Loading…

Worked solutions

IB Computer Science HL · Paper 3

Computer Science, Higher Level

Timezone

TZ1TZ2

Session: May 2024
Marks: 30
Duration: 1 hour
Calculator: Not allowed
Questions: 4

Q1 Q2 Q3 Q4

Keep your copy of the paper open alongside. We don't reproduce the questions: each part below is labelled to match the May 2024 TZ2 paper, which your school or the IB can provide.

Practice mode

Hide our working until you've had a go.

Questions

Questions 1–4 · 30 marks

Question 1

4 marks

Computer vision foundations and robotic wheel odometry tracking

Modelling and simulation
Control systems

(a)

2 marks

Computer vision is an interdisciplinary branch of artificial intelligence and computer science focused on developing algorithmic systems that capture, process, and analyze visual data from digital imagery or video feeds.
Its core purpose is to extract high-level symbolic representations, geometric structures, or contextual understanding from the physical environment to guide autonomous machine decision-making and robotic actuation.

Mentor tip

State both the automated ingestion/analysis of visual inputs and the downstream extraction of symbolic meaning for robotic action.

(b)

2 marks

The robotic processor reads signals from optical wheel encoders, counting pulse transitions that represent discrete rotational increments of each wheel axle.
It calculates the total linear distance traveled by multiplying the total number of complete and fractional revolutions by the known physical circumference of the wheel ( $C = π \cdot d$ ).

d_{traveled} = N_{revolutions} \times (π \cdot D_{wheel})

Mentor tip

Detail how encoder pulse counts determine revolution totals, which are multiplied by the wheel circumference to yield linear distance.

Question 2

8 marks

Pose estimation under occlusion and the role of edge computing

Modelling and simulation
Networks

(a)

4 marks

Missing Keypoints and Kinematic Ambiguity: Physical obstructions (such as fallen ceiling joists, concrete blocks, or furniture) conceal key anatomical landmarks like wrists, knees, or ankles. Deep learning models rely on visible joint connections to construct a kinematic skeletal graph; missing keypoints force the algorithm to infer coordinates, frequently producing anatomically impossible limb configurations or false negative detections.
False Feature Grouping in Multi-Person Scenes: When multiple casualties lie in close proximity or overlap behind collapsed material, the pose estimation algorithm struggles to determine which visible limbs correspond to which torso. It may erroneously bind the arm of one person to the torso of another, corrupting situational triage data.
Visual Context Corruption and Image Degradation: Occluding objects frequently introduce harsh shadows, irregular outlines, and visual clutter that obscure body boundaries. If the victim or the robot is in motion, partial occlusion compounds motion blur, causing the feature extractor to lose track of keypoint boundaries.
Escalating Computational Demands: Resolving occluded joints requires resource-intensive probabilistic inference (such as pictorial structures or temporal tracking across consecutive video frames). This drastically increases processing latency, undermining the real-time responsiveness required during time-sensitive emergency operations.

Mentor tip

Cover missing anatomical keypoints, erroneous limb grouping across adjacent victims, visual boundary confusion, and inflated computational burden.

(b)

4 marks

Bandwidth Independence and Network Resilience: Disaster zones frequently suffer from severed fiber lines, collapsed cellular towers, and severe radio attenuation caused by thick reinforced concrete. Edge computing places processing units directly onboard the robot or at a local tactical command post, enabling uninterrupted visual SLAM and pose estimation without depending on remote cloud connectivity.
Ultra-Low Latency for Real-Time Robotic Navigation: Autonomous navigation, obstacle avoidance, and dynamic motor control require closed-loop feedback latencies below tens of milliseconds. Processing high-resolution visual streams locally eliminates transmission round-trip delays to external data centers, preventing catastrophic collisions with unstable debris.

Mentor tip

Explain edge computing's role in bypassing damaged network infrastructure and eliminating transmission latency for real-time robotic control.

