Ergonomics is not “comfortable chairs”. It is the study of the fit between the human and the work system, measured in injuries, error rates, throughput and effort. HCI inherited its principles — feedback, error prevention, memory load, control — from this discipline.
01What ergonomics is — the definition to reproduce
Ergonomics (or human factors) is the scientific discipline concerned with the understanding of interactions among humans and other elements of a system, and the profession that applies theory, principles, data and methods to design in order to optimise human well-being and overall system performance.
From the Greek ergon (work) + nomos (laws). Two aims, both required, in this order: well-being and performance. That dual aim is the sentence most likely to appear as a two-mark definition.
What it optimises
- Health and safety — prevent injury and illness.
- Comfort — reduce strain, fatigue, discomfort.
- Performance — productivity, accuracy, fewer errors.
- Satisfaction — the system is acceptable to the people who use it.
Where it came from
- Military origins (WWII) — equipment designed for the average soldier failed for everyone else; cockpit layouts, gun sights, control placement.
- Space race — human capability in extreme environments forced explicit models of workload and attention.
- Industry — assembly lines, repetitive strain, shift work.
- Automotive and aviation — the operator is part of a safety-critical system.
- Computing — the operator became a person sitting at a screen for eight hours, and the keyboard became an injury vector.
Human-centred design is the contemporary label: design the work around the human, rather than requiring the human to adapt to the artefact. In HCI terms it is the ancestor of user-centred design, participatory design and usability engineering.
02Physical ergonomics
Concerns the human body: anatomical, anthropometric, physiological and biomechanical characteristics as they relate to physical activity.
| Concept | What it means | Interface example |
|---|---|---|
| Anthropometry | Measurement of the dimensions and capabilities of the human body | Keyboard size, key pitch, screen height, seat depth |
| Biomechanics | Forces on joints, tendons and muscles | Wrist extension while typing; force needed to click |
| Posture | The sustained configuration of the body | Neck flexion looking down at a laptop |
| Repetition | The same movement many times per minute | Mouse clicking, scrolling, typing |
| Force | Effort required to activate a control | Stiff keys, heavy trackball, touchscreen pressure |
| Mechanical stress / vibration | Local pressure or oscillation against tissue | Resting the wrist on a hard desk edge |
Musculoskeletal disorders (MSDs)
- Injuries to muscles, nerves, tendons, joints, cartilage and spinal discs, caused or aggravated by repeated physical stress.
- The upper limb and neck are the computer-era sites: tendinitis, tenosynovitis, epicondylitis (“tennis elbow”), ganglion cysts, and the famous carpal tunnel syndrome.
Awkward postures — the numbers
| Deviation | Threshold quoted in the lecture | Why it matters |
|---|---|---|
| Wrist flexion (bending the hand down/palm-ward) | ≥ 30° | Increases carpal-tunnel pressure |
| Wrist extension (bending the hand up) | ≥ 45° | Classic keyboard angle injury; raised keyboards make it worse |
| Ulnar deviation (hand bending toward the little finger) | ≥ 30° | Caused by a straight keyboard forcing the forearms together |
The design consequences: keep the wrist neutral; keep the elbows near 90°; support the forearms; put the keyboard at negative or flat incline; move the mouse with the whole arm rather than the wrist; take micro-breaks. Every one of these is a design constraint, not a user's discipline problem.
03Keyboards as an ergonomic case study
| Design | Idea | Trade-off |
|---|---|---|
| Straight keyboard | Cheap, familiar, one piece | Forces ulnar deviation and wrist extension |
| Fixed split | Two halves angled to match the forearms; typically 15–25° split angle, 8–66° incline | Keeps the wrist straighter; but a fixed angle fits only one body |
| Adjustable split | The user sets the angle and separation (“tenting”, “tilt”) | Fits more bodies; costs more, takes space, adds setup complexity |
| Sculptured / contoured | Key wells shaped to finger length; thumb clusters | Short travel and less force; steep learning curve, high price |
| Single-handed (e.g. Maltron) | Whole keyboard operable with one hand, for amputation or hemiplegia | Accessibility win; slower for two-handed users, needs training |
The research claim the lecture makes: split and alternative geometries reduce wrist extension and ulnar deviation, but the benefit depends on adjustability, training and the rest of the workstation. A “ergonomic” keyboard on a desk that is too high is still an injury.
04Cognitive ergonomics
Concerns mental processes: perception, memory, reasoning and motor response as they affect interaction among humans and other elements of a system. Its topics: mental workload, decision-making, skilled performance, human–computer interaction, reliability, stress and training.
