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 TZ1 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

Dead reckoning navigation and loop closure challenges in visual SLAM

Control systems
Modelling and simulation

(a)

2 marks

Dead reckoning is a kinematic navigation technique where a mobile robot continuously estimates its current coordinates and heading by integrating incremental motion measurements forward from an established initial reference point.
The processor samples internal sensors, such as rotary wheel encoders, accelerometers, and rate gyroscopes, converting wheel rotations and angular velocities over discrete time intervals into estimated linear displacement and orientation changes.

Δ s = r \cdot Δ θ, x_{t} = x_{t - 1} + Δ s cos (ϕ_{t})

Mentor tip

Explain both the principle (advancing from a known baseline) and the sensor mechanism (integrating wheel rotation or inertial measurements over time).

(b)

2 marks

Reliable loop closure is difficult because perceptual aliasing frequently occurs in disaster zones, where structurally uniform corridors, identical rubble piles, or repetitive brick patterns cause the vision system to match current features against the wrong historical keyframe.
Furthermore, severe lighting variations, smoke occlusions, and accumulated rotational drift warp visual feature descriptors, preventing the matching algorithm from recognizing an identical physical site when revisited from an alternate angle.

Mentor tip

Focus on perceptual aliasing (different locations looking identical) or feature degradation caused by changing illumination and drift.

Question 2

8 marks

Bundle adjustment optimization and ethical deployment of rescue robotics

Modelling and simulation
System fundamentals

(a)

4 marks

Bundle adjustment performs joint non-linear optimization across an entire visual SLAM graph by refining both estimated three-dimensional feature coordinates and historical camera pose parameters concurrently.
It formulates the navigation problem as a non-linear least-squares minimization task (typically using Gauss-Newton or Levenberg-Marquardt algorithms), comparing observed pixel coordinates of visual landmarks across overlapping frames against their projected theoretical positions.
The algorithm systematically minimizes the global reprojection error, which is the geometric distance between detected image features and projected points on the image sensor plane.
This continuous optimization prevents incremental drift errors from compounding, produces a metric and consistent three-dimensional spatial map, and yields refined robot trajectory estimates essential for safe autonomous navigation.

E = i \sum j \sum ∥ x_{ij} - π (C_{i}, X_{j}) ∥^{2}

Mentor tip

Highlight the simultaneous optimization of both camera poses and 3D landmark coordinates to minimize reprojection error across multiple views.

(b)

4 marks

First Consideration: Operational autonomy versus human oversight in life-critical triage decisions. Delegating casualty rescue sequencing entirely to automated algorithms introduces severe moral risk if the software misinterprets victim viability; maintaining a human-in-the-loop ensures ethical accountability and professional clinical judgment during emergency triage.
Second Consideration: Survivor psychological trauma and physical safety during automated interaction. Trapped, injured victims in dark, confined rubble may experience intense terror when approached by a mechanical autonomous machine lacking human empathy; additionally, unintended mechanical malfunction during obstacle clearance could inflict accidental physical harm on fragile individuals.

Mentor tip

State two distinct ethical dimensions, clearly detailing the underlying ethical dilemma and its real-world human impact during disaster operations.

Question 3

6 marks

Algorithmic and environmental constraints on survivor pose estimation

Control systems

6 marks

Severe Environmental Visual Degradation: Collapsed industrial spaces are characterized by thick airborne dust, dense smoke, low ambient illumination, and dirt. These factors degrade optical sensor clarity, diminish image contrast, and introduce heavy noise that disrupts edge detection and convolutional feature extraction.
Physical Occlusion by Debris: Survivors in closed disaster spaces are frequently partially buried beneath drywall, fallen roof beams, or structural furniture. When critical anatomical joints (such as hips, knees, or shoulders) are concealed, keypoint prediction algorithms lack essential visual data, leading to severe predictive failures or omitted skeletal segments.
Atypical Non-Canonical Human Configurations: Pretrained deep learning pose models are predominantly trained on upright, walking, or standard seated individuals. Trapped casualties often adopt twisted, contorted, prone, or fetal postures, producing skeletal joint configurations that standard neural architectures fail to match or mistakenly reject as statistical outliers.
Computational Resource Limitations on Embedded Platforms: Three-dimensional multi-person pose estimation requires heavy neural network inference. Running these complex matrix computations in real time on mobile robotic hardware causes severe processor bottlenecks, generates extreme heat, and accelerates battery drainage, shortening operational deployment windows.
Dynamic Scale and Perspective Distortion: Structural debris forces robots to capture victims from oblique angles and highly variable camera distances. Without reliable stereoscopic depth or LiDAR confirmation, monocular image feeds suffer from depth ambiguity, making it difficult to distinguish a small child nearby from an adult further away.