Question 3

6 marks

Ethical considerations of fairness and non-discrimination in automated casualty triage

System fundamentals

6 marks

Demographic and Anthropometric Training Bias: Machine learning pose estimation models are typically trained on datasets dominated by healthy adult body shapes in conventional postures. If training sets underrepresent children, elderly individuals, individuals with physical disabilities, or amputees, the robot's detection confidence for these cohorts drops significantly. Consequently, the algorithm may inadvertently overlook vulnerable victims or classify them as inanimate debris, directly violating the ethical principle of non-discrimination.
Triage Prioritization Heuristics and Competing Ethical Standards: In automated rescue operations, programmers or learning models must apply criteria to sequence casualty assistance. If the system optimizes solely for maximum rescue volume (favoring victims who are easiest to reach or lightly trapped), it systematically abandons severely pinned victims who require immediate life support. Conversely, prioritizing by estimated survival probability requires automated diagnostic predictions that deep learning systems cannot reliably make, creating intolerable life-or-death moral compromises.
Algorithmic Opacity and the Loss of Human Accountability: Deep neural networks operate as mathematical black boxes, deriving decision boundaries from millions of parameter weights. When a robot bypasses one crying survivor to attend to another, neither human first responders nor surviving families can audit the precise causal logic behind that prioritization. Removing human ethical discretion from triage decisions erodes public trust, obscures legal liability, and eliminates human empathy from life-critical medical interventions.

Mentor tip

Address demographic dataset bias, conflicting algorithmic triage heuristics (utilitarian volume vs vulnerability), and lack of transparent moral accountability.

Question 4

12 marks

Comprehensive evaluation of visual SLAM efficacy in time-critical industrial emergency operations

Control systems

12 marks

Operational Mandate in GPS-Denied Environments: In catastrophic industrial emergencies such as factory fires, chemical spills, or structural collapses, external Global Positioning System (GPS) microwave signals cannot penetrate dense steel roofs, reinforced concrete walls, or subterranean production basements. Under such severe environmental constraints, visual simultaneous localization and mapping (vSLAM) represents an essential enabling technology, allowing search-and-rescue robots to build spatial occupancy maps and localize their trajectory autonomously without relying on external satellite or radio infrastructure.
Critical Operational Liabilities in Fire and Smoke Scenarios: While visual SLAM performs admirably in structured, well-illuminated laboratories, an active factory fire presents catastrophic failure modes for optical computer vision. Intense airborne soot, billowing smoke clouds, and airborne particulates scatter light, causing immediate optical degradation and obscuring the visual texture necessary for feature detection algorithms (such as ORB or FAST). Rapidly flickering flames and sudden electrical blackouts create violent illumination shifts, invalidating the photometric consistency assumption that visual SLAM algorithms depend on across consecutive frames. Furthermore, as thermal currents and fire cause structural elements to warp or collapse dynamically, the static-world assumption underlying geometric bundle adjustment breaks down. When feature tracking fails, the robot enters a relocalization state; however, in a rapidly burning space where landmarks are actively consumed by flames, relocalizing against past keyframes is nearly impossible, leaving the robot disoriented, stationary, and vulnerable to destruction while human survival windows rapidly close.
Timeframe and Computational Bottlenecks: Emergency search and rescue in fires is intensely time-critical; unconscious casualties face asphyxiation from carbon monoxide within minutes. High-accuracy visual SLAM relies on iterative bundle adjustment and dense point cloud updates that demand substantial embedded computing power. If compute resources on the mobile platform become saturated, frame processing rates drop below the camera capture rate, causing frame drops and catastrophic tracking loss. If the robot moves slowly to preserve SLAM tracking accuracy, it fails the operational requirement to locate casualties before smoke inhalation proves fatal.
System Architecture Synthesis and Balanced Recommendation: Concluding that visual SLAM is entirely ineffective would ignore its indispensable mapping capabilities; however, relying strictly on visual optical cameras in factory fires is an engineering failure. The optimal solution requires a multi-modal sensor fusion architecture. Robots must fuse optical cameras with long-wave infrared (LWIR) thermal cameras that see through thick smoke to register human body heat, solid-state LiDAR sensors that generate accurate geometric point clouds regardless of light or smoke, and tactical-grade inertial measurement units (IMUs) that sustain high-frequency dead reckoning through visual blackouts. By deploying an Extended Kalman Filter (EKF) to dynamically down-weight degraded visual inputs in smoke and up-weight thermal and inertial feeds, the robotic system achieves the speed, spatial precision, and reliability required to find survivors within strict emergency timeframes.