When cognitive ergonomics is the whole problem
- Complex, safety-critical interfaces — nuclear control rooms, air traffic control, spacecraft, operating theatres. The operator must build an accurate mental model from an incomplete display, under time pressure, with no second chance.
- Automation surprises — the system acts, the human does not know it acted, and the human's model diverges from the machine's state. This is the Three Mile Island mechanism from chapter 1 in modern clothes.
MCAS automatically pushed the nose down when it detected a high angle of attack, using a single sensor. Pilots were not trained on it, were not told it existed in the manual they had, and the system could repeatedly override their inputs. Two crashes, 346 deaths, and a global grounding.
As cognitive ergonomics: the automation had authority without transparency, no shared mental model, no failure disclosure, and no way for the human to understand or override the machine's belief. That is a mental-workload and reliability failure, not a hardware failure — which is exactly why it belongs in this chapter.
The six HCI principles inherited from ergonomics
These are the design rules the lecture attributes to the ergonomic tradition; they double as a marking checklist for any critique question.
- Consistency — the same action has the same effect everywhere.
- Informative feedback — the system says what it did and what state it is in.
- Error prevention — constrain actions so slips are impossible, not merely detectable.
- Reduce short-term memory load — show, don't require recall.
- User control / internal locus of control — the user initiates and can undo; the system does not take over silently.
- Flexibility and efficiency of use — accommodate novice and expert, and the disabled user, without punishing either.
05System performance: three criteria
| Criterion | Question | Measure |
|---|---|---|
| Effectiveness | Did the user achieve the goal, completely and accurately? | Task success rate, error rate, completeness |
| Efficiency | How much resource did it take? | Time on task, keystrokes, clicks, mental effort (Fitts' IP is an efficiency measure) |
| Effort / satisfaction | How did it feel, and would they do it again? | Subjective rating, fatigue, preference, retention |
Ergonomics adds the human cost to all three: an efficient system that injures the operator has failed, and a comfortable system that cannot complete the task has also failed. The definition's two aims are not a trade-off to be balanced once — they are both constraints.
06Input devices — one advantage, one cost each
| Device | Strength | Cost / limitation |
|---|---|---|
| Keyboard | Fast, precise text entry; many shortcuts; accessible with training | Occupies desk; causes MSD if posture is bad; poor for pointing |
| Mouse | Fast 2D pointing; direct mapping of hand movement to cursor; Fitts-efficient | Needs a surface; wrist extension/ulnar deviation; not usable in the air |
| Trackball | No desk space needed; the device stays still (good in tight or moving environments) | Less familiar; thumb fatigue; coarser pointing |
| Touchpad | Integrated, no extra hardware; gestures | Small target area; accidental touches; slower for precise work |
| Joystick | Continuous control in 2–3 axes; good for vehicles and games | Imprecise for text; strain if held for long; mapping must be learned |
| Graphics tablet / stylus | Natural drawing; pressure and tilt; handwriting | Hand–eye separation unless the screen is the tablet; expensive; learning curve |
| 3D mouse / 6-DoF controller | Six degrees of freedom at once (3 translation + 3 rotation); ideal for CAD and 3D navigation | Only useful in 3D tasks; steep learning; expensive |
| Data glove | Captures finger configuration for VR and rehabilitation | Encumbering; calibration drift; hygiene |
| Touchscreen | Direct manipulation — the finger is the pointer; no intermediate device; flexible labels | No tactile feedback; occlusion by the hand; fat-finger errors (Fitts again); “gorilla arm” when vertical; smudges; accidental touches |
| Speech | Hands-free; fast for commands; natural for users with motor impairment | Noise and privacy; recognition errors that are hard to correct; accents; users must remember the vocabulary |
| Gesture / mid-air | No contact; works at a distance; expressive | No haptics; fatigue; ambiguous segmentation (“was that a click?”) |
Degrees of freedom
Six DoF = three translations (x, y, z) + three rotations (pitch, yaw, roll). A standard mouse gives two (x, y) plus buttons; a 3D mouse gives six simultaneously; a gamepad gives two analogue axes plus triggers. The number of DoF a device offers must match the number the task needs — extra DoF become modes and errors, too few DoF force sequences.
Selection heuristics for an exam scenario
- Text-heavy professional work → keyboard + mouse; ergonomic split if hours are long.
- Hostile or mobile environment (factory floor, ambulance, kiosk) → touchscreen with large targets, or sealed keyboard; avoid precision pointing.
- 3D modelling / surgery simulation → 6-DoF device or haptic stylus; keyboard shortcuts for modes.
- Motor impairment or one hand → single-handed keyboard, speech, eye/head tracking; combine modalities, never rely on one.
