Miril-DroneVLM-2B-2

Drones can talk.

Enabling drones to reason and talk about their environment.

Ask a drone what it sees. Get an answer software can use.

Miril-DroneVLM-2B-2 is an open-weight, 2B-class aerial vision-language model from Miril.ai. Give it an overhead image and an ordinary English question. It returns one of four machine-readable JSON responses: a whole-scene caption, a factual answer, a landing or delivery location, or a point on one visible target.

A drone camera can collect more imagery than a person can inspect frame by frame. This model explores a practical interface for that stream: ask what is present, what is happening, where an object is, or which part of the frame deserves review, then pass the typed result to an operator interface or separately validated application.

The target applications are civilian: delivery, first response, infrastructure inspection, agriculture, construction, mapping, and disaster recovery. Miril-DroneVLM-2B-2 is an edge-oriented research model for testing how those systems can become easier to supervise and integrate. It is a perception component, not a flight controller.

Try the image and video demo:

Miril-DroneVLM-2B-2-Demo

Read The Paper

Download the 12-page PDF: Miril-DroneVLM-2B-2: Turning Aerial Detector Labels into a Structured Vision-Language Interface

The paper explains the complete engineering path behind this release:

• how reviewed WALDO detector labels, counts, and boxes become grounded image-question-JSON supervision without being passed to the model at inference;
• why ordinary English requests route into caption, answer, location, or pointing, and how downstream software validates and uses each response;
• how the instruction corpus was expanded and hardened around ontology, visible target size, paraphrase diversity, grouped image splits, explicit negative cases, and null-coordinate discipline;
• how supervised adaptation, hard-example mining, and preference optimization were evaluated on matched held-out cases without treating valid JSON as proof of visual grounding;
• what the release benchmarks show about descriptive answers, spatial behavior, spoken questions, CUDA and MLX quantization, and the remaining failure modes.

It also documents the model's safety boundary and the next research steps: direct training on paired spoken questions, larger Gemma variants, stronger spatial grounding, and higher-resolution visual backbones.

Four Questions, Four Usable Results

These are deterministic release-checkpoint MLX 8-bit outputs on attributed Pexels images outside the training corpus. Each panel shows the ordinary user question, the parsed response, and any drawable model coordinate. The examples were manually screened for presentation and are not an accuracy sample; the complete held-out results appear below.

One Question In, One Typed Result Out

The user does not choose a schema or append formatting instructions. The user asks a normal question:

What am I looking at?

How many light vehicles are visible?

Choose a place to lower this parcel.

Track the white car.

The application supplies the fixed system contract from router_contract.py. The model interprets the requested operation and returns exactly one bare JSON object.

The public runtime contract is typed_json_router_v2. Pin that contract file with the model revision when integrating the model into another application.

Requested result	JSON type	Secondary key	Downstream use
General read of the whole image	caption	none	Display or index a scene description
Fact, count, presence, appearance, hazard, or verbal explanation	answer	none	Display or consume a textual answer
Choose one landing or delivery area	location	intent: landing or delivery	Review a selected point or broad direction
Locate, mark, or track one target	pointing	action: locate, point, or track	Draw a point or initialize downstream tracking

This separation is the point of the release. The response identifies its own type, so downstream code can dispatch deterministically instead of trying to classify free-form prose after generation.

Prompting In Plain English

Prompt for the result you need, not for the JSON structure:

What you want	Natural examples	Response
A read of the complete frame	“What am I looking at?”, “Describe the whole scene.”	caption
A fact or explanation	“Any buses?”, “How many people?”, “What should I avoid here?”	answer
A candidate landing or delivery area	“Where could I land?”, “Pick a clear place for this parcel.”	location
One visible target	“Where is the white car?”, “Mark the storage tank.”, “Track that truck.”	pointing

These are intent families, not a keyword list. For example, “Where are the cars concentrated?” asks for a verbal answer, while “Where is the white car?” asks for one point.

caption

Use a general request for the whole frame:

What am I looking at?

{
  "type": "caption",
  "caption": "From above, one built-up structure anchors the upper-left corner of a tidy property, bordered by asphalt, bright grass, driveways, and trimmed greenery."
}

answer

Use a question whose result should be words rather than one image point. This includes counts, object presence, colors, scene type, relative position, risks, suitability, and areas to avoid.

Give the visible tally for utility poles.

