AirCopBench

Prompt Design

Our prompt design follows a hierarchical structure with main instructions and specialized templates for different question types. The main instruction prompt Fig. 6 establishes the fundamental framework, defining the role as an expert teacher of multi-view perception and specifying the core requirements for multiple-choice question generation, including JSON output format and answer structure.

MAIN INSTRUCTIONS

Role: You are an expert teacher of the ”Multi-view Perception” course, tasked with creating high-quality multiple-choice
questions that test students’ understanding of multi-UAV collaboration.

Goal: To generate multiple-choice questions about collaborative decision-making, object understanding, perception
assessment, and scene understanding from multi-UAV visual content, adhering to specific requirements.

Content Restrictions: Each question must be strictly restricted to and strongly related to the provided visual content
and annotation data.

Question Structure:

• Each question needs the question itself and 4 choices (A, B, C, D)
• There must be only 1 CORRECT answer and 3 wrong answers
• Plausible but Incorrect: The incorrect choices should be reasonable but factually wrong
• The wrong answers should not be too irrelevant
• Answer Placement: The correct answer can be placed at any position among the choices (A, B, C, D)
• Output Format: Must be valid JSON format

Collaborative Decision

When to Collaborate
TASK EXPLANATION: This type of question requires the student to judge when collaboration between multiple
UAVs is necessary based on the current scene analysis. The proper answer should identify situations where information
is incomplete, targets are occluded, or environmental factors require multi-UAV coordination.

TEMPLATE Question: ”When should [UAV ID] initiate collaboration with other UAVs based on the current scene
analysis?”

TEMPLATE Choices: ”When [specific collaboration trigger condition]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ understanding of when collaboration
between multiple UAVs (up to 3) is necessary.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze annotation →determine collaboration need →formulate question →
create options →verify correctness
3. Questions must be based on annotation data
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

THINKING PROCESS:

1. Analyze the annotation to determine if collaboration is needed
2. Formulate a clear question about the need for collaboration
3. Create 4 distinct options where only one is correct
4. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_when2col_UAV1_1001",
"question_type": "4.1 When to Collaborate (UAV1)",
"question": "When should UAV1 initiate collaboration with
other UAVs based on the current scene analysis?",
"options": {
"A": "When target objects are partially occluded and
require multi-viewpoint verification",
"B": "When the scene is completely clear and all
targets are visible",
"C": "When there are no moving objects in the field
of view",
"D": "When the weather conditions are optimal for
single UAV operation"
},
"correct_answer": "A",
"image_description": "UAV1 shows a drone partially
occluded by a tree, requiring collaboration for
complete target verification"
}

Collaborative Decision

What to Collaborate
TASK EXPLANATION: This type of question requires the student to identify what specific object information should
be shared between multiple UAVs. The proper answer should focus on specific object descriptions with intuitive location
and context details, prioritizing drone detection and tracking as primary targets.

TEMPLATE Question: ”What specific object information should [UAV ID] share with other UAVs about the [tar-
get type] in the marked region?”

TEMPLATE Choices: ”[Target type] [specific condition] in the [location] that needs [information type]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ understanding of what specific object
information should be shared between multiple UAVs (up to 3).”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze images →identify specific object information gaps →formulate
question →create options →verify correctness
3. Questions must be based on actual visual content or provided descriptions
4. Each question should have exactly 4 options (A, B, C, D) with at least one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only
7. Focus on specific object descriptions with intuitive location and context details
8. Use intuitive image locations (upper-left corner, center, near landmarks, etc.) instead of numerical positions
9. Prioritize drone detection and tracking as the primary target in all questions

THINKING PROCESS:

1. Analyze all images to identify specific object information gaps in marked regions across multiple UAV views
2. Focus on drone-related object information as the primary target
3. Identify the focus based on generation index
4. Formulate a clear question about what specific object information to share from marked regions
5. Create 4 distinct options, all related to specific object descriptions with intuitive location and context
6. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_what2col_UAV1_1001",
"question_type": "4.2 What to Collaborate (UAV1)",
"question": "What specific object information should
UAV1 share with other UAVs about the drone in the
marked region?",
"options": {
"A": "Drone occluded by the tree in the upper-left
corner of the image that needs position clarification",
"B": "Drone flying at low altitude near the bottom
edge that requires height verification",
"C": "Drone moving rapidly from left to right across
the center that needs trajectory prediction",
"D": "Drone with similar color to background near the
traffic light that requires contrast enhancement"
},
"correct_answer": "A",
"image_description": "UAV1 shows a drone in the marked
region at position (31.5%, 48.1%) with size 6.3%×3.6%
that is occluded by a tree in the upper-left corner,
requiring detailed position and movement information."
}

Collaborative Decision

Which to Collaborate
TASK EXPLANATION: This type of question requires the student to determine which UAV(s) should be the optimal
collaboration partner in a multi-UAV setup. The proper answer should consider complementary visibility conditions and
the relative strengths or specific needs of the current scenario.