- Public kiosk → touch, large targets, visible timeout, no keyboard hygiene problem.
07Monitoring the user: activity, eyes, head, brain
Activity monitoring
- Control activity — intentional, task-directed input.
- Spurious activity — accidental input, noise, a brush against the device.
- No activity — the user is thinking, reading, or has left. Absence of input is data.
Eye tracking
| Measure | What it shows |
|---|---|
| Fixation | Where attention rests; comprehension happens here |
| Saccade | The jump between fixations; direction and amplitude of search |
| Scanpath | The ordered sequence of fixations — the search strategy, not just the stops |
| Heat map | Aggregate dwell over many users; where the interface actually draws the eye |
| Smooth pursuit | The eye follows a moving target; used for calibration and for gaze-controlled motion |
| Vergence | Both eyes converge/diverge for depth; a cue for 3D displays |
| Vestibulo-ocular reflex | Eyes counter-rotate when the head moves; why head-mounted displays need low latency |
Uses: usability testing (does the user see the button?), reading research, gaze-based input for users with motor impairment, attention-aware interfaces. Limits: the tracker knows where the eye is, not what the mind is doing; calibration drifts; glasses and mascara break it.
Head tracking and brain–computer interfaces
- Head tracker — orientation and position of the head; used for hands-free pointing, VR/AR, and accessibility. More robust than gaze for selection, less precise.
- BCI (EEG) — non-invasive electrodes read cortical potentials. Strengths: no movement needed; used for locked-in patients, neurorehabilitation, and control of prostheses. Limits: very low bandwidth, noisy signal, long setup, fatigue, and the ethical question of neuro-privacy — brain signals can reveal attention, affect or intent that the user did not choose to disclose.
08Affective computing
Rosalind Picard (MIT) founded the field: computing that relates to, arises from, or deliberately influences emotions. The interface should recognise affect and respond appropriately.
What can be measured
- Facial expression — video, facial action coding; the classic channel.
- Voice — prosody, pitch, energy, speech rate.
- Physiology — heart rate, skin conductance, respiration, pupil diameter.
- Behaviour — typing pressure, mouse speed, clicks, pauses, error bursts.
- Self-report — the user says how they feel.
Kuleshov intercut the same neutral face with a plate of soup, a coffin and a woman, and audiences reported hunger, grief and desire. The face did not change; the context did. So a smile read off a camera is not “happy” — it is a contraction of muscles whose meaning depends on the situation, the culture and the person. Affective computing's hard problem is exactly the pragmatics that chapter 4 said models lack.
Additional limits: self-report is unreliable (people misreport and under-report); physiological signals are confounded (arousal from exercise looks like arousal from anger); cultural display rules differ; and inference of emotion from a face raises privacy and consent issues, especially when it is used to judge a worker or a student.
09Concept check
Question 1According to the IEA definition, ergonomics optimises…
Question 2Carpal tunnel syndrome is caused by compression of…
Question 3Which posture threshold pair is quoted in the lecture for wrist deviations?
Question 4A fixed-split keyboard's split angle is typically…
Question 5The 737 MAX MCAS case is primarily an example of…
Question 6A user completes the booking task but takes 22 minutes and 140 clicks, and reports exhaustion. Which combination of performance criteria is described?
Question 7A device offers three translations and three rotations simultaneously. How many degrees of freedom is that?
Question 8Which is not an eye-tracking measure?
10Exam-style questions
A university opens a 40-seat computer lab used eight hours a day. Staff report wrist and neck pain, especially among students using the fixed-height benches with laptops. Identify four ergonomic risks, name the mechanism for each, and propose a fix that addresses the cause rather than the symptom.
Show model answer
| Risk | Mechanism | Fix |
|---|---|---|
| Wrist extension >45° on a raised laptop keyboard | Laptop on a desk tilts the hands up; carpal-tunnel pressure rises with extension | External keyboard at flat/negative incline, or a laptop stand plus separate keyboard, so the wrist stays neutral |
| Ulnar deviation >30° from a straight keyboard | Forearms forced together by a one-piece board | Split keyboard at 15–25°, adjustable if possible |
| Neck flexion from a low screen | The head drops to meet the display; sustained flexion loads the cervical spine | Raise the screen so the top line is at eye height, at ~50–70 cm |
| Static posture and no breaks | Same tissue loaded continuously for hours; circulation and recovery reduced | Height-adjustable desks, sit–stand alternation, scheduled micro-breaks, software reminders |
| Mouse reach and grip force | Shoulder abduction and wrist deviation; small targets increase force and time (Fitts) | Bring the mouse next to the keyboard, use a larger target size, enable keyboard navigation |
The cause-not-symptom point: a wrist rest supports the wrist while it is still extended; the fix is to change the geometry so the wrist does not need support. State this explicitly for the final marks.