{
  "type": "answer",
  "answer": "Six utility poles can be seen around the scene."
}

Where does not automatically mean pointing. “Where are the vehicles concentrated?” asks for a verbal description and routes to answer. “Where is the white car?” asks for one concrete image point and routes to pointing.

location

Use this only when asking the model to choose an area for landing a drone or placing a delivery. A location response may contain a precise point, a broad directional cue, or no point at all. Its status tells the application which interpretation is valid.

Find a usable parcel drop location, if the view contains one.

A broad directional result can look like this:

{
  "type": "location",
  "intent": "delivery",
  "caption": "Coarse grid direction only, not a recommended landing or delivery spot. General safest-looking region: the open bare area on the right side of the road, away from vegetation and the building.",
  "coordinate_system": "gemma_relative_0_1000_yx",
  "point_2d": [500, 750],
  "point_semantics": "representative_point",
  "pointing_mode": "coarse_grid_direction",
  "status": "coarse_direction",
  "x": 750,
  "y": 500
}

That output identifies a direction in the frame. It does not identify an approved parcel drop or landing spot.

pointing

Use this when one concrete visible point is the requested result.

Where in the image is the center of the white-roofed structure?

{
  "type": "pointing",
  "action": "locate",
  "caption": "The white-roofed structure is centered in the lower-right.",
  "coordinate_system": "gemma_relative_0_1000_yx",
  "point_2d": [936, 956],
  "point_semantics": "specific_point",
  "pointing_mode": "precise_point",
  "status": "target_found",
  "x": 956,
  "y": 936
}

The action follows the user’s intent:

User intent	Typical wording	action
Find or look at one target	“Where is…”, “Find…”, “Look at…”	locate
Put a marker on one target	“Point at…”, “Mark…”, “Highlight…”	point
Supply the first point for temporal tracking	“Track…”, “Follow…”, “Lock on…”	track

The model provides only the initial image point for track. Your application performs tracking across later video frames. The included live demo starts from a 60 x 60 region in the processed model frame, maps that region back to source-video pixels, and removes the marker if optical flow fails or any edge of the tracked box touches the frame.

Coordinate Contract

Spatial responses use gemma_relative_0_1000_yx:

• x=0 is the left edge and x=1000 is the right edge.
• y=0 is the top edge and y=1000 is the bottom edge.
• point_2d is [y, x].
• x and y repeat the same values for simpler downstream access.

There are three visual outcomes. They are deliberately different because a broad cue, a grounded point, and an abstention must not look interchangeable in an operator interface.

Precise point

pointing_mode: precise_point is drawable only when the response is schema-valid and uses status: target_found. It represents one image-grounded point. A UI may draw a compact reticle there. For a track request it may also initialize a tracker.

Coarse direction

pointing_mode: coarse_grid_direction is allowed only for location. It uses a 3 x 3 grid where both coordinates are exactly 250, 500, or 750.

Treat it as “look in this part of the frame,” not “land here.” Draw a broad region cue, never a precision reticle. Do not initialize a tracker or send it to a controller as a target.

No point

pointing_mode: none, invalid JSON, an invalid schema, or a non-target status means there is no drawable point. Coordinate fields must be null. If a malformed response combines a non-target status with coordinates, ignore those coordinates.

Interpret status by response type:

Response	Status	Meaning
location	target_found	One image-grounded landing or delivery candidate was selected
location	coarse_direction	Broad 3 x 3 directional context only
location	no_safe_area	No usable area should be marked
location	unknown	The image does not support a reliable judgment
pointing	target_found	One requested target was grounded
pointing	no_target	The requested object is absent
pointing	no_matching_target	Objects may be present, but none satisfy the requested relation
pointing	ambiguous_target	More than one target fits and the request does not identify one
pointing	unknown	The image does not support a reliable point

The included validator enforces these rules before returning a dispatch-safe payload. Applications should still retain the original model text for audit rather than silently repairing it.

Mixed Requests

The fixed router applies this precedence:

1. A request to choose a landing or delivery area routes to location.
2. Otherwise, an explicit request for one concrete point routes to pointing.
3. Otherwise, a factual or verbal question routes to answer.
4. Otherwise, a whole-scene read routes to caption.

If several indistinguishable targets are named, the model should preserve the requested pointing action but return ambiguous_target with null coordinates instead of choosing arbitrarily. If different pointing actions are mixed for one target, track takes priority over point, and point takes priority over locate.