Mentor tip

Evaluate both the necessity of vSLAM in GPS-denied environments and its specific failure modes in fires (smoke, dynamic lighting, moving flames), concluding with a justified multi-sensor fusion recommendation.

Our analysis

Where the marks are

Each part sorted by the topic it mostly tests. Control systems carries the most marks on this paper.

Control systems14 marks · 47%
Q1, Q4
System fundamentals6 marks · 20%
Q3
Modelling and simulation6 marks · 20%
Q1, Q2
Networks4 marks · 13%
Q2

Mentor insights

Traps to watch for

01
Ignoring wheel slippage in encoder calculations · Q1
Pure wheel encoder odometry assumes an ideal condition of rolling without slipping; on slick, wet, or rubble-covered disaster floors, wheel spinning causes catastrophic overestimation of distance traveled unless fused with an IMU.
02
Equating edge computing with passive local storage · Q2
Edge computing is active local execution of demanding algorithms (SLAM optimization, neural network inference) close to the data source to bypass severed network backbones and eliminate latency, not merely storing logs locally.
03
Assuming automated triage algorithms are inherently objective and fair · Q3
Algorithms reflect the biases of their training data and human-coded heuristics; optimizing purely for quick extraction volume can unfairly deprioritize heavily trapped or vulnerable cohorts like children or the elderly.
04
Overestimating optical vSLAM robustness in fire environments · Q4
Optical cameras become blinded in thick smoke or total darkness; high-scoring responses must point out that optical vSLAM must be fused with thermal imaging and LiDAR to operate successfully in burning facilities.

Solutions and tips are GeorgeTheTutor's own; the questions are not reproduced here. Not affiliated with or endorsed by the International Baccalaureate Organization.

Want someone to go through your paper with you?

Our maths mentors took these exams recently. One-to-one, they'll find where your marks are going and fix it before May.

Find your maths mentor More past papers

Worked solutions

IB Computer Science HL · Paper 3

Computer Science, Higher Level

Timezone

TZ1TZ2

Session: May 2024
Marks: 30
Duration: 1 hour
Calculator: Not allowed
Questions: 4

Q1 Q2 Q3 Q4

Keep your copy of the paper open alongside. We don't reproduce the questions: each part below is labelled to match the May 2024 TZ2 paper, which your school or the IB can provide.

Practice mode

Hide our working until you've had a go.

Questions

Questions 1–4 · 30 marks

Question 1

4 marks

Computer vision foundations and robotic wheel odometry tracking

Modelling and simulation
Control systems

(a)

2 marks

Computer vision is an interdisciplinary branch of artificial intelligence and computer science focused on developing algorithmic systems that capture, process, and analyze visual data from digital imagery or video feeds.
Its core purpose is to extract high-level symbolic representations, geometric structures, or contextual understanding from the physical environment to guide autonomous machine decision-making and robotic actuation.

Mentor tip

State both the automated ingestion/analysis of visual inputs and the downstream extraction of symbolic meaning for robotic action.

(b)

2 marks

The robotic processor reads signals from optical wheel encoders, counting pulse transitions that represent discrete rotational increments of each wheel axle.
It calculates the total linear distance traveled by multiplying the total number of complete and fractional revolutions by the known physical circumference of the wheel ( $C = π \cdot d$ ).

d_{traveled} = N_{revolutions} \times (π \cdot D_{wheel})

Mentor tip

Detail how encoder pulse counts determine revolution totals, which are multiplied by the wheel circumference to yield linear distance.