TEMPLATE Question: ”Which UAV should [UAV ID] collaborate with to [specific collaboration goal]?”

TEMPLATE Choices: ”[UAV ID] which [specific advantage or capability]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ understanding of which UAV(s) should
be the collaboration partner in a multi-UAV system (up to 3 UAVs).”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze annotation →determine collaboration partner →formulate question
→create options →verify correctness
3. Questions must be based on annotation data
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

THINKING PROCESS:

1. Analyze the annotation to identify the collaboration partners
2. Formulate a clear question about the collaboration partner
3. Create 4 distinct options where only one is correct
4. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_which2col_UAV1_1001",
"question_type": "4.3 Which to Collaborate (UAV1)",
"question": "Which UAV should UAV1 collaborate with to
get a better viewing angle of the partially occluded
target in the central area?",
"options": {
"A": "UAV2, which has a clear view of the central area
from its positioning",
"B": "UAV3, which is located at a similar angle with
the same viewing obstruction",
"C": "No collaboration needed as the target is fully
visible",
"D": "All UAVs simultaneously for maximum coverage"
},
"correct_answer": "A",
"image_description": "UAV1 has partially occluded view
of central area target, UAV2 has better positioning
with clear view of target"
}

Collaborative Decision

Why to Collaborate
TASK EXPLANATION: This type of question requires the student to analyze the fundamental reasons and motiva-
tions for collaboration between multiple UAVs. The proper answer should explain the specific benefits of collaboration
decisions and evaluate the specific benefits brought by collaboration.

TEMPLATE Question: ”Why is collaboration necessary between [UAV ID] and other UAVs in this scenario?”

TEMPLATE Choices: ”To [specific collaboration benefit or reason]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ understanding of which UAV(s) should
be the collaboration partner in a multi-UAV system (up to 3 UAVs).”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze annotation →determine collaboration rationale →formulate question
→create options →verify correctness
3. Questions must be based on annotation data
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

THINKING PROCESS:

1. Analyze the annotation to identify the collaboration partner
2. Formulate a clear question about the collaboration partner
3. Create 4 distinct options where only one is correct
4. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_why2col_UAV1_1001",
"question_type": "4.4 Why to Collaborate (UAV1)",
"question": "Why is collaboration necessary between
UAV1 and other UAVs in this scenario?",
"options": {
"A": "To overcome visual occlusion caused by
environmental obstacles and improve target detection
accuracy",
"B": "To reduce battery consumption by distributing
the workload",
"C": "To increase flight speed and cover more ground
area",
"D": "To test communication systems between UAVs"
},
"correct_answer": "A",
"image_description": "UAV1 encounters visual occlusion
of key targets due to environmental objects, requiring
collaborative input from other UAVs to maintain
complete situational awareness"
}

Object Understanding

Object Recognition
TASK EXPLANATION: This type of question requires the student to identify targets from UAV perspectives, focusing
specifically on drone detection, vehicle recognition, and pedestrian identification. The proper answer should emphasize
the UAV perspective and aerial view characteristics.

TEMPLATE Question: ”From the UAV’s aerial perspective, what type of target is [specific characteristic] in this
scene?”

TEMPLATE Choices: ”[Specific target description] from [UAV perspective characteristic]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”UAV Multi-view Perception”
course. Your role is to create high-quality multiple-choice questions that test students’ ability to identify targets from
UAV perspectives, focusing specifically on drone detection, vehicle recognition, and pedestrian identification.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Focus on UAV-specific target perception: drones, vehicles, pedestrians
3. Questions must emphasize the UAV perspective and aerial view characteristics
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

THINKING PROCESS:

1. First, describe the key targets visible from the UAV perspective
2. Focus on UAV-specific target types: drones, vehicles, pedestrians
3. Formulate a clear, specific question about target recognition from aerial view
4. Create 4 distinct options where only one is correct
5. Verify the question emphasizes UAV perspective and target perception

EXAMPLE OUTPUT:

{
"question_id": "sim3_OR_UAV1_1001",
"question_type": "2.1 Object Recognition (UAV1)",
"question": "From the UAV’s aerial perspective, what type of
target is most prominently visible in this scene?",
"options": {
"A": "A white delivery van",
"B": "A surveillance drone",
"C": "A pedestrian crossing the road",
"D": "A stationary traffic light"
},
"correct_answer": "A",
"image_description": "The UAV captures a white delivery van
from above, clearly visible on the multi-lane road with
other vehicles nearby."
}

Object Understanding

Object Counting
TASK EXPLANATION: This type of question requires the student to count specific target types in scenes from UAV
perspectives. The proper answer should be based on annotation data and ensure counting accuracy for drones, vehicles,
pedestrians, and bicycles.