Mentor tip

Structure your explanation across sensory/environmental conditions, occlusion and non-canonical posture challenges, and onboard embedded hardware constraints.

Question 4

12 marks

Strategic evaluation of visual SLAM and pose estimation rescue robotics in closed environments

Control systems

12 marks

Operational and Technical Benefits: Deploying autonomous rescue robots equipped with visual simultaneous localization and mapping (vSLAM) and skeletal pose estimation offers transformative operational value in dangerous closed spaces such as collapsed factories or structurally compromised commercial buildings. By penetrating unstable environments subject to imminent collapse, toxic fumes, or intense heat, robots eliminate direct human risk for first responders. Unlike human rescuers who tire quickly and face sensory overload, autonomous robots operate persistently, systematically mapping unknown architectural layouts where prior blueprints are unavailable or rendered obsolete by structural collapse. Furthermore, deep learning pose estimation moves beyond basic bounding-box object detection; by identifying precise joint coordinates, it provides first responders with critical anatomical insight regarding whether a survivor is trapped, immobilized, conscious, or compressed under load, thereby accelerating targeted extrication.
Technical Liabilities and Computational Costs: Despite compelling benefits, substantial technical and financial hurdles impede pure robotic deployment. In closed disaster zones, visual SLAM algorithms face severe reliability bottlenecks. Thick smoke, dust clouds, and total illumination loss neutralize optical cameras, triggering catastrophic tracking failure. When tracking fails, the robot must execute computationally expensive relocalization routines, which struggle in visually degraded or structurally altered spaces. Furthermore, full 3D bundle adjustment and real-time multi-person pose inference demand immense parallel processing power. Equipping mobile robots with high-performance embedded graphics hardware drastically inflates manufacturing and maintenance costs, while draining onboard battery reserves within short operational windows. If robots suffer tracking drift or software crashes inside rubble, operations stall, wasting irreplaceable survival time. Additionally, procurement, personnel simulation training, and specialized maintenance impose significant economic burdens on municipal emergency budgets.
Architectural Synthesis and Recommended Framework: To resolve the tension between capability and environmental limitations, emergency response agencies should not deploy visual-only robotic systems in isolation. Instead, robots must implement a robust multi-sensor fusion architecture, supplementing RGB cameras with long-wave infrared thermal sensors (which penetrate dense smoke to reveal human body heat), solid-state LiDAR (which measures structural geometry independent of ambient light), and tactical inertial measurement units to bridge brief optical dropouts. High-intensity optimization tasks should be offloaded via wireless mesh links to portable tactical edge computing nodes stationed safely outside the structure. Finally, operational protocols should adhere to a shared-autonomy model: robots handle dangerous spatial mapping, environmental hazard telemetry, and victim localization, while experienced human rescue commanders interpret high-level situational data and direct physical triage. This hybrid paradigm maximizes survivor discovery while safeguarding public resources and human life.

Mentor tip

Organize your response with balanced depth: articulate specific technical benefits (safety, mapping, skeletal tracking), analyze concrete failure modes (smoke, computational drain, tracking drift), and synthesize an authoritative recommendation featuring multi-sensor fusion and edge computing.

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 systems20 marks · 67%
Q1, Q3, Q4
Modelling and simulation6 marks · 20%
Q1, Q2
System fundamentals4 marks · 13%
Q2

Mentor insights

Traps to watch for

01
Conflating dead reckoning with absolute positioning · Q1
Dead reckoning relies strictly on internal sensors (encoders, IMUs) to sum relative displacements from a start point; it receives zero external reference signals, meaning its position uncertainty compounds continuously over time without external drift correction.
02
Oversimplifying bundle adjustment as simple moving averages · Q2
Bundle adjustment does not merely smooth trajectory lines; it solves a large non-linear least-squares optimization problem that simultaneously refines both 6-DOF camera poses and 3D landmark coordinates to minimize reprojection errors across multiple overlapping views.
03
Assuming standard computer vision datasets transfer to disaster scenes · Q3
Standard pose estimation datasets assume clean lighting, upright posture, and unhindered visibility. In disaster zones, casualties are frequently contorted, prone, or heavily occluded by debris, causing standard models to fail unless specially fine-tuned on non-canonical postures.
04
Relying solely on optical cameras in smoke-filled environments · Q4
Pure visual SLAM collapses when smoke, dust, or power loss eliminates optical contrast; high-scoring essays must emphasize sensor fusion with thermal imaging, LiDAR, and IMUs to maintain spatial tracking under severe visual impairment.

