Building a simulator involves fundamental tradeoffs: a system that models physics perfectly may not run fast enough for real-time training; a simplified physics model may not prepare operators for real conditions. Understanding these tradeoffs helps designers make informed choices about where fidelity matters and where simplification is acceptable.
Why This Exists
Every simulation involves simplifications. Some simplifications matter enormously for training effectiveness; others are irrelevant. Knowing which is which allows resources to be directed toward the fidelity that counts, and prevents false confidence that “more realistic” always means “better training.”
Who This Is For
- Simulation designers building training systems
- Training managers evaluating simulation platforms
- Engineers integrating real hardware with simulation environments
- Researchers studying simulation effectiveness
The Fidelity Spectrum
Physics Fidelity
Physics fidelity refers to how accurately the simulation models the behaviour of physical systems:
- High physics fidelity — Accurately models fluid dynamics, structural mechanics, acoustic propagation
- Low physics fidelity — Uses simplified models (linear dynamics, lookup tables) that approximate behaviour
Computational cost: High-fidelity physics (e.g., computational fluid dynamics) is computationally expensive and may not run at real-time rates.
Control System Fidelity
Control system fidelity refers to how accurately the simulation models the control software and hardware:
- High control fidelity — Runs actual vehicle control software against simulated sensor inputs
- Low control fidelity — Uses simplified models of vehicle response that may not match the real control system
Testing value: High control fidelity is essential for testing control system software before deployment. Low control fidelity is acceptable for training operator skills that do not depend on control system specifics.
Where Fidelity Matters for Training
Task-Dependent Fidelity Requirements
The required fidelity depends on what is being trained:
| Training Objective | Required Fidelity |
|---|---|
| Procedure knowledge | Low — simulator only needs to present the right situations |
| Emergency decision-making | Medium — situation development must be plausible |
| Manual vehicle control skills | High — vehicle response must match reality |
| Fault diagnosis | High — failure modes must manifest as they do in reality |
| Environmental hazard recognition | High — environmental cues must be representative |
The Negative Transfer Problem
Negative transfer occurs when simulation training teaches the wrong habits — habits that degrade performance in the real environment. Negative transfer is caused by:
- Simulator controls that differ from real controls — Operators learn control habits on the simulator that are wrong for the real system
- Physics that differs significantly from reality — Operators learn to anticipate vehicle response that does not match real behaviour
- Consequence-free failure — Operators learn that they can recover from errors that are unrecoverable in reality
Principle: It is better to train with a lower-fidelity simulation that does not cause negative transfer than a high-fidelity simulation that teaches wrong habits for the wrong reasons.
Hardware-in-the-Loop (HIL)
HIL simulation runs actual vehicle hardware (control electronics, actuator drivers) against a simulated environment. This provides:
- Real control system behaviour — Exactly the same response characteristics as the real vehicle
- Real failure modes — Hardware failures manifest as they do in deployment
- Software validation — Control software can be tested without deploying the vehicle
Cost: HIL requires procuring additional hardware and integrating it with the simulation environment. For high-value vehicles, this investment is typically justified.
Software-in-the-Loop (SIL)
SIL runs the vehicle control software in simulation without physical hardware. This provides:
- Control software testing — Software bugs can be found without hardware
- Faster iteration — Easier to reset and re-run than HIL
- Limited hardware validation — Does not validate hardware implementation
Physics Simplifications and Their Implications
Simplified Hydrodynamics
Real vehicle hydrodynamics are nonlinear and coupled. Simplified models (linear drag, decoupled axes) are adequate for:
- High-level mission planning
- Operator training in stable conditions
Simplified models fail for:
- Training in strong currents or near seabed
- Control system design and tuning
- Fault behaviour in unusual attitudes
Simplified Acoustic Models
Acoustic propagation in the ocean depends on the sound velocity profile, bathymetry, and sea state. Simplified models may not accurately represent:
- Multipath effects that cause ranging errors
- Communication blackout zones
- Sonar image artefacts
Training implication: Operators trained only on simplified acoustic models may not recognise or respond correctly to multipath-induced errors in real operations.
Choosing the Right Level of Fidelity
A practical framework:
- Define training objectives — What skills and knowledge must the training develop?
- Identify critical fidelity requirements — Which aspects of the simulation must be realistic to achieve those objectives?
- Identify acceptable simplifications — Which aspects can be simplified without affecting training effectiveness?
- Validate the training — Measure training transfer to confirm the chosen fidelity is adequate