TEMPLATE Question: ”From the UAV’s aerial perspective, how many [target type] can be detected in [UAV ID]’s
field of view?”

TEMPLATE Choices: ”[Number] [target type]”

CORE PROMPT STRUCTURE:

[Rule-Based] Generate UAV target counting questions based on
all_samples.json annotation data.
Now generates even if count=0.

EXAMPLE OUTPUT:

{
"question_id": "sim3_OC_UAV1_1001",
"question_type": "2.2 UAV Target Counting (UAV1)",
"question": "From the UAV’s aerial perspective, how many
targets (drones, vehicles, pedestrians) can be detected
in UAV1’s field of view?",
"options": {
"A": "3",
"B": "4",
"C": "5",
"D": "6"
},
"correct_answer": "A",
"source": "Rule-Based from all_samples.json"
}

Object Understanding

Object Grounding
TASK EXPLANATION: This type of question requires the student to understand spatial positions of targets in scenes
from UAV perspectives. The proper answer should analyze relative positional relationships between targets and evaluate
spatial perception capabilities.

TEMPLATE Question: ”Where is the [target type] located relative to [other objects] in [UAV ID]’s field of view?”

TEMPLATE Choices: ”[Target type] [spatial relationship] [reference objects]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”UAV Multi-view Perception”
course. Your role is to create high-quality multiple-choice questions that test students’ understanding of target spatial
positioning from UAV aerial perspectives.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Focus on UAV-specific target positioning: drones, vehicles, pedestrians from aerial view
3. Questions must emphasize the UAV’s spatial perception capabilities
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

THINKING PROCESS:

1. First, describe the key targets and their spatial positions from the UAV perspective
2. Focus on UAV-specific target types: drones, vehicles, pedestrians
3. Formulate a clear, specific question about target grounding from aerial view
4. Create 4 distinct options where only one is correct
5. Verify the question emphasizes UAV spatial perception

EXAMPLE OUTPUT:

{
"question_id": "sim3_OG_UAV1_1001",
"question_type": "2.3 Object Grounding (UAV1)",
"question": "Where is the drone located relative to other
objects in UAV1’s field of view?",
"options": {
"A": "Above the intersection, hovering near the traffic
light",
"B": "Behind the building, partially obscured from view",
"C": "On the ground near the sidewalk",
"D": "Inside the vehicle on the road"
},
"correct_answer": "A",
"image_description": "The drone is positioned above the
intersection, hovering near the traffic light structure."
}

Object Understanding

Object Matching
TASK EXPLANATION: This type of question requires the student to match identical targets across multi-UAV per-
spectives, analyzing the impact of viewpoint changes on target appearance. The proper answer should focus on appear-
ance differences caused by viewpoint changes rather than simple recognition.

TEMPLATE Question: ”The [target description] seen from [perspective1] in [UAV1]’s view appears as what in
[UAV2]’s perspective?”

TEMPLATE Choices: ”[Target description] seen from [perspective2] with [specific changes]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”UAV Multi-view Perception”
course. Your role is to create high-quality multiple-choice questions that test students’ ability to match targets across
multiple UAV perspectives, focusing on drone detection, vehicle tracking, and pedestrian identification.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Focus on UAV-specific target matching: drones, vehicles, pedestrians across aerial views
3. Questions must emphasize the UAV’s multi-perspective target tracking capabilities
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

A simple question like ”What is the white truck in image 1?” with the answer ”The white truck in image 2” is USELESS.
Avoid this.
Instead, follow this reasoning process to create a high-quality question:

THINKING PROCESS:

1. Identify a Candidate Target: In the first image (uav id), find a distinct target (drone, vehicle, pedestrian) that is also
clearly visible in one of the subsequent images (other UAVs). Let’s call this the ”target object”.
2. Analyze the Change: Critically compare the target object’s appearance and context between the UAV views. Focus
on what has CHANGED. Examples of changes include:

• Perspective: ”The vehicle seen from the side” in [uav id] is now ”seen from the rear” in another UAV view.
• Relative Position: ”The car behind the bus” in [uav id] is now ”the car beside a red sedan” in another UAV view.
• Action/State: ”The pedestrian walking towards the crosswalk” in [uav id] is now ”the pedestrian waiting at the
crosswalk” in another UAV view.
• Occlusion: ”The partially occluded blue car” in [uav id] is now ”fully visible” in another UAV view.

