See the interfaces among data, control, hardware and people.
Robotics technology stack · Layer 05
Simulation and synthetic data
A disciplined way to use physics engines, digital twins, domain randomisation and software-in-the-loop tests without confusing simulation with field evidence.
Evidence statusEstablished engineering tool; transfer quality remains task-dependentLast reviewed: 11 August 2026
05
FUURAA thesis
Simulation is most valuable as a controlled experiment and coverage multiplier; every consequential claim still needs a calibrated bridge to physical tests.
Reading method
Turn a technology label into an engineering chain, testable claims and explicit boundaries.
Translate capability into task, latency, failure and recovery.
Separate standards, independent measurement, research and first-party claims.
State what cannot be inferred from a demo, benchmark or interface.
System breakdown
Four interdependent layers determine whether the technology can enter real work.
Each layer shows its role and the failure signal most worth watching.
Assets and geometry
Robot, tool, object and site models define the world the simulator can represent.
Physics and sensors
Contact, friction, compliance, actuators and sensor noise turn geometry into behavior.
Scenario generation
Randomization and adversarial cases expand coverage across objects, layouts, lighting and faults.
Reality calibration
Paired real measurements estimate which simulated metrics transfer and where correction is needed.
Engineering evaluation
Five checks turn abstract capability into reviewable system evidence.
Record normal performance, failure, recovery and human cost—not only the best-looking result.
- 01
State the simulation purpose
Separate training, design exploration, regression testing, safety analysis and throughput estimation.
- 02
Validate component by component
Compare sensor outputs, contact events, motion, energy and timing against physical references.
- 03
Track parameter provenance
Record whether values are measured, manufacturer-specified, estimated or tuned to fit an outcome.
- 04
Reserve physical holdouts
Do not use every real test to tune the simulator; preserve independent cases for transfer evaluation.
- 05
Report the reality gap
Publish where ranking, failure frequency or dynamics diverge instead of only showing matched examples.
Scope boundaries
State what the evidence supports—and what it does not.
- 01
Photorealism does not establish accurate contact, sensor timing or actuator dynamics.
- 02
Synthetic data can amplify assumptions and omit hazards absent from the scene generator.
- 03
A digital twin is only as current as its assets, calibration and configuration-control process.
Sources and evidence status
Keep the source, date, evidence identity and reading boundary visible.
This page prioritises standards bodies, public measurement programmes, official project documentation and original research disclosures.
01
NVIDIA · reviewed 2026-08-11
↗02Isaac Sim — Robotics Simulation and Synthetic Data Generation
Documents a workflow spanning asset import, physics and sensor configuration, synthetic data and software-in-the-loop tests.
Google DeepMind · 2021-10-18
↗Opening up the MuJoCo physics simulator
Explains MuJoCo's focus on contact-rich physics and its release for broader research use.