Marking note. 2 marks per risk (mechanism + fix), plus 2 for the “fit the task to the human” framing.
An art museum wants an interactive wall for visitors: browsing a collection, zooming into paintings, and leaving a comment. Visitors stay for 2–3 minutes, may be wearing gloves, and the wall is 2.2 m tall.
(a) Recommend an input technology and justify it with two advantages and two limitations. (6)
(b) Explain how Fitts' law changes your design of the touch targets. (4)
Show model answer
(a) Large-format capacitive or infrared touch wall (a mid-air gesture layer is a possible addition for zoom). Advantages: direct manipulation — the visitor points at the painting itself, so no mapping must be learned and no device must be held; no movable hardware — nothing to steal, sanitise between users is easy, and it survives a public space; large targets suit a standing, moving audience; the labels can change per screen.
Limitations: no tactile feedback, so confirmation must be visual and generous; occlusion and “gorilla arm” — the hand hides the target and a 2.2 m wall forces reaching, which fatigues in under a minute; also glove compatibility (capacitive fails with gloves, infrared does not) and hygiene. A trackpad or physical controller at standing height avoids the reach problem but adds hardware and a mapping. A defensible answer names the glove issue and picks infrared for that reason.
(b) Fitts: T = a + b·log₂(2D/W). The wall is large, so D is large for a visitor standing at the centre; increasing W is the only lever. Make targets ≥ 2 cm at the viewing distance, place primary actions low and central (short D), avoid corners, keep the comment keyboard as a large-key on-screen layout, and accept that a 2.2 m wall has a “reach envelope” that no target size can fix — put the interactive band within 0.9–1.6 m. Adjacent targets need a gap, or the effective W drops and the error rate rises (fat-finger).
Marking note. (a) needs a technology and a consequence, not a list. (b) needs the formula and the two levers D and W, plus the reach-envelope insight for the top band.
A vendor claims its EEG headset “detects when a student is confused and adapts the lesson”. Evaluate the claim as a cognitive-ergonomics intervention: what would it need to measure, what would it get wrong, and what are the ethical conditions of use?
Show model answer
What it measures. EEG gives cortical potentials correlated with arousal, attention and some error-related signals (e.g. error-related negativity). “Confusion” would be inferred from a classifier over those patterns, ideally combined with behaviour (pause length, error rate, revisits) and self-report.
What it gets wrong. The signal is noisy and low-bandwidth; artefacts from blinking, jaw clenching and movement contaminate it. Arousal is not confusion — the same pattern can mean engagement, frustration or fatigue. The classifier is trained on a population, so it is least reliable for the atypical student it is meant to help. And Kuleshov applies: context determines meaning, and the headset has no context.
Ethical conditions. Informed consent (and assent, for minors); a clear purpose limitation (neuro-data must not become a grading signal); the right to opt out without penalty; data minimisation and local processing; transparency about what is inferred; and a human in the loop — the adaptation must be a suggestion to the teacher, not an automatic judgement of the student. Without these, the system converts a learning difficulty into a surveillance record.
Marking note. Three parts, roughly 3 + 3 + 2. “It reads thoughts” scores zero — say “classifies correlates of mental state”.
Fitts' law was established with a mouse on a flat screen. A rover operator on Earth must pick a rock with a robot arm on Mars, with a signal delay of 12 minutes. Does Fitts' law still apply? What does the delay do to the interaction design?
Show model answer
The law still applies in form. The task is still pointing: a distance to cover, a tolerance (the rock's size), a device, a human. T = a + b·log₂(2D/W) still describes the operator's movement, and ID still rises with distance and falls with target size. What changes is that a and b must be re-estimated for the hand controller, the camera view and the operator's training, and W must be the projected tolerance on the 2D camera image, not the rock's true size — a rock seen obliquely has a smaller effective width.
The delay changes the design, not the law. A 12-minute round trip means no closed-loop corrective movements: the operator cannot see the result and adjust, so the movement becomes one ballistic plan. Fitts' prediction still holds for the planned movement, but the error rate rises because visual feedback arrives after the fact. Design responses: slow the arm and use waypoints; move the camera to give a perpendicular view (maximise effective W); enlarge the tolerance (a “snap” or affordance zone); provide a predictive preview/simulation; and separate the planning from the execution so the operator confirms a plan rather than steering in real time.
Marking note. “Form remains, constants change” is the lecture's own phrasing for Fitts under altered conditions. The delay argument must mention the loss of the feedback loop, not just “it's slower”.