3. Formulate Question: Ask about the target object’s appearance or context in the first image, with the answer being
how it appears in the second image.
4. Create Options: Make all 4 options plausible descriptions of the target object in the second image, with only one
being correct.
5. Verify: Ensure the question tests understanding of perspective changes, not just object recognition.

EXAMPLE OUTPUT:

{
"question_id": "sim3_OM_UAV1_1001",
"question_type": "2.4 Object Matching (UAV1)",
"question": "The red car seen from the side in UAV1’s view
appears as what in UAV2’s perspective?",
"options": {
"A": "A red car seen from the rear with visible
taillights",
"B": "A blue car seen from the front",
"C": "A red car seen from above with roof visible",
"D": "A red car seen from the opposite side"
},
"correct_answer": "A",
"image_description": "UAV1 shows a red car from the side,
while UAV2 shows the same car from the rear with visible
taillights."
}

Perception Assessment

Quality Assessment
TASK EXPLANATION: This type of question requires the student to assess image quality for perception tasks in
multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection. The proper answer should evaluate
factors such as clarity, noise, and distortion that affect target detection.

TEMPLATE Question: ”How would you rate the [quality factor] for detecting [target types] in this scene?”

TEMPLATE Choices: ”[Quality level] with [specific characteristics]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ ability to assess image quality for per-
ception tasks in multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze →identify quality factors →formulate question →create options →
verify correctness
3. Questions must be based on actual visual content or provided description
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only
7. Focus on quality factors that affect detection of drones, vehicles, pedestrians, and bicycles

THINKING PROCESS:

1. First, describe the quality factors (clarity, noise, color balance, etc.) in the image or description
2. Identify the focus based on generation index
3. Formulate a clear, specific question about image quality for target detection
4. Create 4 distinct options where only one is correct
5. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_QA_UAV1_1001",
"question_type": "3.1 Quality Assessment (UAV1)",
"question": "How would you rate the image clarity for
detecting drones and vehicles in this scene?",
"options": {
"A": "Excellent with sharp details on all targets",
"B": "Good with minor blur on some objects",
"C": "Fair with noticeable distortion affecting
detection",
"D": "Poor with significant artifacts obscuring targets"
},
"correct_answer": "A",
"image_description": "The image shows excellent clarity
with sharp details on drones, vehicles, pedestrians,
and bicycles."
}

Perception Assessment

Usability Assessment
TASK EXPLANATION: This type of question requires the student to assess image usability for perception tasks in
multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection and tracking. The proper answer
should evaluate whether images are suitable for specific tasks and consider matching between task requirements and
image characteristics.

TEMPLATE Question: ”Is the image captured by [UAV ID] usable for [specific task]?”

TEMPLATE Choices: ”[Usability level] for [specific reason]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ ability to assess image usability for
perception tasks in multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection and tracking.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze →identify usability factors →formulate question →create options
→verify correctness
3. Questions must be based on actual visual content or provided description
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only
7. Focus on usability factors that affect detection and tracking of drones, vehicles, pedestrians, and bicycles

THINKING PROCESS:

1. First, describe the usability factors (suitability for target detection, tracking, etc.) in the image or description
2. Identify the focus based on generation index
3. Formulate a clear, specific question about image usability for target tasks
4. Create 4 distinct options where only one is correct
5. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_UA_UAV1_1001",
"question_type": "3.2 Usability Assessment (UAV1)",
"question": "Is the image captured by UAV1 usable for
detecting drones, vehicles, pedestrians, and bicycles?",
"options": {
"A": "Yes, highly usable",
"B": "Yes, usable",
"C": "Yes, partially usable",
"D": "No, not usable"
},
"correct_answer": "A",
"source": "Rule-Based from JSON"
}

Perception Assessment

Causal Assessment
TASK EXPLANATION: This type of question requires the student to analyze causes of perception quality issues in
multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection. The proper answer should identify
key factors affecting perception effectiveness and understand impact of causal relationships on perception quality.

TEMPLATE Question: ”What is the primary cause of [perception issue] in [UAV ID]’s image?”