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 TZ1 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

Dead reckoning navigation and loop closure challenges in visual SLAM

Control systems
Modelling and simulation

(a)

2 marks

Dead reckoning is a kinematic navigation technique where a mobile robot continuously estimates its current coordinates and heading by integrating incremental motion measurements forward from an established initial reference point.
The processor samples internal sensors, such as rotary wheel encoders, accelerometers, and rate gyroscopes, converting wheel rotations and angular velocities over discrete time intervals into estimated linear displacement and orientation changes.

Δ s = r \cdot Δ θ, x_{t} = x_{t - 1} + Δ s cos (ϕ_{t})

Mentor tip

Explain both the principle (advancing from a known baseline) and the sensor mechanism (integrating wheel rotation or inertial measurements over time).

(b)

2 marks

Reliable loop closure is difficult because perceptual aliasing frequently occurs in disaster zones, where structurally uniform corridors, identical rubble piles, or repetitive brick patterns cause the vision system to match current features against the wrong historical keyframe.
Furthermore, severe lighting variations, smoke occlusions, and accumulated rotational drift warp visual feature descriptors, preventing the matching algorithm from recognizing an identical physical site when revisited from an alternate angle.

Mentor tip

Focus on perceptual aliasing (different locations looking identical) or feature degradation caused by changing illumination and drift.

Question 2

8 marks

Bundle adjustment optimization and ethical deployment of rescue robotics

Modelling and simulation
System fundamentals

(a)

4 marks

Bundle adjustment performs joint non-linear optimization across an entire visual SLAM graph by refining both estimated three-dimensional feature coordinates and historical camera pose parameters concurrently.
It formulates the navigation problem as a non-linear least-squares minimization task (typically using Gauss-Newton or Levenberg-Marquardt algorithms), comparing observed pixel coordinates of visual landmarks across overlapping frames against their projected theoretical positions.
The algorithm systematically minimizes the global reprojection error, which is the geometric distance between detected image features and projected points on the image sensor plane.
This continuous optimization prevents incremental drift errors from compounding, produces a metric and consistent three-dimensional spatial map, and yields refined robot trajectory estimates essential for safe autonomous navigation.

E = i \sum j \sum ∥ x_{ij} - π (C_{i}, X_{j}) ∥^{2}

Mentor tip

Highlight the simultaneous optimization of both camera poses and 3D landmark coordinates to minimize reprojection error across multiple views.

(b)

4 marks

First Consideration: Operational autonomy versus human oversight in life-critical triage decisions. Delegating casualty rescue sequencing entirely to automated algorithms introduces severe moral risk if the software misinterprets victim viability; maintaining a human-in-the-loop ensures ethical accountability and professional clinical judgment during emergency triage.
Second Consideration: Survivor psychological trauma and physical safety during automated interaction. Trapped, injured victims in dark, confined rubble may experience intense terror when approached by a mechanical autonomous machine lacking human empathy; additionally, unintended mechanical malfunction during obstacle clearance could inflict accidental physical harm on fragile individuals.

Mentor tip

State two distinct ethical dimensions, clearly detailing the underlying ethical dilemma and its real-world human impact during disaster operations.

Question 3

6 marks

Algorithmic and environmental constraints on survivor pose estimation

Control systems

6 marks

Severe Environmental Visual Degradation: Collapsed industrial spaces are characterized by thick airborne dust, dense smoke, low ambient illumination, and dirt. These factors degrade optical sensor clarity, diminish image contrast, and introduce heavy noise that disrupts edge detection and convolutional feature extraction.
Physical Occlusion by Debris: Survivors in closed disaster spaces are frequently partially buried beneath drywall, fallen roof beams, or structural furniture. When critical anatomical joints (such as hips, knees, or shoulders) are concealed, keypoint prediction algorithms lack essential visual data, leading to severe predictive failures or omitted skeletal segments.
Atypical Non-Canonical Human Configurations: Pretrained deep learning pose models are predominantly trained on upright, walking, or standard seated individuals. Trapped casualties often adopt twisted, contorted, prone, or fetal postures, producing skeletal joint configurations that standard neural architectures fail to match or mistakenly reject as statistical outliers.
Computational Resource Limitations on Embedded Platforms: Three-dimensional multi-person pose estimation requires heavy neural network inference. Running these complex matrix computations in real time on mobile robotic hardware causes severe processor bottlenecks, generates extreme heat, and accelerates battery drainage, shortening operational deployment windows.
Dynamic Scale and Perspective Distortion: Structural debris forces robots to capture victims from oblique angles and highly variable camera distances. Without reliable stereoscopic depth or LiDAR confirmation, monocular image feeds suffer from depth ambiguity, making it difficult to distinguish a small child nearby from an adult further away.

