Training simulators
Simulations with a person in the loop: cockpits, operating theatres, control rooms and bridges.
These are simulations with a human operator in the loop. The point isn't to predict the world. It's practice: letting people rehearse rare, costly and dangerous situations before they meet them for real.
Full-flight simulators
A replica cockpit on a motion platform with the aircraft’s flight model, systems and failures, good enough that pilots qualify on new types in them.
Question it answers: Can the crew handle an engine failure on take-off?
Leaves out: Motion platforms approximate acceleration; sustained acceleration cannot be reproduced.
Runs on: about 1013 FLOPS rough a simulator's dedicated computer, real timeSince: 1929 Edwin Link's pneumatic pilot trainerIn the simulations tree: Link Trainer
Driving and rail simulators
A car or train cab with a projected world, used for research and teaching.
Question it answers: How do fatigue, distraction or a new display change how people drive?
Leaves out: Drivers know the environment is not real, so risk-taking does not fully match that on the road.
Runs on: about 1013 FLOPS rough a dedicated simulator computer, real timeSince: 1970 by 1970, early university driving simulators for road safety research
Ship bridge simulators
A full ship’s bridge with realistic handling, weather and harbour traffic.
Question it answers: Can the pilot bring this ship into a narrow harbour in a crosswind?
Leaves out: Rare mechanical failures and true sea conditions are hard to fully replicate.
Runs on: about 1013 FLOPS rough a dedicated simulator computer, real timeSince: 1972 by 1972, early full-mission ship bridge simulators for officer training
Spaceflight and mission control
Spacecraft and mission-control simulators running every phase of a flight, with failures injected on purpose.
Question it answers: Can the crew and the ground team handle what goes wrong?
Leaves out: Only failures the designers expected are injected. Real missions throw up ones they didn't.
Runs on: about 1013 FLOPS rough mission-control simulator computers, real timeSince: 1959 early NASA simulators trained the first astronauts and ground teams
Heavy equipment simulators
Cranes, excavators and mining trucks with realistic controls and loads, for training operators away from a real site.
Question it answers: Can the operator handle this load or this ground safely?
Leaves out: Real ground conditions and machine wear are harder to replicate than the model assumes.
Runs on: about 1012 FLOPS rough the simulator's own computer, real timeSince: 1990 by 1990, early crane and mining equipment simulators for operator training
Diving and submersible simulators
Underwater conditions, buoyancy and vehicle handling for divers and crewed submersible pilots.
Question it answers: Can the diver or pilot handle an emergency at depth?
Leaves out: Real water pressure and cold on the body are things a simulator on land cannot recreate.
Runs on: about 1012 FLOPS rough the simulator's own computer, real timeSince: 1978 by 1978, early submersible pilot training simulators
Surgical simulators
Virtual tissue and instruments with force feedback, for practising procedures.
Question it answers: Is this trainee ready to operate on a real patient?
Leaves out: Virtual tissue does not bleed, tear or smell as the real thing does.
Runs on: about 1012 FLOPS rough a desktop with haptic hardware, real timeSince: 1993 by 1993, early virtual-reality simulators for laparoscopic surgery
Patient simulators
Manikins that breathe, bleed, talk and respond to drugs, for emergency and team training.
Question it answers: Does the team recognise the problem and act in time?
Leaves out: Physiology is scripted to the scenario; unscripted complications do not emerge on their own.
Runs on: about 1010 FLOPS rough an embedded computer inside the manikin, real timeSince: 1969 the first anaesthesia training manikin, built in Los Angeles
Dental simulators
Virtual teeth and gums with force feedback, for practising drilling, fillings and extractions.
Question it answers: Is this trainee's hand steady and careful enough for a real patient's mouth?
Leaves out: Virtual tooth and gum tissue still feels more uniform than a real patient's mouth.
Runs on: about 1011 FLOPS rough a dental school workstation, real timeSince: 1997 by 1997, early haptic dental training simulators
Childbirth simulators
A birthing manikin with a model baby, for practising normal delivery and rare emergencies like a stuck shoulder.
Question it answers: Does the team recognise and manage a birth emergency quickly enough?
Leaves out: The manikin's fixed set of complications does not cover every real birth emergency.