TEMPLATE Choices: ”[Specific cause] [affecting factor]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ ability to analyze causes of perception
quality issues in multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze image →identify potential causes →formulate question →create
options →verify correctness
3. Questions must be based on actual visual content or provided description
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only
7. Focus on causes that affect detection of drones, vehicles, pedestrians, and bicycles

THINKING PROCESS:

1. First, describe the potential causes of perception issues (occlusion, lighting, resolution, etc.) in the image or description
2. Identify the primary cause based on generation index
3. Formulate a clear, specific question about the cause of perception issues for target detection
4. Create 4 distinct options where only one is correct
5. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_CA_UAV1_1001",
"question_type": "3.3 Causal Assessment (UAV1)",
"question": "What is the primary cause of reduced visibility for drones and vehicles in this scene?",
"options": {
"A": "Heavy fog obscuring distant objects",
"B": "Bright sunlight causing glare on metallic surfaces",
"C": "Low resolution making small objects indistinct",
"D": "Camera angle limiting field of view"
},
"correct_answer": "A",
"image_description": "The image shows heavy fog reducing visibility for drones, vehicles, pedestrians, and bicycles."
}

Perception Assessment

Improvement Assessment
TASK EXPLANATION: This type of question requires the student to suggest improvements for perception quality in
multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection. The proper answer should propose
practical methods to enhance perception effectiveness and understand how to mitigate identified issues.

TEMPLATE Question: ”How can the perception quality for detecting [target types] be improved in [UAV ID]’s
image?”

TEMPLATE Choices: ”By [specific improvement method] to [benefit]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”Multi-view Perception” course.
Your role is to create high-quality multiple-choice questions that test students’ ability to suggest improvements for
perception quality in multi-UAV views, with focus on drone, vehicle, pedestrian, and bicycle detection.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Follow a structured thinking process: analyze issues →identify improvement methods →formulate question →create
options →verify correctness
3. Questions must be based on actual visual content or provided description
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only
7. Focus on improvement methods that enhance detection of drones, vehicles, pedestrians, and bicycles

THINKING PROCESS:

1. First, identify the perception issues (poor lighting, occlusion, low resolution, etc.) in the image or description
2. Suggest practical improvement methods
3. Formulate a clear, specific question about improving perception for target detection
4. Create 4 distinct options where only one is correct
5. Verify the question is unambiguous and answerable

EXAMPLE OUTPUT:

{
"question_id": "sim3_IA_UAV1_1001",
"question_type": "3.4 Improvement Assessment (UAV1)",
"question": "How can the perception quality for detecting drones and vehicles be improved in this scene?",
"options": {
"A": "By adjusting the UAV altitude to reduce occlusion",
"B": "By changing the color filter to enhance contrast",
"C": "By increasing the frame rate for better motion capture",
"D": "By adding more UAVs for multi-angle views"
},
"correct_answer": "A",
"image_description": "The image shows occlusion issues that can be improved by adjusting UAV altitude."
}

Scene Understanding

Scene Description
TASK EXPLANATION: This type of question requires the student to describe overall scenes from multi-UAV perspec-
tives, integrating information from multiple views. The proper answer should provide comprehensive scene descriptions
that capture key elements and dynamics.

TEMPLATE Question: ”What is the overall scene description integrating views from all UAVs?”

TEMPLATE Choices: ”[Comprehensive scene description]”

CORE PROMPT STRUCTURE: system prompt = ”You are an expert teacher of the ”UAV Multi-view Perception”
course. Your role is to create high-quality multiple-choice questions that test students’ ability to describe overall scenes
from multi-UAV perspectives, integrating information from multiple views.”

CRITICAL RULES:

1. ALWAYS respond in English only
2. Focus on integrating multi-UAV views for comprehensive scene understanding
3. Questions must emphasize key scene elements: environment, targets, dynamics
4. Each question should have exactly 4 options (A, B, C, D) with only one correct answer
5. Options should be plausible, distinct in meaning, and avoid minor rephrasing
6. Output must be valid JSON format only

THINKING PROCESS:

1. First, integrate descriptions from all UAV views
2. Formulate a clear question about overall scene description
3. Create 4 distinct options where only one is correct
4. Verify the question captures multi-view integration

EXAMPLE OUTPUT:

{
"question_id": "sim3_SD_1001",
"question_type": "4. Scene Description",
"question": "What is the overall scene description integrating views from all UAVs?",
"options": {
"A": "Busy urban intersection with multiple vehicles, pedestrians, and a hovering drone",
"B": "Quiet rural road with few vehicles and no pedestrians",
"C": "Industrial area with heavy machinery and workers",
"D": "Park setting with people walking and bicycles"
},
"correct_answer": "A",
"image_description": "Integrated view shows a busy urban intersection with vehicles, pedestrians, and a drone."
}