure": {
"cx": 35.71,
"cy": 33.16,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
31,
37.77
],
"plume_radius": 6,
"victims": [
[
41.71,
33.69,
5
],
[
38.08,
28.16,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.1,
-... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28.7,
"safety_violations": 0,
"battery_end": 84.8,
"path_len_m": 76.4
} | [
{
"t": 0,
"obs": [
0.7142,
0.6632,
0,
0.8187,
0.62,
0.7553,
0.8342,
0.6739,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00007 |
fire_assess | Determine fire intensity and spread direction | scripted | 1,008 | {
"structure": {
"cx": 36.67,
"cy": 33.13,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
31.11,
36.16
],
"plume_radius": 6,
"victims": [
[
42.67,
36.03,
5
],
[
38.74,
28.13,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.2,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 29.6,
"safety_violations": 0,
"battery_end": 84.3,
"path_len_m": 76.9
} | [
{
"t": 0,
"obs": [
0.7335,
0.6626,
0,
0.8324,
0.6221,
0.7233,
0.8535,
0.7206,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00008 |
fire_assess | Survey east wing for fire progression | scripted | 1,009 | {
"structure": {
"cx": 35.57,
"cy": 28.27,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
29.66,
32.71
],
"plume_radius": 6,
"victims": [
[
41.57,
30.22,
5
],
[
36.22,
23.27,
5
]
],
"gps_denied_radius": 4,
"wind": [
0.12,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 27,
"safety_violations": 0,
"battery_end": 85.7,
"path_len_m": 74.6
} | [
{
"t": 0,
"obs": [
0.7113,
0.5654,
0,
0.7525,
0.5933,
0.6541,
0.8313,
0.6045,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00009 |
fire_assess | Evaluate wind effect on fire spread pattern | scripted | 1,010 | {
"structure": {
"cx": 35.95,
"cy": 30.36,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
31.79,
33.4
],
"plume_radius": 6,
"victims": [
[
41.95,
32.45,
5
],
[
37.43,
25.36,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.18,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28,
"safety_violations": 0,
"battery_end": 85.2,
"path_len_m": 75.8
} | [
{
"t": 0,
"obs": [
0.719,
0.6072,
0,
0.7849,
0.6357,
0.6679,
0.839,
0.649,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00010 |
fire_assess | Assess defensive vs offensive attack feasibility | scripted | 1,011 | {
"structure": {
"cx": 37.84,
"cy": 30.63,
"w": 12,
"d": 10,
"h": 8
},
"fire_xy": [
32.27,
35.47
],
"plume_radius": 6,
"victims": [
[
43.84,
28.25,
5
],
[
37.23,
25.63,
5
]
],
"gps_denied_radius": 4,
"wind": [
-0.16,
... | {
"success": true,
"reason": "assessment band held 8 s",
"time_s": 28.1,
"safety_violations": 0,
"battery_end": 85.3,
"path_len_m": 77.4
} | [
{
"t": 0,
"obs": [
0.7568,
0.6127,
0,
0.8175,
0.6453,
0.7094,
0.8768,
0.565,
0.1667,
1,
0,
0,
1,
1,
0,
1,
0,
1,
0,
0
],
"action": {
"vx": 2,
"vy": 2,
"vz": 1,
... | ep_00011 |
End of preview.
DroneVLA DFR Episodes — Language-Conditioned Aerial Drone-First-Responder Actions
The first open language-conditioned action dataset for aerial drone first response. 330 flight episodes · 18,565 decision ticks · 3 DFR mission categories · natural-language instructions from the Language Table for Drones benchmark · full state–action trajectories with safety and outcome labels.
Companion dataset to DroneVLA (Frontier Tower Research Lab): the open vision–language–action model and benchmark for autonomous drone first response. A July 2026 survey of NVIDIA's Physical AI collection (46 datasets/models; individual sets exceed 150M downloads) found zero aerial datasets — this release begins closing that gap, following the episodes + language-annotations schema of the PhysicalAI-Autonomous-Vehicles (Alpamayo) release, adapted to aerial DFR.
Files
| File | Contents |
|---|---|
episodes.jsonl |
One JSON object per episode (schema below) |
stats.json |
Dataset-level statistics |
flight_gallery.png |
Keyframe gallery — learned policy flying all 3 missions |
Episode schema
{
"episode_id": "ep_00000",
"task": "fire_assess", // fire_assess | victim_search | return_home
"instruction": "Assess the structure fire from the east side",
"policy": "scripted", // scripted teacher | learned (behavior-cloned head)
"world_seed": 1000, // regenerates the exact world
"world": { // scene geometry (ENU metres)
"structure": {"cx", "cy", "w", "d", "h"},
"fire_xy": [x, y], "plume_radius": 6.0,
"victims": [[x, y, z], ...],
"gps_denied_radius": 4.0, "wind": [wx, wy]
},
"outcome": {
"success": true, "reason": "assessment band held 8 s",
"time_s": 28.9, "safety_violations": 0,
"battery_end": 82.1, "path_len_m": 61.4
},
"ticks": [ // 2 Hz decision ticks
{
"t": 0.0,
"obs": [/* 20-D observation: relative geometry, battery, GPS flag,
heading, clock, task one-hot */],
"action": {"vx", "vy", "vz", "yaw_rate", "mission_state"},
"pos": [x, y, z], "yaw": 0.0, "gps": true
}
]
}
Dynamics & safety envelope
Kinematic point-mass integration at 10 Hz (policy at 2 Hz), BSIS-style
velocity limits (2 m/s GPS / 0.5 m/s GPS-denied), structure collision =
crash, smoke-plume cylinder = safety violation, battery model with the
20% return-home rule. Mission success criteria per task are documented in
the DroneVLA repository (dronevla/kinesim.py).
Baselines trained on this data
| Policy | fire_assess | victim_search | return_home | Mean | Crashes |
|---|---|---|---|---|---|
| Scripted teacher (data source) | 100% | 100% | 100% | 100% | 0% |
| Behavior-cloned action head (0.15M params) | 98.0% | 94.0% | 100% | 97.3% | 0% |
| Zero-shot VLM policy (open-loop, n=30) | 0% | 0% | 0% | 0% | 0% |
Closed-loop mission success over 100 unseen worlds per task. The 0% open-loop row is the headline: language-level competence does not survive closed-loop flight without a trained action head.
Intended uses
Behavior cloning and offline RL for aerial control; language-conditioned policy learning; sim benchmark reproduction; schema template for future aerial datasets (a vision release with onboard-camera frames from Isaac Sim is in preparation).
License & citation
Apache-2.0. Please cite:
@misc{dronevla2026dataset,
author = {Annan, Yoofi},
title = {DroneVLA DFR Episodes: Language-Conditioned Aerial
Drone-First-Responder Actions},
year = {2026},
note = {Frontier Tower Research Lab},
}
- Downloads last month
- 52