Runs on: about 109 FLOPS rough an embedded controller inside the manikin, real timeSince: 1996 by 1996, early computerised birthing manikins for obstetric training
Anaesthesia simulators
A virtual patient's breathing, heart and drug response, controlled by a trainee giving anaesthetic drugs.
Question it answers: Does the trainee dose correctly and catch a complication before it becomes dangerous?
Leaves out: The model's drug responses are typical ones; real patients react more variably.
Runs on: about 1011 FLOPS rough a training centre desktop, real timeSince: 1967 Sim One, an early anaesthesia training manikin with simulated physiology
Paramedic and mass-casualty triage training
Simulated casualties with different injuries, used to practise sorting patients by urgency under time pressure.
Question it answers: Does the trainee sort patients correctly when there are more casualties than hands to help?
Leaves out: Actors and manikins cannot fully convey the disorder of a real mass-casualty scene.
Runs on: about 1010 FLOPS rough a training exercise's basic equipment, little computingSince: 1988 by 1988, early computer-assisted mass-casualty triage exercises
Virtual human patients
Animated characters respond to a trainee's questions with scripted or generated dialogue to practise clinical skills.
Question it answers: How well can a trainee interview a patient who is not real?
Leaves out: Dialogue range is limited, and nonverbal cues are shallow.
Runs on: about 109 FLOPS rough Real-time rendering and dialogueSince: 2007 Virtual patients for clinical therapist skills training (2007) source
Agent hospitals
A simulated hospital in which language-model doctors, nurses and patients carry out the care process and improve with experience.
Question it answers: Can agents practise treatment in a virtual hospital and improve?
Leaves out: Patients and diseases are generated, so realism of outcomes is unproven.
Runs on: about 1014 FLOPS rough Many model calls per simulated caseSince: 2024 Agent Hospital: a simulacrum of hospital with evolvable medical agents (2024) source
Catheter and endovascular simulators
A haptic device mimics a catheter and a virtual X-ray image of the vessels so trainees can practise cardiac or vascular procedures.
Question it answers: Can a trainee thread the catheter without damaging the vessel?
Leaves out: The feel of tissue is approximate and unusual anatomy is limited.
Runs on: about 1010 FLOPS rough real-time physics and haptics on a workstation.Since: 2000 Dawson and colleagues, designing a computer-based simulator for interventional cardiology training. source
Structured clinical examinations
Trainees move through a series of stations with trained actors playing patients and are scored against a checklist.
Question it answers: Can this trainee take a history and examine a patient to the standard?
Leaves out: An actor cannot produce real physical signs such as a heart murmur or an enlarged liver.
Runs on: no computer. no computation, a human role-play exercise.Since: 1975 Harden and colleagues, assessment of clinical competence using an objective structured examination. source
Ultrasound training simulators
A dummy body and a probe, with a computer showing the ultrasound image for that position, drawn from stored scans of real patients.
Question it answers: Can a trainee learn to find and read ultrasound views without a patient?
Leaves out: A limited library of stored cases cannot show the variety of real bodies or the feel of pressing the probe.
Runs on: about 109 FLOPS rough real time on a workstation.Since: 2004 Maul and colleagues' experience with SonoTrainer and comparative review of other ultrasound simulators. source
Robotic surgery simulators
A virtual copy of a surgical robot's console in which trainees practise moving instruments, cutting and suturing in simulated tissue, with scores for skill.
Question it answers: Is a surgeon safe to operate with the robot, as shown by practice in virtual tissue?
Leaves out: Virtual tissue is simpler than the real thing, so bleeding and complications are not well trained.
Runs on: about 109 FLOPS rough real time on a workstation.Since: 2009 Kenney and colleagues' face, content and construct validity study of the dV-Trainer virtual reality robotic simulator. source
Eye surgery simulators
A model eye viewed through a real microscope with virtual instruments inside it, to practise steps such as removing a cataract, with feedback on errors.
Question it answers: Can a trainee learn delicate eye surgery before touching a patient?
Leaves out: The eye's tissue is modelled coarsely, so rare complications are not rehearsed.
Runs on: about 109 FLOPS rough real time on a workstation.Since: 2009 Feudner and colleagues' randomised trial showing virtual reality training improves wet-lab performance of capsulorhexis. source
Endoscopy simulators
A real scope is passed into a model of the gut that displays a computer-drawn view of the lining and responds to turning and pushing.