Loader Compatibility

Use transformers>=5.12.1. Gemma 4 E2B declares 35 language layers and 20 shared-KV layers. Layers 15 through 34 reuse key/value states, so a correct export does not contain separate k_proj, v_proj, or k_norm tensors for those shared layers. Release validation checks this layout directly and rejects both genuinely missing owner tensors and redundant shared-layer tensors.

The retained shared-KV export audit records the expected and observed tensor owners for this exact artifact.

Quickstart

Download the small runtime helpers from this repository, install their dependencies, and ask a plain-English question:

hf download MirilAI/Miril-DroneVLM-2B-2 inference.py router_contract.py requirements.txt --local-dir miril-drone-runtime
python -m pip install -r miril-drone-runtime/requirements.txt
python miril-drone-runtime/inference.py \
  --image drone_frame.jpg \
  --prompt "Where is the white car?"

The script:

• sends the exact fixed router prompt as a system message,
• sends the image and ordinary English question as the user message,
• decodes deterministically,
• requires valid bare JSON,
• validates the complete response schema,
• refuses to dispatch or draw malformed coordinates.

For direct integration, import TYPED_JSON_ROUTER_SYSTEM_PROMPT, typed_response_errors, and drawable_point from router_contract.py.

Experimental spoken question

The inherited Gemma audio path can be exercised with the same image and router contract. This checkpoint was fine-tuned with typed questions, so use the published typed-versus-spoken benchmark to judge the modality gap rather than assuming parity. Install ffmpeg, then run:

python miril-drone-runtime/inference.py \
  --image drone_frame.jpg \
  --audio spoken_question.mp3 \
  --audio-transcript "Where is the white car?"

--audio-transcript is optional provenance for the returned audit record. It is never supplied to the model; the user turn contains the image and decoded audio waveform only.

Image And Video Overlays

The same black-side-panel presentation used by the demo is included as runnable code. Labels are bold green, every response is shown beside the question that produced it, and the panel is half the image width.

hf download MirilAI/Miril-DroneVLM-2B-2 \
  inference.py router_contract.py image_overlay.py video_overlay.py requirements.txt \
  --local-dir miril-drone-runtime

Image:

python miril-drone-runtime/image_overlay.py \
  --image drone_frame.jpg \
  --output overlay.jpg \
  --prompt "What am I looking at?" \
  --prompt "Are any people or light vehicles visible?" \
  --prompt "Where is the white car?"

Video, with one inference pass every five seconds:

python miril-drone-runtime/video_overlay.py \
  --input drone_video.mp4 \
  --output drone_video_overlay.mp4 \
  --interval-seconds 5 \
  --crop center-square \
  --crop-size 1000 \
  --prompt "What am I looking at?" \
  --prompt "Track the white car."

The renderer preserves the source frame rate, reveals each answer over 12 frames without changing its font size during the reveal, tracks valid precise points between inference calls, hides failed or edge-adjacent tracks, and never tracks a coarse direction or non-target response.

Evaluation

Release evaluation has three complementary scopes:

1. Complete held-out validation: every V2 release artifact is evaluated on every accepted validation request. This is the primary quantization-regression check for merged BF16, CUDA bnb8, CUDA bnb4, MLX 8-bit, and MLX 4-bit.
2. Independent route benchmark: a separately authored held-out benchmark balances captioning, factual answers, landing selection, delivery selection, object location, visual pointing, and tracker initialization. It supports controlled V1/V2 and cross-runtime comparison without reusing training questions.
3. Cleaned held-out audit: every unchanged release artifact is evaluated after a stricter review removes pointing targets that are too small under the model's actual image-preprocessing contract. Validation and test are complete, and strict-cleaned and coverage-matched views are reported separately.

All scopes check route selection, strict JSON, schema validity, text overlap, status and mode agreement, null-coordinate discipline, coarse-grid agreement, and precise-coordinate recovery. No single aggregate can show all of those behaviors, so the release reports them separately. Publication requires complete prediction coverage and the exact case-file hash for every reported artifact.