Mentor tip

Structure your explanation across sensory/environmental conditions, occlusion and non-canonical posture challenges, and onboard embedded hardware constraints.

Question 4

12 marks

Strategic evaluation of visual SLAM and pose estimation rescue robotics in closed environments

Control systems

12 marks

Operational and Technical Benefits: Deploying autonomous rescue robots equipped with visual simultaneous localization and mapping (vSLAM) and skeletal pose estimation offers transformative operational value in dangerous closed spaces such as collapsed factories or structurally compromised commercial buildings. By penetrating unstable environments subject to imminent collapse, toxic fumes, or intense heat, robots eliminate direct human risk for first responders. Unlike human rescuers who tire quickly and face sensory overload, autonomous robots operate persistently, systematically mapping unknown architectural layouts where prior blueprints are unavailable or rendered obsolete by structural collapse. Furthermore, deep learning pose estimation moves beyond basic bounding-box object detection; by identifying precise joint coordinates, it provides first responders with critical anatomical insight regarding whether a survivor is trapped, immobilized, conscious, or compressed under load, thereby accelerating targeted extrication.
Technical Liabilities and Computational Costs: Despite compelling benefits, substantial technical and financial hurdles impede pure robotic deployment. In closed disaster zones, visual SLAM algorithms face severe reliability bottlenecks. Thick smoke, dust clouds, and total illumination loss neutralize optical cameras, triggering catastrophic tracking failure. When tracking fails, the robot must execute computationally expensive relocalization routines, which struggle in visually degraded or structurally altered spaces. Furthermore, full 3D bundle adjustment and real-time multi-person pose inference demand immense parallel processing power. Equipping mobile robots with high-performance embedded graphics hardware drastically inflates manufacturing and maintenance costs, while draining onboard battery reserves within short operational windows. If robots suffer tracking drift or software crashes inside rubble, operations stall, wasting irreplaceable survival time. Additionally, procurement, personnel simulation training, and specialized maintenance impose significant economic burdens on municipal emergency budgets.
Architectural Synthesis and Recommended Framework: To resolve the tension between capability and environmental limitations, emergency response agencies should not deploy visual-only robotic systems in isolation. Instead, robots must implement a robust multi-sensor fusion architecture, supplementing RGB cameras with long-wave infrared thermal sensors (which penetrate dense smoke to reveal human body heat), solid-state LiDAR (which measures structural geometry independent of ambient light), and tactical inertial measurement units to bridge brief optical dropouts. High-intensity optimization tasks should be offloaded via wireless mesh links to portable tactical edge computing nodes stationed safely outside the structure. Finally, operational protocols should adhere to a shared-autonomy model: robots handle dangerous spatial mapping, environmental hazard telemetry, and victim localization, while experienced human rescue commanders interpret high-level situational data and direct physical triage. This hybrid paradigm maximizes survivor discovery while safeguarding public resources and human life.

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 systems20 marks · 67%
Q1, Q3, Q4
Modelling and simulation6 marks · 20%
Q1, Q2
System fundamentals4 marks · 13%
Q2

Mentor insights

Traps to watch for

01
Conflating dead reckoning with absolute positioning · Q1
Dead reckoning relies strictly on internal sensors (encoders, IMUs) to sum relative displacements from a start point; it receives zero external reference signals, meaning its position uncertainty compounds continuously over time without external drift correction.
02
Oversimplifying bundle adjustment as simple moving averages · Q2
Bundle adjustment does not merely smooth trajectory lines; it solves a large non-linear least-squares optimization problem that simultaneously refines both 6-DOF camera poses and 3D landmark coordinates to minimize reprojection errors across multiple overlapping views.
03
Assuming standard computer vision datasets transfer to disaster scenes · Q3
Standard pose estimation datasets assume clean lighting, upright posture, and unhindered visibility. In disaster zones, casualties are frequently contorted, prone, or heavily occluded by debris, causing standard models to fail unless specially fine-tuned on non-canonical postures.
04
Relying solely on optical cameras in smoke-filled environments · Q4
Pure visual SLAM collapses when smoke, dust, or power loss eliminates optical contrast; high-scoring essays must emphasize sensor fusion with thermal imaging, LiDAR, and IMUs to maintain spatial tracking under severe visual impairment.

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