Question it answers: Can a trainee learn to steer a scope and spot disease without a patient?
Leaves out: The gut's stretch and the patient's discomfort are poorly reproduced, so the feel of real procedures is missing.
Runs on: about 109 FLOPS rough real time on a workstation.Since: 1992 Beer-Gabel and colleagues' computer-assisted training in endoscopy, from a simulator to a learning station. source
Language-model simulated students
Language-model characters play confused or struggling learners so that trainee teachers can practise explaining and giving feedback.
Question it answers: Can new teachers practise responding to learners before meeting real ones?
Leaves out: The model's students are more polite and consistent than real ones and make mistakes real ones would not.
Runs on: about 1014 FLOPS rough a few model calls per turn of dialogue.Since: 2023 Markel and colleagues' GPTeach, interactive teaching-assistant training with GPT-based students. source
Control-room simulators
Exact copies of the control rooms of power stations, chemical plants and grids, driven by a model of the process.
Question it answers: Can the operators bring the plant back safely after an upset?
Leaves out: The model cannot include every valve and failure the real thing has.
Runs on: about 1013 FLOPS rough the simulator's own dedicated computer, real timeSince: 1958 by 1958, early full-scope simulators for nuclear power plant operators
Emergency exercises
A scripted disaster worked through by the people who would respond to it, around a table or across a real city.
Question it answers: Where does the plan break down?
Leaves out: Participants know it is an exercise, so real panic and confusion are underplayed.
Runs on: no computer. paper and people; sometimes a simple computer to track eventsSince: 1954 by 1954, civil defence tabletop exercises rehearse disaster response
Business simulations
Teams running a simulated company or market over several rounds, taught in business schools.
Question it answers: Can the team read the market and manage the company through it?
Leaves out: Simplified market rules can reward exploitation of the model, not necessarily sound management.
Runs on: about 1010 FLOPS rough a modest computer; most of the exercise is discussionSince: 1956 the American Management Association ran the Top Management Decision Game, an early computer-based business game
Air traffic control simulators
A simulated sky full of aircraft, radar screens and radio calls, for training and testing controllers.
Question it answers: Can the controller keep enough aircraft safely apart as traffic gets busier?
Leaves out: Simulated pilots follow instructions more predictably than real ones under pressure.
Runs on: about 1012 FLOPS rough an air navigation service provider's server, real timeSince: 1966 by 1966, early computer-driven air traffic control training simulators
Negotiation and diplomacy training simulations
Trainee diplomats, mediators or executives role-playing a dispute against instructors or each other, practising how to read the other side and reach a deal.
Question it answers: Did the trainee find a workable deal, and where did their approach go wrong?
Leaves out: Role-players usually compromise more readily than real counterparts under real domestic pressure.
Runs on: about 1010 FLOPS rough little computing; mostly people, rules and a scorer's desktopSince: 1963 by 1963, early simulated negotiation exercises for diplomatic training
Emergency incident command simulators
A live, timed exercise where commanders and staff receive escalating, unscripted updates about an unfolding emergency over real radios and screens.
Question it answers: Can the command team make good decisions in real time, not just with hindsight?
Leaves out: However realistic, everyone involved still knows no one will really get hurt.
Runs on: about 1011 FLOPS rough a training centre's server, real timeSince: 1974 by 1974, early functional emergency-management exercises with live radio traffic
Language-model conflict rehearsal
A language-model counterpart plays a difficult person so a trainee can rehearse a conflict and get feedback.
Question it answers: How can a hard conversation be practised safely?
Leaves out: The partner's responses are not those of a real person.
Runs on: about 1013 FLOPS rough Model calls per sessionSince: 2023 Rehearsal: simulating conflict to teach conflict resolution (2023) source
Mass-casualty exercises
Actors or manikins stand for many victims, and hospitals, ambulances and teams go through a staged disaster to test triage and capacity.
Question it answers: Would a hospital cope if dozens of seriously injured people arrived at once?
Leaves out: Staff know it is an exercise, so fear, fatigue and chaos of a real event are absent.
Runs on: no computer. no computation; a day of staff, actors and observers.Since: 1989 Simon and colleagues' simulated mass casualty evaluation at Beilinson Hospital. source