Generation comparison

Metric	V1 merged BF16	Merged BF16
Valid JSON	92.0%	100.0%
Schema valid	21.8%	96.1%
Route accuracy	21.8%	94.8%
Caption / answer F1	26.2%	38.2%
Spatial status	2.9%	79.2%
Precise target retained	7.5%	50.2%
Point within 100	3.9%	33.4%
Coarse direction exact	0.0%	35.0%
No-target discipline	43.9%	93.4%

Deployment comparison

Metric	Merged BF16	CUDA bnb8	CUDA bnb4	MLX 8-bit	MLX 4-bit
Valid JSON	100.0%	99.1%	96.7%	100.0%	100.0%
Schema valid	96.1%	93.6%	92.1%	94.6%	55.8%
Route accuracy	94.8%	91.6%	91.2%	93.4%	56.9%
Caption / answer F1	38.2%	19.6%	37.2%	37.3%	35.1%
Spatial status	79.2%	63.4%	69.7%	77.9%	77.3%
Precise target retained	50.2%	42.4%	53.1%	51.2%	46.9%
Point within 100	33.4%	8.0%	29.7%	31.0%	25.1%
Coarse direction exact	35.0%	9.3%	30.2%	32.1%	27.6%
No-target discipline	93.4%	89.6%	87.5%	92.6%	91.6%

Complete held-out validation

Metric	Merged BF16	CUDA bnb8	CUDA bnb4	MLX 8-bit	MLX 4-bit
Valid JSON	99.9%	99.7%	99.0%	100.0%	100.0%
Schema valid	99.9%	99.7%	98.7%	100.0%	53.0%
Route accuracy	99.9%	99.6%	98.6%	100.0%	53.2%
Caption / answer F1	38.6%	23.0%	37.2%	37.8%	35.6%
Spatial status	86.1%	70.1%	79.8%	85.3%	82.8%
Precise target retained	71.7%	60.7%	73.6%	71.2%	68.8%
Point within 100	38.3%	9.2%	34.0%	33.6%	27.8%
Coarse direction exact	34.4%	11.3%	31.2%	33.0%	25.8%
No-target discipline	96.3%	93.3%	92.4%	95.6%	95.5%

Cleaned held-out deployment audit

After training, a stricter held-out audit removed pointing rows whose targets fall below the model-visible size threshold, then ran every release artifact on the complete revised validation and test splits. Strict-cleaned rows use only accepted evidence. Coverage-matched rows add evidence-preserving questions on the same held-out images to restore the earlier route and pointing action/status mix; they do not recreate the earlier object-class histogram.

Final held-out test

Strict-cleaned evidence

Metric	Merged BF16 - Strict cleaned test	CUDA bnb8 - Strict cleaned test	CUDA bnb4 - Strict cleaned test	MLX 8-bit - Strict cleaned test	MLX 4-bit - Strict cleaned test
Valid JSON	99.8%	99.6%	89.3%	100.0%	100.0%
Schema valid	99.8%	99.6%	89.2%	100.0%	62.8%
Route accuracy	99.8%	99.5%	89.1%	100.0%	63.3%
Reference text F1	60.9%	47.4%	53.5%	60.0%	57.2%
Spatial status	88.0%	76.7%	72.0%	86.1%	83.8%
Coordinate quality	89.7%	84.0%	78.9%	88.4%	86.1%
Precise target retained	65.2%	52.6%	60.3%	64.8%	61.3%
Point within 100	42.6%	9.4%	34.3%	39.5%	31.4%
No-target discipline	96.5%	97.1%	85.1%	95.1%	93.4%

Coverage-matched evidence

Metric	Merged BF16 - Coverage-matched test	CUDA bnb8 - Coverage-matched test	CUDA bnb4 - Coverage-matched test	MLX 8-bit - Coverage-matched test	MLX 4-bit - Coverage-matched test
Valid JSON	99.9%	99.6%	89.4%	100.0%	100.0%
Schema valid	99.9%	99.5%	89.3%	100.0%	50.2%
Route accuracy	99.9%	99.5%	89.1%	100.0%	50.5%
Reference text F1	64.3%	50.3%	56.6%	63.4%	60.6%
Spatial status	84.2%	64.1%	70.8%	82.5%	79.0%
Coordinate quality	84.9%	77.6%	74.7%	83.6%	80.9%
Precise target retained	63.5%	49.9%	60.0%	62.8%	59.7%
Point within 100	42.8%	7.8%	34.4%	39.2%	31.4%
No-target discipline	96.4%	97.6%	84.1%	95.2%	93.1%

Validation

Strict-cleaned evidence

Metric	Merged BF16 - Strict cleaned validation	CUDA bnb8 - Strict cleaned validation	CUDA bnb4 - Strict cleaned validation	MLX 8-bit - Strict cleaned validation	MLX 4-bit - Strict cleaned validation
Valid JSON	99.9%	99.6%	96.4%	100.0%	100.0%
Schema valid	99.9%	99.5%	96.1%	100.0%	64.3%
Route accuracy	99.9%	99.5%	96.0%	100.0%	64.6%
Reference text F1	59.9%	48.1%	56.8%	59.0%	56.7%
Spatial status	89.6%	80.1%	80.2%	88.5%	86.3%
Coordinate quality	89.8%	84.9%	85.5%	88.9%	87.4%
Precise target retained	65.8%	55.7%	67.8%	65.8%	62.9%
Point within 100	39.3%	9.6%	34.6%	35.0%	28.6%
No-target discipline	96.7%	97.2%	91.8%	96.1%	95.2%