Question 2

8 marks

Pose estimation under occlusion and the role of edge computing

Modelling and simulation
Networks

(a)

4 marks

Missing Keypoints and Kinematic Ambiguity: Physical obstructions (such as fallen ceiling joists, concrete blocks, or furniture) conceal key anatomical landmarks like wrists, knees, or ankles. Deep learning models rely on visible joint connections to construct a kinematic skeletal graph; missing keypoints force the algorithm to infer coordinates, frequently producing anatomically impossible limb configurations or false negative detections.
False Feature Grouping in Multi-Person Scenes: When multiple casualties lie in close proximity or overlap behind collapsed material, the pose estimation algorithm struggles to determine which visible limbs correspond to which torso. It may erroneously bind the arm of one person to the torso of another, corrupting situational triage data.
Visual Context Corruption and Image Degradation: Occluding objects frequently introduce harsh shadows, irregular outlines, and visual clutter that obscure body boundaries. If the victim or the robot is in motion, partial occlusion compounds motion blur, causing the feature extractor to lose track of keypoint boundaries.
Escalating Computational Demands: Resolving occluded joints requires resource-intensive probabilistic inference (such as pictorial structures or temporal tracking across consecutive video frames). This drastically increases processing latency, undermining the real-time responsiveness required during time-sensitive emergency operations.

Mentor tip

Cover missing anatomical keypoints, erroneous limb grouping across adjacent victims, visual boundary confusion, and inflated computational burden.

(b)

4 marks

Bandwidth Independence and Network Resilience: Disaster zones frequently suffer from severed fiber lines, collapsed cellular towers, and severe radio attenuation caused by thick reinforced concrete. Edge computing places processing units directly onboard the robot or at a local tactical command post, enabling uninterrupted visual SLAM and pose estimation without depending on remote cloud connectivity.
Ultra-Low Latency for Real-Time Robotic Navigation: Autonomous navigation, obstacle avoidance, and dynamic motor control require closed-loop feedback latencies below tens of milliseconds. Processing high-resolution visual streams locally eliminates transmission round-trip delays to external data centers, preventing catastrophic collisions with unstable debris.

Mentor tip

Explain edge computing's role in bypassing damaged network infrastructure and eliminating transmission latency for real-time robotic control.

Question 3

6 marks

Ethical considerations of fairness and non-discrimination in automated casualty triage

System fundamentals

6 marks

Demographic and Anthropometric Training Bias: Machine learning pose estimation models are typically trained on datasets dominated by healthy adult body shapes in conventional postures. If training sets underrepresent children, elderly individuals, individuals with physical disabilities, or amputees, the robot's detection confidence for these cohorts drops significantly. Consequently, the algorithm may inadvertently overlook vulnerable victims or classify them as inanimate debris, directly violating the ethical principle of non-discrimination.
Triage Prioritization Heuristics and Competing Ethical Standards: In automated rescue operations, programmers or learning models must apply criteria to sequence casualty assistance. If the system optimizes solely for maximum rescue volume (favoring victims who are easiest to reach or lightly trapped), it systematically abandons severely pinned victims who require immediate life support. Conversely, prioritizing by estimated survival probability requires automated diagnostic predictions that deep learning systems cannot reliably make, creating intolerable life-or-death moral compromises.
Algorithmic Opacity and the Loss of Human Accountability: Deep neural networks operate as mathematical black boxes, deriving decision boundaries from millions of parameter weights. When a robot bypasses one crying survivor to attend to another, neither human first responders nor surviving families can audit the precise causal logic behind that prioritization. Removing human ethical discretion from triage decisions erodes public trust, obscures legal liability, and eliminates human empathy from life-critical medical interventions.

Mentor tip

Address demographic dataset bias, conflicting algorithmic triage heuristics (utilitarian volume vs vulnerability), and lack of transparent moral accountability.