Coverage-matched evidence

Metric	Merged BF16 - Coverage-matched validation	CUDA bnb8 - Coverage-matched validation	CUDA bnb4 - Coverage-matched validation	MLX 8-bit - Coverage-matched validation	MLX 4-bit - Coverage-matched validation
Valid JSON	99.9%	99.6%	96.3%	100.0%	100.0%
Schema valid	99.9%	99.5%	96.0%	100.0%	53.0%
Route accuracy	99.9%	99.4%	95.9%	100.0%	53.2%
Reference text F1	62.6%	50.3%	59.4%	61.8%	59.5%
Spatial status	84.5%	66.7%	76.6%	83.5%	80.7%
Coordinate quality	84.5%	78.7%	80.1%	83.4%	81.7%
Precise target retained	60.5%	49.7%	62.1%	60.7%	58.8%
Point within 100	38.4%	8.1%	34.0%	33.9%	28.0%
No-target discipline	96.1%	97.4%	90.0%	95.5%	94.1%

Validation supports comparison and model selection; test is the final held-out report. These automated scores measure contract and reference agreement, not flight safety.

Experimental spoken-query input

Metric	Typed question	Spoken question
Valid JSON	99.9%	100.0%
Schema valid	99.9%	98.1%
Route accuracy	99.9%	97.6%
Caption / answer F1	39.3%	39.2%
Spatial status	71.3%	65.3%
Precise target retained	50.5%	39.3%
Point within 100	36.3%	31.9%
Coarse direction exact	34.7%	34.3%
No-target discipline	88.2%	88.2%

The image, ordinary-English wording, accepted JSON target, route, and case identity are matched; only the question modality changes. This checkpoint was trained on typed questions, so spoken input is an experimental zero-shot capability boundary rather than parity-tested speech support. A future generation will train directly on paired image-plus-audio questions.

The independent benchmark rows use identical cases. Higher is better. Reference-overlap and coordinate scores are regression signals, not physical-world safety evidence.

These automated metrics are regression signals, not proof of physical-world safety. Aerial reasoning and pointing still require human review and scenario-specific testing.

Citation

@techreport{sturges2026mirildrone2b2,
  title       = {Miril-DroneVLM-2B-2: Turning Aerial Detector Labels into a Structured Vision-Language Interface},
  author      = {Sturges, Stephan},
  institution = {Miril.ai},
  year        = {2026},
  url         = {https://huggingface.co/MirilAI/Miril-DroneVLM-2B-2/blob/main/technical_report.pdf}
}

GPT-5.5 and GPT-5.6 Sol contributed to the research, engineering, analysis, and editing. Stephan Sturges directed the work and accepts responsibility for the report and release.

Deployment Variants

• Miril-DroneVLM-2B-2-bnb8: 8-bit CUDA deployment.
• Miril-DroneVLM-2B-2-bnb4: 4-bit CUDA deployment.
• Miril-DroneVLM-2B-2-MLX-8bit: 8-bit Apple Silicon deployment.
• Miril-DroneVLM-2B-2-MLX-4bit: compact Apple Silicon deployment with a documented strict-schema routing regression.

Variant cards report artifact size, recommended deployment memory, and complete same-case benchmark deltas.

Treat those measured deltas as authoritative. The CUDA 8-bit and 4-bit packages use different bitsandbytes algorithms, so bit width alone does not rank their multimodal fidelity.

The MLX 4-bit artifact remains available as a measured compact variant, not as an equivalent replacement for MLX 8-bit. On the complete balanced independent benchmark it preserves valid JSON at 100.0% after rounding but drops to 55.8% exact schema validity and 56.9% route accuracy, predominantly because pointing responses use intent where the contract requires action.

The export tensor audit passes. All expected key/value-owner and vision k_norm tensors are present; later language layers intentionally reuse shared key/value states. The leading hypothesis is therefore precision loss from uniform affine 4-bit conversion of token embeddings and linear modules, not a missing tensor. The variant card contains the full evidence and operational warning.

Next Experiments

• Mixed-bit MLX: keep token embeddings, the language head, and selected projections at 8-bit, then require recovery of the action key, exact route, and exact schema on the same benchmark cases.
• Gemma 4 E4B: repeat the four-schema training and evaluation protocol at the larger effective model scale while holding the text corpus, grouped splits, and release gates fixed.
• Paired spoken input: replace a deterministic subset of typed questions with ElevenLabs-generated speech while preserving the image and JSON target. Split-isolated synthesis voices, varied voice settings, local acoustic augmentation, held-out voices, and later human recordings will test whether audio training closes the observed spoken-input gap without degrading typed requests.

WALDO Lineage

Miril-DroneVLM-2B-2 is a separate VLM, but its aerial vocabulary grows from the WALDO30 overhead detector. The two systems are complementary: WALDO supplies fast detection boxes; Miril-DroneVLM-2B-2 supplies language reasoning and typed responses.

During dataset construction, WALDO annotations are treated as authoritative evidence for class presence, counts, and geometry after audit. A broad class name is not treated as proof of a narrower noun: an UPole box may contain a utility pole, antenna, dish, mast, or related fixture, and a Boat box may contain different kinds of watercraft. Narrow language is used only when reviewed image evidence supports it. This preserves the geometric value of the detector labels without asking the model to memorize invented detail.

WALDO class	What the class covers
LightVehicle	Cars, sedans, rickshaws, vans, and similar small personal road vehicles. Trucks and buses remain separate classes.
Person	People and human figures visible from above.
Building	Buildings and roofed structures, including sheds, shacks, pergolas, canopies, sunshades, and tents when supported by the image.
UPole	Utility and telegraph poles, power-line fixtures, lamp posts, street fixtures, satellite dishes, antennas, and thin truss structures.
Boat	Boats and watercraft such as ships, canoes, kayaks, and sailboats; surfboards may also occur in the source ontology.
Bike	Bicycles, motorbikes, mopeds, scooters, and other two-wheeled vehicles.
Container	Storage and shipping containers.
Truck	Trucks, lorries, transport vehicles, semitrailers, articulated trucks, flatbeds, and larger industrial road vehicles.
Gastank	Storage tanks, including liquid-storage and cylindrical tanks.
Digger	Diggers, excavators, tractors when explicitly supported, and other construction vehicles or equipment.
Solarpanels	Individual solar panels and panel arrays.
Bus	Buses and bus-like passenger vehicles.

The VLM receives no hidden detector boxes or object-count metadata during inference. It sees only the image, fixed router system message, and user question.

Model Scope

This checkpoint is derived from google/gemma-4-E2B-it and distributed as merged weights. Its aerial fine-tuning used drone-view images with typed English questions and the typed JSON contract above.

The inherited Gemma architecture also accepts audio. We evaluate image-plus-spoken-question inference separately from matched typed questions and report that result as an experimental zero-shot capability, not as speech-training parity. Native temporal video understanding is not part of this release contract: the video demo samples individual frames and uses OpenCV between model calls.

Limits And Safety

Miril-DroneVLM-2B-2 is a perception and reasoning model. It is not a flight controller, certified detect-and-avoid system, landing certifier, or standalone safety system.

It can miss small people or objects, confuse similar classes, misread scale, return an incorrect answer, select an unsuitable region, or produce invalid JSON. Performance varies with altitude, compression, motion blur, weather, camera angle, and domain shift.

For physical deployment:

• validate the exact camera, altitude range, geography, and operating conditions,
• independently check terrain, obstacles, people, vehicles, trees, wires, weather, and maps,
• keep geofencing and vehicle constraints outside the model,
• reject invalid JSON and non-target coordinates,
• treat coarse directions as review cues only,
• keep a human or separately validated planner in the decision loop,
• log prompts, outputs, validation failures, and downstream actions.

Intended Use

Good research and prototyping fits include:

• searching and summarizing aerial imagery for human review,
• inspection triage for infrastructure, agriculture, construction, and mapping,
• first-response and disaster-recovery situational awareness,
• delivery or landing-area pre-screening with explicit abstention,
• object localization and tracker initialization under application control, and
• typed-output experiments for physical AI.

Do not use the model for direct unsupervised flight control, certified safety decisions, identifying people, or safety-critical operation without independent sensing and validation.

License

Apache License 2.0. See LICENSE and NOTICE.

This model is derived from google/gemma-4-E2B-it. Review the upstream model documentation and applicable terms before deployment.