Question 4

12 marks

Comprehensive evaluation of visual SLAM efficacy in time-critical industrial emergency operations

Control systems

12 marks

Operational Mandate in GPS-Denied Environments: In catastrophic industrial emergencies such as factory fires, chemical spills, or structural collapses, external Global Positioning System (GPS) microwave signals cannot penetrate dense steel roofs, reinforced concrete walls, or subterranean production basements. Under such severe environmental constraints, visual simultaneous localization and mapping (vSLAM) represents an essential enabling technology, allowing search-and-rescue robots to build spatial occupancy maps and localize their trajectory autonomously without relying on external satellite or radio infrastructure.
Critical Operational Liabilities in Fire and Smoke Scenarios: While visual SLAM performs admirably in structured, well-illuminated laboratories, an active factory fire presents catastrophic failure modes for optical computer vision. Intense airborne soot, billowing smoke clouds, and airborne particulates scatter light, causing immediate optical degradation and obscuring the visual texture necessary for feature detection algorithms (such as ORB or FAST). Rapidly flickering flames and sudden electrical blackouts create violent illumination shifts, invalidating the photometric consistency assumption that visual SLAM algorithms depend on across consecutive frames. Furthermore, as thermal currents and fire cause structural elements to warp or collapse dynamically, the static-world assumption underlying geometric bundle adjustment breaks down. When feature tracking fails, the robot enters a relocalization state; however, in a rapidly burning space where landmarks are actively consumed by flames, relocalizing against past keyframes is nearly impossible, leaving the robot disoriented, stationary, and vulnerable to destruction while human survival windows rapidly close.
Timeframe and Computational Bottlenecks: Emergency search and rescue in fires is intensely time-critical; unconscious casualties face asphyxiation from carbon monoxide within minutes. High-accuracy visual SLAM relies on iterative bundle adjustment and dense point cloud updates that demand substantial embedded computing power. If compute resources on the mobile platform become saturated, frame processing rates drop below the camera capture rate, causing frame drops and catastrophic tracking loss. If the robot moves slowly to preserve SLAM tracking accuracy, it fails the operational requirement to locate casualties before smoke inhalation proves fatal.
System Architecture Synthesis and Balanced Recommendation: Concluding that visual SLAM is entirely ineffective would ignore its indispensable mapping capabilities; however, relying strictly on visual optical cameras in factory fires is an engineering failure. The optimal solution requires a multi-modal sensor fusion architecture. Robots must fuse optical cameras with long-wave infrared (LWIR) thermal cameras that see through thick smoke to register human body heat, solid-state LiDAR sensors that generate accurate geometric point clouds regardless of light or smoke, and tactical-grade inertial measurement units (IMUs) that sustain high-frequency dead reckoning through visual blackouts. By deploying an Extended Kalman Filter (EKF) to dynamically down-weight degraded visual inputs in smoke and up-weight thermal and inertial feeds, the robotic system achieves the speed, spatial precision, and reliability required to find survivors within strict emergency timeframes.

Mentor tip

Our analysis

Where the marks are

Each part sorted by the topic it mostly tests. Control systems carries the most marks on this paper.

Control systems14 marks · 47%
Q1, Q4
System fundamentals6 marks · 20%
Q3
Modelling and simulation6 marks · 20%
Q1, Q2
Networks4 marks · 13%
Q2

Mentor insights

Traps to watch for

01
Ignoring wheel slippage in encoder calculations · Q1
Pure wheel encoder odometry assumes an ideal condition of rolling without slipping; on slick, wet, or rubble-covered disaster floors, wheel spinning causes catastrophic overestimation of distance traveled unless fused with an IMU.
02
Equating edge computing with passive local storage · Q2
Edge computing is active local execution of demanding algorithms (SLAM optimization, neural network inference) close to the data source to bypass severed network backbones and eliminate latency, not merely storing logs locally.
03
Assuming automated triage algorithms are inherently objective and fair · Q3
Algorithms reflect the biases of their training data and human-coded heuristics; optimizing purely for quick extraction volume can unfairly deprioritize heavily trapped or vulnerable cohorts like children or the elderly.
04
Overestimating optical vSLAM robustness in fire environments · Q4
Optical cameras become blinded in thick smoke or total darkness; high-scoring responses must point out that optical vSLAM must be fused with thermal imaging and LiDAR to operate successfully in burning facilities.

Solutions and tips are GeorgeTheTutor's own; the questions are not reproduced here. Not affiliated with or endorsed by the International Baccalaureate Organization.

Want someone to go through your paper with you?

Our maths mentors took these exams recently. One-to-one, they'll find where your marks are going and fix it before May.

Find your maths mentor More past papers